Exploring the Link between Structural Adjustment Programs, - - PDF document

Exploring the Link between Structural Adjustment Programs, Educational Discontinuities and Stalled Fertility in Sub-Saharan Africa

Endale Kebedea, Anne Goujona, Wolfgang Lutza Affiliation:

aWittgenstein Centre for Demography and Global Human Capital (IIASA, VID/OeAW, WU)

This paper is under review. Please don’t cite it.

Abstract Recently, most population projections for Sub-Saharan Africa have been corrected upwards because in a number of countries the earlier declining trends in fertility rates had stalled in the years around 2000. While most studies so far have focused on economic, political or other factors prevalent around 2000, here we show that the phenomenon is likely caused by the cohorts of girls who should have entered school in the 1980s but could not do so due to cuts in social expenditures following the Structural Adjustment Programs (SAP) imposed on several African countries. These programs affected expenditures on education by governments and households and had an impact on school intake, particularly for girls. Because of the well- established strong link between female education and fertility in Africa, the fertility decline stalled when these relatively less educated young cohorts entered their prime childbearing ages around 2000. To study this hypothesis in more detail we combine individual level data from series of DHS (Demographic and Health Surveys) for 18 African countries – with and without fertility stalls – thus creating a pooled dataset

• f 1,831,022 births to 573,091 women for cohorts born from 1950 to 1990 by level of education. Visual

and multivariate statistical analyses clearly confirm the hypothesis that the fertility stalls in the affected countries were largely caused by the cohorts of women that experienced education stalls in the context of

• SAP. Conversely, most countries that did not undergo such education stalls in the 1980s also did not face

fertility stalls around 2000. Significance Statement Whether Sub-Saharan Africa’s population will “only” double from 1 billion today or increase to 3 or 4 billion by 2100 will crucially depend on the future course of fertility. This will have massive implications for Africa and for the rest of the world, not least through international migration pressure and difficulties in meeting the sustainable development goals. Hence it is of highest priority to better understand the drivers

• f fertility decline and the reasons for its recent stall in some countries. Our finding that stalls are the late

consequences of cuts in the education system in the 1980s helps to resolve the puzzle. It clearly points at the necessity to prioritize universal female education on the African continent. Keywords Fertility, Sub-Saharan Africa, education, structural adjustment programs, population projections

Introduction All human populations have entered the process of demographic transition in which first death rates start to fall due to socio-economic development and improved public health and after a lag of several decades, birth rates start to decline. During the period when death rates are already low and birth rates are still high, the populations grow rapidly. This was the case in Europe and North America around 1900 and the process subsequently spread to Latin America, East Asia and then Southern and Western Asia. In most of these regions fertility rates have already fallen to low levels, even when the population still continues to grow due to the age-structural momentum in which larger cohorts of young women still enter the reproductive ages and death rates continue to fall. Sub-Saharan Africa has been the last region to enter this demographic transition. It was only in the 1980s that birth rates started to fall in most countries, but these declines have been uneven and stalling at times. Particularly in the late 1990s and early 2000s, some sub-Saharan African countries have experienced a levelling off of their fertility decline and in some cases even saw a reversal leading to an increase( as shown in figure 1). Much has been written and speculated about this so-called stalled African fertility transition (1). The reasons for this interruption of the fertility decline in many Sub-Saharan African countries have remained a demographic mystery as little consensus exists on the causes of the stalls (2). Most of the existing studies try to link the fertility stalls to some specific period factors such as the slower trends in socio-economic development prevalent in the stalling countries (3), the low priority assigned to family planning programs in the beginning of the 21st Century (4, 5), the impact of HIV/AIDS mainly through its effect on child mortality (2, 6) and other factors related to public and reproductive health. In contrast to Figure 1: Period birth rates (births per 1000 of population) for selected African countries with stalled and non-stalled period TFR declines. these explanations , recently, Goujon et al. (7) proposed another plausible explanation focusing on cohort rather than period effects. They linked the fertility stalls around 2000 to the fact that some cohorts of women were subject to an education stall, due most likely to the adverse effects on education of the structural adjustment programs launched by the Bretton wood’s institutions in the 1980s. While Goujon et al (7) gave

• nly a tentative descriptive analysis based on cross-sectional period data, here we test the hypothesis

systematically on cohort data constructed from a pooled individual level data of 1.831,022 births to 573,091 women based on series of Demographic and Health Surveys (DHS) in 18 countries. IMF policies, education and fertility The International Monetary Fund (IMF) and the World Bank provide low-cost financing to governments in economic turmoil, notably via lending programs, usually attached with conditionalities with which recipient countries have to comply. In the 1980s, these institutions launched the Structural Adjustment Programs (SAP), which made it possible for governments of many African countries with fiscal problems to get loans. It was a response to the post-independence African economic crises in many countries for which the World Bank mostly blamed excessive state interventions in the economy (8). Consequently, most of the conditionalities attached to the structural adjustment loans and facilities aimed at achieving balance of payments and usually included austerity measures through cutting government expenditures predominantly in the social sector or introducing user fees for those services. These budget cuts affected in particular the education and health systems, which had never been very strong in these poverty stricken countries (9, 10). The social impacts of these SAPs recommended and often imposed by IMF and World Bank have been highly controversial. While few studies have shown that adjustment lending had a positive effect on economic growth (11), there is a larger body of literature discussing the adverse consequence of SAPs on social and economic developments of the recipient countries starting soon after their implementation (12– 15). It has also been demonstrated that the progress of improving the quantity and quality of education that had started immediately after the independence of many countries has been seriously hampered through these budget cuts (12, 16, 17). At household level it affected children’s education negatively and girls suffered more as a result of discrimination in with-in household resource allocation in time of economic stress (13). These effects are also clearly visible in the cohort data presented in this study. Recently, the possible negative impact of IMF sponsored SAPs on child mortality has given rise to a controversial discussion in the pages of PNAS. Daoud et al (18) argue on the basis of a large pooled dataset from DHSs that SAPs reduce the protective effect of parental education on child health, especially in rural

