Abstract

Demographic scholarship suggests that schooling plays an important role in transforming fertility preferences in the early stages of fertility decline. However, there is limited evidence on the relationship between schooling and fertility preferences that addresses the endogeneity of schooling. I use the implementation of Universal Primary Education (UPE) policies in Malawi, Uganda, and Ethiopia in the mid-1990s to conduct a fuzzy regression discontinuity analysis of the effect of schooling on women’s desired fertility. Findings indicate that increased schooling reduced women’s ideal family size and very high desired fertility across all three countries. Additional analyses of potential pathways through which schooling could have affected desired fertility suggest some pathways—such as increasing partner’s education—were common across contexts, whereas other pathways were country-specific. This analysis contributes to demographic understandings of the factors influencing individual-level fertility behaviors and thus aggregate-level fertility decline in sub-Saharan Africa.

Introduction

A well-documented negative association exists between schooling and fertility in sub-Saharan Africa (Ainsworth et al. 1996; Bongaarts 2010; Castro-Martin 1995; Kravdal 2002; Lloyd et al. 2000). Considerable research attention has been devoted to understanding why this negative relationship persists in Africa and elsewhere in the world. Demographic scholarship suggests that preference for smaller families is an important precursor to fertility decline (Coale 1973; Cochrane 1979; Easterlin 1975; Knodel and van de Walle 1986). Further research indicates that schooling plays an important role in changing fertility preferences through the diffusion of new information (Reed et al. 1999), promotion of new norms (Caldwell 1976, 1980), and exposure to new social networks and social interactions (Bongaarts and Watkins 1996; Mare 1991). Nonetheless, establishing a causal relationship between schooling and fertility preferences is difficult given the many characteristics—such as socioeconomic status (SES), cognitive ability, or family preferences—that likely predict both schooling and fertility preferences.

A growing number of studies on the relationship between schooling and behavioral fertility outcomes in sub-Saharan Africa have attempted to deal with the endogeneity of schooling by employing methods for causal inference (McQueston et al. 2013). These studies have taken two forms: (1) randomized controlled trials of school incentive programs (Baird et al. 2010; Duflo et al. 2012; Dupas 2011); and (2) quasi-experimental studies using instrumental variable and regression discontinuity techniques that exploit random variation in access to schooling via introduction of school reforms (Ferre 2009; Osili and Long 2008; Zanin et al. 2015), distance from school (Alsan and Cutler 2013), or performance on qualifying examinations (Ozier 2010). Overall, the evidence indicates a robust negative relationship between schooling and fertility at both primary and secondary school levels in a number of African countries.1

Few studies on the relationship between schooling and fertility preferences have used methods for causal inference to address the endogeneity of schooling. Building on this research gap, I use the implementation of Universal Primary Education (UPE) policies in Malawi, Uganda, and Ethiopia in the mid-1990s to explore the effect of schooling on women’s desired fertility. These three countries were chosen because they were among the first adopters of UPE policies in Africa, allowing for comparison among women who had completed their schooling careers and started childbearing. Comparing multiple countries provides a way of assessing the generalizability of the findings and allows for exploration of whether there was something universal about mass schooling that affected desired fertility (Caldwell 1980), or whether effects were context specific (Moultrie and Timaeus 2014).

Using a fuzzy regression discontinuity research design, my analysis focuses on three research questions. First, did increased schooling have an effect on women’s ideal family size and very high desired fertility (six or more children) in Malawi, Uganda, and Ethiopia? Second, what were the potential pathways through which schooling could have affected desired fertility in each country? Third, to what extent were schooling effects consistent across countries as opposed to country-specific?

Schooling, Fertility Preferences, and Fertility Decline

Demographers assert that diffusion of new ideas and preferences related to contraception use, marital fertility limitation, and ideal family size play an important role in precipitating fertility decline (Cleland and Wilson 1987; Coale 1973; Cochrane 1979; Easterlin 1975; Knodel and van de Walle 1986). Schools have been hypothesized to be important sites for socializing students into new ideas about sexual behavior, fertility regulation, and family norms (Caldwell 1976, 1980). In mid-nineteenth century Europe, schools played an essential role in promoting the cultural and linguistic homogeneity that would culminate in the onset of the fertility transition (Watkins 1991). Caldwell (1980) asserted a causal link between the introduction of mass schooling and fertility decline in contemporary countries in the early stages of the fertility transition.