• areas. Their models show that without SAPs living in a household with educated parents reduces the odds
• f child malnourishment by 38 percent, whereas in the presence of these programs the protective effect is

reduced to 21 percent. Similar adverse effects are found in sanitation, shelter and health care access. In a reaction to this paper Subramanian and De Neve (19) point at the shortcomings of a purely cross-sectional approach and defend IMF policies. In particular, they point at the fact that after 2000 IMF policies and lending mechanisms have considerably changed and that under the Millennium Development Goals child health has become a priority. Interestingly, these two papers only address the changes of the protective effects of educated parents and do not discuss the first order effect that SAPs had on hampering education in the 1980s and thus reducing the proportion of educated parents, which by itself would have a sizable negative effect on child health. In addition, the papers do not address the lasting cohort effects of the SAPs

• n education structure nor do they mention their effects on fertility. They cover issues of the 1980s and

1990s and discuss how blame should be assigned regarding a phenomenon that is supposedly over and history.

Our cohort approach shows that the issue is well alive today and will remain so as long as the unfortunate cohorts that in their childhood in the 1980s were deprived of their education opportunities will bear the consequences of this over their entire life cycle. These life-long consequences do not only concern health, they are also relevant for fertility, which in course has also externalities to population growth and society at large. In the literature, so far, fertility trends in Sub-Saharan Africa have hardly been linked to SAPs. The two phenomena are, indeed, almost two decades apart, which may have seemed too long for any direct causal effect from, for example, reduced reproductive health spending on fertility. This twenty-year lag is, however, precisely the timing that would be expected for an effect on fertility operating through female education: declining primary school enrolment rates for girls during the 1980s would result in lower education, and hence higher fertility, for women in their prime childbearing ages after 2000. Given the strong differentials in fertility by level of female education in all African countries and the extensive body

• f literature that explains the causal mechanisms behind the pervasive negative association between the two

(20–24), it seems to be a plausible hypothesis to assume a direct causal relationship between the stalled trend in female education and the subsequent stall in fertility decline in the countries affected by the former. So far, there has been no satisfactory explanation for this apparent discontinuity in around a dozen (depending on definition) African countries despite of a comprehensive effort by the NAS Committee on Population to assess the drivers and consequences of the fertility transition in Africa (25). The volume also highlights the fact that at any given level of socio-economic development fertility in Africa is higher than in other world region at comparable stages of development, something that Bongaarts calls the “Africa effect” (26). He also shows that this Africa effect of on average about one child more is entirely explained by higher desired family size rather than higher proportions of unwanted births. And he also shows that desired family size varies greatly with level of mothers’ education and in a regression model including also GDP per capita, life expectancy, percent urban and an Africa effect, education turns out to be clearly the most important determinant of all three dependent variables used: fertility, desired family size and contraceptive prevalence (26). Contraceptive prevalence is certainly a decisive proximate determinant of fertility. It is directly associated with lower desired family size and lower actual fertility. Hence, attempts to understand the reasons for the stalled fertility declines in the affected countries also gave particular attention to this factor. Askew et al (27) highlight the fact that in Kenya there was a complete stall in the contraceptive prevalence rate from 1998 to 2003 (at a level of 39 percent) which corresponds exactly to the fertility stall. The paper speculates that this stall may be due in part to a shift in in government and donor priorities towards the HIV/AIDS

• pandemic. However, why would such an effect be limited to only five years after which the previous trend

resumed? An alternative explanation can be given by the pattern of education improvements by successive birth cohorts of women, as shown in Figure 2 for Kenya. The purpose of our study is to systematically test this hypothesis by constructing real cohort data from the broadest available datasets. We investigate the link between longitudinal cohort education trends and longitudinal cohort specific fertility trends from consecutive DHSs conducted in the region. We also apply multivariate statistical analyses adding national level indicators to micro-level data in order to explore alternative explanations and quantify the relative impact of the different potential determinants of fertility.

Figure 2: Proportions of women with at least some formal education for 5-year birth cohorts of women in Kenya

Data used and cohorts reconstructed This study combines micro- and macro-level data to explore the link between the educational discontinuities and the period of fertility stalls in sub-Saharan Africa1. The micro data were pooled from DHSs collected

• ver the year 1990-2015. Within each country, the surveys made use of a two-stage cluster sampling

technique to collect comparable, reliable and nationally representative data on living conditions and demographic characteristics of households. Our dataset includes 18 sub-Saharan African countries representing about 66 percent of the population of the region in 2015 (28). The selected countries are Benin, Cote d’Ivoire, Cameroon, Congo, Ethiopia, Gabon, Ghana, Guinea, Gambia, Kenya, Lesotho, Niger, Nigeria, Senegal, Tanzania, Uganda, Zambia and Zimbabwe. The DHS surveys retrospectively collect and report full birth histories of women aged 15–49. The pooled dataset for the 18 countries under study includes 1,831,022 births to 573,091 women. DHSs also provide the micro-level data on mothers’ education as well as on the education of partners, and

• ther household members. DHS household files were used to construct an indicator for cohort educational

attainment, namely the proportion of women with at least some formal education by single-year birth cohorts. National macro-level data were also collected to control the possible impacts of period-related factors (such as economic slow-downs, child mortality and urbanization) in explaining the observed period fertility stalls. Data on real GDP/capita are obtained from the Penn World Tables, for the period 1960-2014. The impact

• f the economic situation in previous years is also controlled for by taking the lagged values of the log of

real GDP/capita. To address the possible impact of health crises (HIV/AIDS, malaria, malnutrition, etc.) we took information on country level child mortality rates for the period 1960-2016 from the World

1 Appendix Table S1 reports variable definitions and data sources

Development Indicators (29). Furthermore, we control for trends in urbanization by controlling for annual urban growth rates with data collected from the World Development Indicators (29). At the individual level, the models also include the possible influence of other background characteristics

• f women such as place of residence and religion. In addition, the individual membership of women in

specific birth cohorts is included in the multivariate models. We have reconstructed two series of cohort trends: (1) trends in age-specific fertility rates, and (2) trends in the proportion of women with at least some education, both by 5-year birth cohorts of women born over the 1950-1990 period. This helps in examining whether the patterns of fertility transitions and stalls (if any) are cohort- or period-related, and if they match with inter-cohort educational progress/discontinuities. These trends are reconstructed and compared for two different groups of countries: countries, which experience a levelling-off or reversal in the decline of period total fertility rate (TFR) around the year 2000 (called ‘fertility stalled’ countries) and countries which did not experience the stall (‘fertility non-stalled’ countries). Following the criteria defined by Goujon et al. (7) a country is labelled as ‘fertility stalled’ if the ratio of TFR between two consecutive periods (2000-2005/1995-2000 or 2005-2010/2000-2005) is 0.98 or

• above. Eight of the 18 sample countries (i.e. Congo, Kenya, Gambia, Niger, Nigeria, Tanzania, Zambia and

Zimbabwe) are labelled as ‘fertility stalled’ countries. Progress in cohort educational attainment is measured by the percentage growth in the share of women in each cohort with at least some education between two consecutive 5-year birth cohorts.

Inter-cohort changes in education and fertility

Figure 3 visualizes the pattern of inter-cohort changes in educational attainment for two groups of countries, based on their fertility stalling status around the year 2000 as defined above. It shows that for all countries the percentage increase in the proportions of women with at least some formal education had been higher in the 1960s than over the 1970s, although one has to keep in mind that levels of education were very low in the 1950s and therefore any improvement resulted in large increases. For the ‘fertility stalled’ countries this diminishing rate of improvement was more pronounced and the rate was already very low for the cohort born in 1975-80. This could either be a consequence of the developing macro-economic turmoil that lead to the adoption of SAPs or the first consequences of SAP implemented in the early 1980s. For the female cohort born during the period 1980-85 the figure shows negative growth in educational attainment for the ‘fertility stalled’ countries. This is a remarkable discontinuity and the only episode of its kind since the 1950s. Women born in 1980-85 are precisely the ones who were of primary school age at the end of the 1980s and early 1990s when the education systems of those countries were affected by SAPs introduced by the Bretton Woods institutions in the region. After that, the more recent cohort born in 1985- 90 again show an increasing growth rates for intake in education both in the ‘fertility stalled’ and ‘fertility non-stalled’ countries.

Figure 3: Inter-cohort growth rates of share women with some education by fertility stalling status of countries for 5-year birth cohorts of women. [It is measured as the share of women with some education in cohort (t) relative to share of women with some education in the previous cohort (t-1).]

Figure 4 shows that these patterns of inter-cohort progress in education are fully consistent with the inter- cohort changes in age-specific fertility rates. In all African countries considered, fertility rates have been declining gradually in each successive birth cohort up to the above discussed poorly educated cohort born in 1980-85 in the ‘fertility stalled’ countries. Indeed, for this group of countries, the previously rather smooth decline in the pattern of age-specific fertility already slowed down for the cohort born in 1975-1980 and even reversed for women born in 1980-85 who had actually higher fertility rates at ages 15-22 than the previous birth cohort. In contrast, ‘fertility non-stalled’ countries show no discontinuity in the declining pattern of age-specific fertility rates.

Figure 4: Age specific fertility rates per 1000 women for 5-year birth cohorts of women and fertility stalling status

• f countries (‘fertility non-stalled’ and ‘fertility stalled’ countries)

The above discussed results clearly show that the stall in the decline of age-specific fertility rates originates from the same cohort of women that earlier had also been affected by the discontinuity in the education

• expansion. This is the birth cohort of women who entered into reproductive age at the beginning of the 21st

century and who is associated with the stall in the decline of the overall period TFR. However, it also indicates that the effect is most visible for younger women aged 15 to 22 than for older women. This is likely due to the fact that women of the 1980-85 birth cohort who did not get a chance to go to school entered marriage and childbearing at an earlier age. To study this issue in more detail Figure 5 plots the proportions of childless women by age and 5-year birth cohort, separately for both ‘fertility stalled’ and ‘fertility non-stalled’ countries. Again, the inter-cohort pattern of transitions to first birth confirms our

• hypothesis. At each age, in the ‘fertility non-stalled’ countries the proportion of childless women is

increasing from one birth cohort to the next with the increases being stronger for more recent cohorts. In the ‘fertility stalled’ countries essentially the same pattern of increase is observed until it is halted for the 1980-85 birth cohort. For that cohort, at each single age, the proportion of childless women is similar to that of the previous cohort (1975-80). Among the most recent birth cohorts, again, the proportion of childless women at each age resumes its earlier increase, irrespective of the fertility stalling status of countries.