Schooling could affect desired fertility and other related fertility preferences through a number of pathways. Sociocognitive theory posits that changes in preferences arise from changes in knowledge, changes in attitudes, or some combination of the two (Davies and Macdowall 2006). In this case, changes in knowledge could include knowledge about family planning and fertility regulation (Cochrane 1979), and changes in attitudes could include the acceptability of using family planning (Cleland and Wilson 1987) or the acceptability of having smaller families (Caldwell 1980). Drawing on the literature, I highlight four major pathways through which schooling could affect women’s desired fertility via changing knowledge and attitudes. These pathways are not necessarily mutually exclusive, and this is not an exhaustive list of all potential pathways but is instead the start of an exploratory model to be built upon in further research.

First, schooling may directly improve student’s knowledge about family planning and reproductive health through sexual education or information about biology, reproduction, and transmission of sexually transmitted infections (Cochrane 1979). HIV risk–reduction programs in African schools have been shown to positively affect knowledge about HIV transmission (Gallant and Maticka-Tyndale 2004). School-based sexual education and condom use demonstrations were positively associated with knowledge of correct condom use in Malawi, Uganda, Burkina Faso, and Ghana (Bankole et al. 2007).

Second, schooling may expose students to new gender and family norms via textbooks, Euro-centric curriculums, exposure to female teachers as role models, and so on (Caldwell 1976, 1980). Caldwell theorized that mass schooling socializes children into a new type of morality that prioritizes the model of the Western nuclear family over the traditional extended family, leading to preference for smaller families. Although this hypothesis has received considerable research attention, it has remained difficult to test empirically (Lloyd et al. 2000).

Third, schooling may positively affect women’s access to and engagement with mass media—and by extension, new gender and family norms— if mass media is used in classrooms or if school provides women with the resources needed to access mass media outside the classroom (e.g., literacy and higher income) (Reed et al. 1999). Evidence from India, Indonesia, and Brazil suggests that increased exposure to television is associated with significant increases in modern contraceptive use and decreases in pregnancy (Dewi et al. 2013; La Ferrara et al. 2012; Jensen and Oster 2009). Researchers have attributed these changes to the increased acceptability of small families visible in soap operas and other popular television programs.

Fourth, schooling may alter women’s relationships with peers and romantic partners, which in turn may shape attitudes and knowledge about fertility. Social interaction plays an important role in transforming ideas about fertility and contraception (Bongaarts and Watkins 1996), and a large literature examines the effects of school-based social and sexual networks on later life outcomes (Mare 1991; Mare and Maralani 2006; Mouw 2006). Evidence from an experimental study in Malawi indicates that increased schooling led to decreases in the age difference between adolescent women and their partners and decreases in overall sexual frequency (Baird et al. 2012). In an experimental study in Kenya, increased schooling led to adolescent relationships characterized by higher levels of commitment (Duflo et al. 2012).

This is not to suggest that the only way school could affect fertility decline is through changing desired fertility. An alternative perspective is offered by neoclassical economic models of fertility, which argue that increases in women’s schooling lead to wage increases and, correspondingly, increases in the opportunity costs of having a large number of children (Becker 1981; Becker and Lewis 1973). This increase in opportunity costs induces women to shift from having a large quantity of children to having fewer higher-quality children. In the strictest interpretation of neoclassical models, preferences are treated as fixed and exogenous, although subsequent scholarship has attempted to incorporate endogenous preferences into neoclassical models (Easterlin and Crimmins 1985; Pollak and Watkins 1993).

Over the years, demographers have moved away from the notion that any one monolithic theory will explain fertility decline, stressing instead the importance of multiple contextual factors, including changing preferences and changing economic opportunities (Hirschman 1994; Mason 1997). Caldwell (1980) attempted to bridge ideational and economic theories of fertility decline by arguing that mass schooling initiates the fertility transition through a combination of ideational factors (such as preference for smaller families) and economic factors (particularly, the high costs of children’s schooling).