Figure 5: Proportions of childless women by age, for 5-year birth cohorts of women and fertility stalling status of countries.

It is too early to explore whether this clear effect of the education stall for women born in 1980-85 on their early childbearing behavior will also result in higher completed cohort fertility at the end of their reproductive career or whether this cohort will later compensate this early boost by lower fertility at higher

• ages. Unfortunately, data for this cohort are truncated at age 25-30. But already the last available data for

this cohort seem to indicate that at least to some degree the declining fertility trends resumed after women reached their mid-20s to bring the cohort back in line with the general trend of declining cohort fertility. However in terms of period TFRs it is clear that this very early childbearing experience of this cohort has

contributed to higher period fertility. If this should indeed be compensated at higher ages, this would strengthen the trend toward lower period fertility in the near future. Why would the effect of stalled education mostly affect young women? A plausible path of causation would be that the cohorts of women who dropped out of school or were never enrolled because of SAPs have entered childbearing earlier than previous cohorts and thus bear more children at earlier ages. In order to test this hypothesis against the possible alternative explanation that the fertility stall was caused by macro level factors such as a slowdown in socio-economic development around 2000 or other period effects (that would have somehow affected younger women more than women above age 25) we will next explore the relative impacts of cohort effects due to an educational stall versus possible period-related shocks through a multivariate analysis.

Multivariate Analysis

In order to disentangle and compare the relative effects of cohort- and period-related factors in explaining the observed fertility stalls, we have employed a duration analysis to study the time to first birth. The duration of interest is the time between 𝑢0 , the age when a woman is at risk of giving the first birth, and the age when she actually has her first birth. Here, 𝑢0 is assumed to be age 12 (30). In this context, the notion

• f duration analysis refers to studying the age-dependent probability of moving from the status of being

childless to the status of having the first child. From the retrospectively collected DHSs data, we have followed 280,776 women from age 12 to the occurrence of first birth, from the eight ‘fertility stalled’

• countries. The follow-up windows have an equal length of 1 year. Those women who did not give their first

birth by the age of 24 or at the time of the interview are right censored. We have fitted a mixed effect discrete time duration model. The choice of the model is justified by the hierarchical structure of the data where individual women 𝑗 are nested within their respective country 𝑑. More information about the model and variables is presented in the appendix. The dependent variable is the relative risk that a woman 𝑗 in country 𝑑 gives her first birth at age 𝑢, given that the birth has not occurred before age 𝑢. The key independent variables at individual level are cohort membership and education attainment, here defined as a categorical variable: whether a woman has at least completed primary education or not. In addition, there are country level indicators related to the following macro-level period changes real GDP/capita, urban population growth (annual percentage) and annual percentage change in under five mortality rates. These macro-level variables are time variant for each age 𝑢 of women at risk of giving birth and a one-year lagged values are considered. Four related duration model have been estimated and compared. First, in the ‘baseline model’, we have constructed a mixed-effect model including only birth cohorts and other individual background characteristics of women as fixed effects. This allows understanding the pattern of inter-cohort progress in the timing of first births without controlling for cohort education and period related factors. Second, in the ‘education-adjusted model’, we add the partial effect of education by including an education variable as an additional individual-level predictor. The third model, ‘period-related model’ includes the unadjusted effects of macro-level period factors (GDP, urbanization and child mortality rates) on top of predictors controlled in the baseline model. These first three models allow to measure the relative impact of education and period factors in explaining the inter-cohort fertility transitions to the first child. Finally, the net effect

and robustness of education and period-related factors are compared in the ‘full model’ that includes individual cohort membership, individual education, period-related macro factors as well as further individual background characteristics(such as such as area of residence and religion of the mother) as fixed effects. The above specification permits to test the hypothesis whether cohorts of women who were potentially affected by SAPs have had their first birth earlier compared to other cohorts. However, we are also interested in the quantum effect of SAPs. Thus, we have further specified and fitted a mixed effect Poisson regression model with the cumulative number of births at age 25 as dependent variables and independent variables as in the duration model (more information in the appendix).

Figure 6. Marginal effect (odds ratio) of belonging to 5-year birth cohort on timing of first birth (since age 12) estimated from mixed effect logistic regression models (as specified in the appendix) and estimates are based on data from ‘fertility stalled’ countries. Odds ratios below 1 show that the cohort needed more years to get first birth than the 1950-55 reference cohort. Error bars are 95% CIs.

Figure 6 presents the estimated marginal effect of membership in 5-year birth cohorts on the probability of giving the first birth in the form of odds ratios. The results of the different models are represented by colors. The complete numerical regression results are presented in appendix table S.2. The results clearly show a distinctive pattern in all models in which the odds of having a first birth first increase sharply for the cohorts born since 1955 in relation to the 1950-55 reference cohort and thereafter show a clearly declining trend. This initial increase is a familiar pattern in the early stage of the demographic transition in which birth rates first rise due to improved health, changes in breastfeeding patterns and disappearance of traditional taboos before they then enter a lasting decline. This steady inter-cohort decline from the 1955-60 cohort onwards come to halt between the cohorts born in 1975-80 and 1980-85. Considering other background characteristics, women who were born in the period 1980-85 have taken even fewer years (since age 12) than the 1975-80 birth cohort to give a first birth. As expected, the odds ratio for the most educated and most recent birth cohort (1985-90) is the lowest. Comparing results across all four models reveals that progress in education was the main factor triggering the inter-cohort declines in the timing of first birth. Once we control for education, the declining patterns in odds ratios across successive birth cohorts born after 1965-70 becomes much weaker. For instance, in the baseline model, the odds of first birth for the

1970-75 cohort was about 15 percent lower than the reference cohort while in the education adjusted model, being born in 1970-75 is associated with a zero drop in the odds of first birth compared to the 1950-55 reference cohort. Similarly, the odds ratio for the cohorts born before 1970-75 increases significantly indicating that those women would have their first birth much sooner than the reference cohort if they had not received better education. On the other hand, the addition of period related macro-level indicators such as GDP on the baseline model hardly affects the marginal effect of belonging to a certain birth cohort on the timing of first birth. Of the three period-related indicators, only child mortality rates yield significant

• effects. In addition, from the full model where we include both period macro variables and individual

education variables, the marginal effect of women’s education remains strong and robust while the effect

• f period related factors are substantially weaker. Further sensitivity runs with somewhat modified models

and variable definitions are given in the appendix. These results all support the hypothesis that education rather than period effects operating at the national level were the key drivers of the timing of entering reproduction. In a final step we also assess the relative effects on the quantum of early childbearing by estimating a model explaining the cumulative number of births experienced by women have up to the age of 25. The results are presented in the Appendix. The incidence rate for each successive birth cohort is significantly lower than for the previous birth cohort. Here again, the 1980-85 birth cohort comes out as an exception having as many births up to age 25 as compared to the 1975-80 birth cohort. In the education-adjusted model, we add women’s individual educational attainment as an additional predictor and observe that the cross-cohort decline becomes much less pronounced evidencing the role of cohort educational progress in triggering the

• bserved cohort fertility declines.

Conclusions Due to the great momentum of population changes the prospects for future population growth in Africa and consequently in the world greatly depends of the fertility trajectory in Africa over the coming years. For assessing those trends it greatly matters what general approach is taken to population projections and in particular whether heterogeneity of the population with respect to its changing education composition is taken into account. The projections produced by the United Nations Population Division only consider the age and sex structure of the population and base their assumptions about future fertility trends on an extrapolative statistical model of the TFR, which is highly sensitive to recent trends in the TFR. The medium scenario of the most recent assessment of 2017 projects based on an extrapolative statistical model

• f TFR that the number of people on the planet will rise to 9.8 billion in 2050 and that Sub-Saharan Africa

will be responsible for more than half of the world’s population growth over the next 35 years (31). Compared to the 2010 UN revision, these recent projections result in world population that is 0.5 billion larger in 2050. The difference in the projection outcomes stems for a large part from the extrapolation of the recent trends in fertility levels in many sub-Saharan African countries that experienced slowed or stagnating fertility declines as discussed in this paper. However, the relevance of these stalls for future fertility trends greatly depends on the nature and causes of the stalls. In the above given analysis we have found strong empirical support for the hypothesis that this fertility stall was caused by a temporary stall in the education of female cohorts born in the early 1980s. We have shown clear evidence that the for most of the countries experiencing fertility stalls around 2000 there have been stalled in the education improvement of the female cohorts that entered the prime childbearing ages around

that time. More detailed analysis of cohort-specific age patterns of first births as well as higher order birth and multivariate models including possible macro-level period effects further confirm the pattern that indeed exceptional education experience of the birth cohort 1980-85 is a driver of the observed fertility

• stalls. Since the more recent cohorts of young women have again picked up in terms of education this

finding suggests that for the future we may expected an acceleration of the fertility decline as the subsequent better-educated cohorts of women move into the main childbearing ages. This finding is also relevant for the ongoing discussion as to whether population projections should be carried out by only breaking down by age and sex or whether educational attainment should be included as a third demographic dimension (32).The evidence discussed in this paper illustrates well that in the case of education discontinuities the assumed future fertility trajectories are different when heterogeneity by level of education is explicitly factored into the model as compared to disregarding heterogeneity and only observing aggregate TFR trends. The most recent findings from the Kenya DHS are a point in case. A country that in the late 1970s had a TFR of 8.1 – considered to be the highest in the world – and then after an initial decline has a decade of stagnant fertility around 2000 at a TFR of 4.6-4.9 all of a sudden had a rather steep dive to 3.9 for the three years preceding the 2014 DHS (33, 34). No statistical extrapolation model based on TFR alone could predict this recent decline. The education-specific analysis discussed in this paper makes it plausible and expected because the female cohorts that had to experience the stall in education expansion are being replaced in the prime childbearing ages by new and better-educated cohorts. While the evidence for this education-fertility link at the cohort level seems to be very robust, the evidence for directly blaming the SAPs for the education stalls is less certain. This is due to the lack of reliable statistical information about how precisely the SAPs in individual countries lead to cuts in the education budgets and how these cuts triggered down to effects on school enrolment rates. While the timing of the known implementation of SAPs and that of the education stalls clearly make it plausible to assume a direct causal link this is still short of a proof. In any case, such assignment of blame for the education stall is only

• f historical interest, given that IMF and World Bank have in the meantime learned their lessons (11).

However, what is still relevant for the future is the continuing cohort effect of educational attainment, which will be relevant far beyond fertility and affect the future health, and income of the effected cohorts for decades to come. The future population growth in Africa will also be relevant for the rest of the world. Whether it will “only” increase to two billion as shown by optimistic scenarios that assume successful implementation of the sustainable development goals (25 and PNAS-SDG paper) or more pessimistic scenarios assuming slow or stalled development and thus resulting in 4-5 billion Africans by the end of the century, will first of all impact on the future of living conditions and quality of life of Africans. But it will also affect other continents due to out-migration pressure, global environmental impacts and of course the efforts needed to live up to the promise to eradicate poverty, hunger and premature death in all corners of the planet. Continued rapid population growth will make this an up-hill battle. Education is currently underfunded, particularly in Africa (Ref Lutz and Klingholz). A new effort for massive investment in education - particularly of girls - will not only help to slow this growth but will also empower Africans and create the human capital needed for rapid social and economic development and sustainable increases in human well- being.