Conceptual and Methodological Issues in Measuring Desired Fertility

Fertility preferences remain key to demographic understandings of a wide range of fertility behaviors. However, there is no one consensus for how to interpret stated fertility preferences, such as desired fertility. Conceptually, demographers distinguish among fertility preferences, intentions, and expectations, although in practice the three terms are often conflated (Ryder and Westoff 1967; Thomson 1997). Changes in fertility preferences are typically viewed as a prerequisite for changing intentions and expectations (Cochrane 1979; Easterlin 1975). Nonetheless, there is ongoing dispute about the extent to which preferences correspond with behaviors and whether high fertility reflects high desire for children (Pritchett 1994) or high levels of unmet need for contraception (Bongaarts 1994).

A number of methodological and theoretical issues arise when trying to assess desired fertility through questions about ideal family size. Debate exists about whether fertility is consciously regulated and what that means for interpreting stated fertility preferences. Historically, demographers viewed fertility decline as possible only when fertility regulation enters into an individual’s “calculus of conscious choice” (Coale 1973:65). However, a lack of conscious thinking about family size does not necessarily mean that fertility is unregulated, particularly given social customs that influence fertility via marriage and divorce (Hirschman 1994). Cross-national evidence indicates a positive relationship between women’s schooling and likelihood of providing a numeric, as opposed to nonnumeric, response to ideal family size (Bachan and Frye 2013; Hayford and Agadjanian 2011; McCarthy and Oni 1987; Olusanya 1971; Riley et al. 1993). These findings provide some evidence that schooling influences women’s conscious thinking about desired fertility. Nonetheless, women may not have a concrete numerical reproductive target, even if they are actively engaged in fertility regulation (Ryder 1973).

Furthermore, women’s responses to ideal family size may change over the life course. Some researchers attribute changes in ideal family size to ex-post facto rationalization based on current number of offspring (Pritchett 1994) or measurement error (Bankole and Singh 1998). Other researchers have argued that ideal family size and other fertility preferences evolve continuously and dynamically throughout the life course based on employment, relationships, existing offspring characteristics, and other contextual factors (Bulatao 1981; Hayford 2009; Heiland et al. 2008; Lee 1980; Liefbroer 2009; Udry 1983; Yeatman et al. 2013).

Finally, the extent to which ideal family size reflects personal preference as opposed to broader societal norms and institutions remains unclear. Ryder (1973:505) argued that the concept of ideal family size is inherently sociological and is shaped by the dominant norms and paradigms of the times. Johnson-Hanks (2007) asserted that researchers need to understand the complex systems of meaning that underlie declining fertility rather than treating fertility decline as a rational process driven by desire for smaller families. Building on this argument, Moultrie and Timaeus (2014) called for greater understanding of the country-specific norms and historical institutions that regulate fertility and lead to decline. Although my analyses cannot capture all aspects of these important critiques, they do proceed from the postulate that changes in stated fertility preferences reflect important underlying social processes that have implications for changing fertility.

Universal Primary Education in Malawi, Uganda, and Ethiopia

I use the implementation of Universal Primary Education (UPE) policies in Malawi, Uganda, and Ethiopia to assess the effect of schooling on desired fertility. The defining feature of UPE policies in all three countries was the elimination of school fees. The direct and indirect costs of schooling are large obstacles to school enrollment for poor populations (Avenstrup et al. 2004; World Bank 2009). User fees, in particular, have been identified as a major obstacle to school enrollment (Bentaouet Kattan and Burnett 2004). Malawi, Ethiopia, and Uganda were among the earliest adopters of UPE in sub-Saharan Africa. In the following decades, school fees have also been eliminated in Cameroon (1999/2000), Lesotho (2000), Tanzania (2001), Zambia (2002), Kenya (2003), Mozambique (2004), and Ghana (2005) (Bentaouet Kattan 2006; World Bank 2009).

Starting in 1991, Malawi adopted a sequential approach to the elimination of primary school fees that entailed provision of fee waivers for grade 1 in the first year of implementation and for grade 2 in the second year (Bentaouet Kattan 2006; World Bank 2009). In 1994, this policy was replaced by a “big bang” approach in which primary school fees were eliminated across all grades starting in September 1994. In Malawi, I use 1994 as the year of discontinuity for the analysis because this is the year when girls at older ages who are at highest risk of dropping out of school were affected by the reform.