References

1. Bongaarts J, Casterline J (2013) Fertility transition: Is sub-Saharan Africa different? Popul Dev Rev 38:153–168. 2. Moultrie TA, et al. (2008) Refining the criteria for stalled fertility declines: an application to rural KwaZulu-Natal, South Africa, 1990-2005. Stud Fam Plann 39(1):39–48. 3. Shapiro D, Gebreselassie T (2008) Fertility transition in sub-Saharan Africa: Falling and stalling. Afr Popul Stud 23:3–23. 4. Agyei-Mensah S (2007) New Times, new families: The Stall in ghanaian fertility (Union for African Studies, Arusha, Tanzania), p 24. 5. Bongaarts J (2008) What can fertility indicators tell us about pronatalist policy options? Vienna Yearb Popul Res 2008:39–55. 6. Westoff CF, Cross AR (2006) The stall in the fertility transition in Kenya (Macro International Inc, Calverton, Maryland, USA) Available at: https://www.dhsprogram.com/publications/publication- as9-analytical-studies.cfm. 7. Goujon A, Lutz W, KC S (2015) Education stalls and subsequent stalls in African fertility: A descriptive overview. Demogr Res 33(47):1281–1296. 8. World Bank (1981) Accelerated development in Sub-Saharan Africa: An agenda for action (The World Bank, Washington, D.C.) Available at: http://documents.worldbank.org/curated/en/702471468768312009/pdf/multi-page.pdf. 9. Husain I (1994) Adjustment in Africa: Lessons from country case studies ed Faruquee R (World Bank, Washington, D.C.).

• 10. lhonvbere J (1996) Economic crisis, structural adjustment and Africa’s future? Women Pay the

Price: Structural Adjustment in Africa and the Caribbean, ed Thomas-Emeagwali G (Africa World Press, Trenton, NJ), pp 133–154.

• 11. World Bank (1992) Adjustment lending and mobilization of private and public resources for growth

(The World Bank, Washington, D.C.) Available at: http://documents.worldbank.org/curated/en/370601468739765599/pdf/multi-page.pdf.

• 12. Cornia GA, Jolly R, Stewart F eds. (1988) Adjustment with a Human Face: Volume 2: Country

Case Studies (Clarendon Press, Oxford Oxfordshire; New York). 1 edition.

• 13. Stromquist NP (1999) The impact of structural adjustement programmes in Africa and Latin
• America. Gender, Education and Development: Beyond Access to Empowerment, eds Heward C,

Bunwaree S (Zed Books Ltd, London and New York), pp 17–32.

• 14. Rose P (1995) Female education and adjustment programs: A crosscountry statistical analysis.

World Dev 23(11):1931–1949.

• 15. Przeworski A, Vreeland JR (2000) The effect of IMF programs on economic growth. J Dev Econ

62(2):385–421.

• 16. Lockheed ME, Verspoor AM (1990) Improving Primary Education in Developing Countries

(Oxford University Press, Washington, D.C.).

• 17. Alexander NC (2001) Paying for Education: How the World Bank and the International Monetary

Fund Influence Education in Developing Countries. Peabody J Educ 76(3–4):285–338.

• 18. Daoud A, et al. (2017) Impact of International Monetary Fund programs on child health. Proc Natl

Acad Sci 114(25):6492–6497.

• 19. Subramanian SV, Neve J-WD (2017) Social determinants of health and the International Monetary
• Fund. Proc Natl Acad Sci 114(25):6421–6423.
• 20. Bongaarts J (2010) The causes of educational differences in fertility in Sub-Saharan Africa. Vienna

Yearb Popul Res 8:31–50.

• 21. Cochrane SH (1979) Fertility and education. What do we really know? (Johns Hopkins University

Press, Baltimore, Maryland).

• 22. Castro Martin T (1995) Women’s education and fertility: Results from 26 demographic and health
• surveys. Stud Fam Plann 26(4):187–202.
• 23. Fuchs R, Goujon A (2014) Future fertility in high fertility countries. World Population and Human

Capital in the 21st Century, eds Lutz W, Butz WP, KC S (Oxford University Press, Oxford), pp 147–225.

• 24. Lutz W, Skirbekk V (2014) How education drives demography and knowledge informs projections.

World Population and Human Capital in the 21st Century, eds Lutz W, Butz WP, KC S (Oxford University Press, Oxford), pp 14–38.

• 25. Casterline JB, Bongaarts JP (2017) Fertility Transition in Sub-Saharan Africa (The Population

Council).

• 26. Bongaarts J (2017) Africa’s Unique Fertility Transition. Popul Dev Rev 43:39–58.
• 27. Askew I, Maggwa N, Obare F (2017) Fertility Transitions in Ghana and Kenya: Trends,

Determinants, and Implications for Policy and Programs. Popul Dev Rev 43:289–307.

• 28. United Nations (2015) World Population Prospects: The 2015 Revision (Department of Economic

and Social Affairs, Population Division, New York, NY) Available at: http://esa.un.org/unpd/wpp/.

• 29. World Bank (2017) World Development Indicators. Available at: http://data.worldbank.org/data-

catalog/world-development-indicators [Accessed July 18, 2017].

• 30. Bongaarts J, Frank O, Lesthaeghe R (1984) The proximate determinants of fertility in sub-Saharan
• Africa. Popul Dev Rev 10(3):511–537.
• 31. United Nations (2017) World Population Prospects: The 2017 Revision, Key Findings and Advance

Tables (United Nations Population Division | Department of Economic and Social Affairs, New York) Available at: https://www.popline.org/node/639412 [Accessed September 25, 2017].

• 32. Lutz W, Butz WP, KC S (2014) World Population and Human Capital in the Twenty-First Century

(Oxford University Press, Oxford).

• 33. Lutz W (1991) Future demographic trends in Europe and North America: What can we assume

today? (Academic Press, London, UK).

• 34. DHS Kenya (2015) Kenya Demographic and Health Survey 2014 (ICF International, Calverton,

Maryland, USA) Available at: http://www.dhsprogram.com/publications/publication-FR308-DHS- Final-Reports.cfm.

Supporting Information

Appendix Table S1.

Variable definitions and Sources

Variable /Indicator Definition Source

𝐷𝑝ℎ𝑝𝑠𝑢𝑑,𝑗

Birth cohort of women 𝑗 of country 𝑑 in five- year intervals (1950-55, 1955-60, 1960- 65,…1985-90) Demographic and Household Surveys Cohort ASFR Age specific fertility rates by single year of age and 5-year birth cohort of women Computed from Demographic and Household Surveys Cohort education Share of women in the cohort with at least some primary education Computed from Demographic and Household Surveys

𝐹𝑒𝑣𝑑𝑑,𝑗

Dummy variable: = 1 if women 𝑗 from country 𝑑 have some formal education at the time of the survey ;=0, otherwise Demographic and Household Surveys

𝐻𝐸𝑄/𝑑𝑏𝑞𝑗𝑢𝑏 𝑑,𝑢−1

Logarithm of Real Gross domestic product per capita (in mil. 2011US$) of the country 𝑑 at age 𝑢 − 1 of the woman under consideration It is log transformed to correct for the skewed distribution. Penn World Table 9.1.

𝑉𝑠𝑐𝑏𝑜𝑑,𝑢−1

urban population growth (annual percentage change) of country 𝑑 at age 𝑢 − 1 of the woman under consideration WDI, 2017 𝑉5𝑁𝑆𝑑,𝑢−1 Annual growth rate of under-five mortality rates of country 𝑑 at age 𝑢 − 1 of the woman under consideration WDI, 2017 Religion Dummy variable: = 1 if women was Muslim at the time of the survey.=0, otherwise Demographic and Household Surveys Place of residence Area of residence (rural/urban) of women at the time of the survey Demographic and Household Surveys

Model Specifications

• 1. The duration analysis of time to first birth

In order to disentangle and compare the relative effects of cohort- and period-related factors in explaining the observed fertility stalls, we have employed a duration analysis to study the time to first birth. The duration of interest is the time between 𝑢0 , the age when a woman is at risk of giving the first birth, and the age when she actually had her first birth or right censored2. Here, 𝑢0 is assumed to be age 12. In order to estimate the model, we first convert the data into person-year format. A woman who has her first birth at age n contributes 𝑜 − 𝑢0 observations to the sample, one observation for each year at-risk until she has her first birth and exits the sample. The duration models are specified in the following way:

Logit(hi,c,t) = log ( first birthi,c,t 1 − first birthi,c,t ) = α(𝑢) + 𝛾1𝐷𝑝ℎ𝑝𝑠𝑢𝑑,𝑗 + β2𝐹𝑒𝑣𝑑𝑑,𝑗 + 𝐻𝐸𝑄/𝑑𝑏𝑞𝑗𝑢𝑏𝑑,𝑢−1 + 𝛾4𝑉𝑠𝑐𝑏𝑜𝑑,𝑢−1 + β5𝑉5𝑁𝑆𝑑,𝑢−1 + 𝑌′

𝑗𝜌 + 𝑍′ 𝑑𝜈𝑑 − − − − − − − (1)

Where the dependent variable is the relative risk that a women 𝑗 in country 𝑑 gives her first birth at age 𝑢 , given that the birth has not occurred before age 𝑢: ℎ𝑢𝑑𝑗 = 𝑞𝑠(𝑍

𝑗,𝑑,𝑢 = 1 / 𝑍 𝑗,𝑑,𝑡 = 0, 𝑡 < 𝑢) − − − − − − − − − − − (2)

Where 𝑍

𝑗,𝑑,𝑢 is a binary response variable indicating whether a women 𝑗 from country 𝑑 has given the first

birth at age 𝑢. The baseline hazard α(𝑢) is specified as piecewise constant with 12 equally spaced (single year) age intervals. The key independent categorical variable 𝐷𝑝ℎ𝑝𝑠𝑢𝑑,𝑗 is the 5-year birth cohort of women 𝑗 in country 𝑑. Women’s educational attainment 𝐹𝑒𝑣𝑑𝑑,𝑗 is defined as a categorical variable: whether a woman has at least completed primary education or not. Predictors for period related changes 𝑆𝐻𝐸𝑄/𝑑𝑏𝑞𝑗𝑢𝑏𝑑,𝑢−1 , 𝑉𝑠𝑐𝑏𝑜𝑑,𝑢−1 and 𝑉5𝑁𝑆𝑑,𝑢−1 𝑏re country level variables measuring respectively real GDP/capita , urban population growth (annual percentage) and annual percentage change in under five mortality rates, in the year before the follow-up age of women (t). These three variables enter the model as continuous variables in their one year lagged form. Matrix X contains other individual background characteristics such as area of residence (rural/urban) and religion (Muslim/non-Muslim). The parameter 𝜈𝑑 corresponds to a country random effect, which assumes a normal distribution with constant variance.

2 Those women who hadn’t had a birth by age 24 or at the time of the interview are right censored. It is considered as

if they were followed until age 24 but not observed thereafter. Similarly those aged below 24 but hadn’t had a child at the time of the interview are also right censored at the exact age at interview.

• 2. Mixed effect Poisson Regression model

The Mixed effect duration analysis permits to the testing of the hypothesis whether the cohorts of women who were potentially affected by SAPs have had their first birth earlier compared to other cohorts. However, we are also interested on the quantum effect of educational discontinuities. Thus, we have specified and fitted a mixed effect Poisson regression model as follow. log(CNB25𝑗,𝑑/𝑢) = 𝐷𝑝ℎ𝑝𝑠𝑢𝑑,𝑗 + 𝐹𝑒𝑣𝑑𝑑,𝑗 + 𝑄′𝑑,𝑢−1𝜌+𝑌′𝑗𝜌 + 𝑍′𝑑𝜈 − − − − − − − − − (3) The dependent variable is the cumulative number of births by age 25, from women 𝑗 residing in country 𝑑. The key independent variables are 𝐷𝑝ℎ𝑝𝑠𝑢𝑑,𝑗 is the 5-year birth cohort of women 𝑗 in country 𝑑 and Woman’s educational attainment 𝐹𝑒𝑣𝑑𝑑,𝑗 defined as a categorical variable: whether a woman has at least some formal education or not. Matrix X contains other individual background characteristics such as area

• f residence (rural/urban) and religion (Muslim/non-Muslim). The parameter 𝜈𝑑 corresponds to a country

random effect, which assumes a normal distribution with constant variance

Appendix Table S.2: Mixed effect logistic regression results: Transition to First Birth (odds

ratios). The dependent variable is the Relative risk that a woman 𝑗 in country 𝑑 had her first birth at

age 𝑢

Baseline Model Education Adjusted Period Adjusted Mutual Adj. VARIABLES OR(SE) OR(SE) OR(SE) OR(SE) Birth cohort (ref=1950- 55) 1955-60 1.125*** 1.182*** 1.146*** 1.19*** (0.0169) (0.0170) (0.0194) (0.0194) 1960-65 1.016 1.126*** 1.054** 1.150*** (0.0154) (0.0155) (0.0177) (0.0177) 1965-70 0.939*** 1.101** 0.979 1.142*** (0.0150) (0.0151) (0.0172) (0.0172) 1970-75 0.851*** 1.007 0.911*** 1.089*** (0.0148) (0.0149) (0.0170) (0.08) 1975-80 0.808*** 0.966** 0.860*** 1.049* (0.0148) (0.0150) (0.0171) (0.0171) 1980-85 0.813*** 0.977** 0.846*** 1.043 (0.0148) (0.0150) (0.0171) (0.0171) 1985-90 0.752*** 0.925** 0.766*** 0.962** (0.0150) (0.0152) (0.0169) (0.0171) Religion (ref= non-Muslim) Muslim 1.342*** 1.169*** 1.406*** 1.197*** (0.006) (0.007) (0.0273) (0.007) Residence (ref=Urban) Rural 1.565*** 1.357*** 1.597*** 1.378*** (0.005) (0.0051) (0.0110) (0.005) Women’s education (ref= no) At least some primary education 0.590*** 0.589*** (0.057) (0.006) Period Factorsi U5MRt_1 0.973*** 0.977*** (0.001) (0.0185) GDP/capitat_1 0.993 1.025* (0.019) (0.0185) Urban t_1 0.993 1.007 (0.002) (0.0185) Baseline(PWC) YES YES YES YES Constant YES YES YES YES Random Effects MOR MOR MOR MOR Level 2 (Country) 1.20 1.21 1.16 1.16 Country 8 8 8 8

Person year Observation 1930025 1928899 1787191 1793740

i. variables are standardized Appendix Table S.3: Mixed effect Poisson regression results. The dependent variable is Cumulative number of births by age 25 (Incidence Rate Ratio)

Baseline Model Educ Adjust VARIABLES IRR(SE) IRR(SE) Birth cohort (ref=1950- 55) 1955-60 1.017 1.043*** (0.0091) (0.0094) 1960-65 0.951*** 1.012 (0.0084) (0.0087) 1965-70 0.881*** 0.972** (0.0083) (0.0087) 1970-75 0.835*** 0.923 (0.0083) (0.0087) 1975-80 0.816*** 0.913** (0.0084) (0.0088) 1980-85 0.819*** 0.918** (0.0086) (0.0090) 1985-90 0.793*** 0.897** (0.0098) (0.0102) Religion (ref= non-Muslim) Muslim 1.230*** 1.130*** (0.0046) (0.005) Residence (ref=Urban) Rural 1.320*** 1.196*** (0.0036) (0.004) Women’s education (ref= no ) At least some primary education 0.681*** (0.004) Baseline(PWC) YES YES Constant YES YES

• Country 8 8

Observations AIC 182,700 628278.7 167,082 565,108.9

Appendix Figure S.1: Average number of births by age 25 for subsequent cohorts of women in selected African countries, according to fertility stalling status of countries Appendix Figure S.2: Marginal effect (odds ratio) of birth cohort on timing of first birth (since age 12) estimated from the ‘baseline’ mixed effect logistic regression model as specified in the method section. Only women with at least completed primary education are considered. Odds ratios below 1 show that the cohort needed more years to first birth than the 1950-55 reference cohort.

Exploring the Link between Structural Adjustment Programs, Educational Discontinuities and Stalled Fertility in Sub-Saharan Africa