By contrast, Uganda adopted a “big bang” approach to elimination of primary school fees from the start, effectively eliminating fees at all levels of primary school starting January 1997 (Bentaouet Kattan 2006; Deininger 2003; Grogan 2008). For Uganda, I use 1997 as the year of discontinuity for the analysis. Both Uganda and Malawi also eliminated indirect fees, such as contributions to school development funds, and the requirement to wear uniforms to school (Kadzamira and Rose 2003; World Bank 2009).

In Ethiopia, fees were eliminated for primary school and the first two years of secondary school in 1994 (Method et al. 2010). However, it took about a year for fee abolition to be forwarded to rural areas; thus, complete policy implementation did not occur until 1996 (World Bank 2009). In Ethiopia, I take 1994 as the year of implementation for the regression discontinuity analysis with the caveat that some respondents may not have been exposed to UPE until 1996; the implications of this for the analysis are discussed in the upcoming section on first-stage results. Nonetheless, the UPE policy in Ethiopia also eliminated fees for two years of secondary school, which means that the same respondents who were age 13 and younger in 1994 would have been age 15 and younger in 1996 and still would have been eligible for free schooling when the policy was universally implemented. The fact that the Ethiopian sample was exposed to free primary school and two years of free secondary school does have certain implications for comparability of findings between countries, and I explore this further in the Discussion section.

Evidence from all three countries suggests that UPE programs were successful at improving enrollment. In the year following the 1994 “big bang” implementation in Malawi, 1 million new children entered primary school, and total primary school enrollment increased by more than 50 % (World Bank 2009). In Uganda, primary school attendance rates went from 62 % in 1992 (before UPE) to 84 % by 1999 (Deininger 2003). In Ethiopia, the primary school gross enrollment rate increased from 26 % in the year when UPE was implemented to 80 % by 2004/2005 (World Bank 2009). UPE policies allowed entry at any grade, which led to large increases in enrollment amongst older children at higher grades. For example, enrollment increased by 79 % in the final year of primary school (Standard 8) in Malawi following UPE introduction (World Bank 2009).

UPE policies were particularly effective at improving enrollment among poor households and in poor regions, especially for girls. Prior to elimination of fees in Malawi, in 1990/1991, the primary school gross enrollment rate of children from the wealthiest income quintile was nearly double that of children from the poorest income quintile (Al Samarrai and Zaman 2007).2 Among children in the poorest income quintile, the primary school gross enrollment rate of males was 65 %, compared with 51 % for females. Following elimination of fees, in 1997/1998, the wealth gap in the primary school gross enrollment rate between the poorest and wealthiest income quintiles had been almost totally eliminated, and the primary school gross enrollment rates of males and females in the poorest wealth quintile stood at 125 % and 109 %, respectively (Al Samarrai and Zaman 2007).

In Uganda before UPE, primary school attendance of children in the poorest income quintile was 46 %, compared with 82 % for children in the wealthiest income quintile (Deininger 2003); school attendance for girls in the poorest income quintile was 40 %, compared with 51 % for boys in the poorest income quintile and 80 % for girls in the wealthiest income quintile (Deininger 2003). In the year after UPE implementation, school attendance for children in the poorest income quintile rose to 84 %, compared with 89 % for children in the wealthiest income quintile (Deininger 2003). Following UPE, school attendance for girls in the poorest wealth quintile rose to 75 % (Deininger 2003).

In the period directly following UPE implementation in Ethiopia, the average annual growth in enrollment in Afar and Somali—two of most socioeconomically disadvantaged regions of the country—was twice that of the national average (World Bank 2009). In Afar and Somali, the female primary school gross enrollment rate went from 6 % to 18 % in the decade following UPE implementation (World Bank 2009). At the national level, the female primary school gross enrollment rate went from 20 % to 71 % over the same period, whereas the male primary school gross enrollment rate went from 32 % to 88 % (World Bank 2009).

UPE policies in all three countries have been criticized for improving access to school at the expense of deteriorating school quality (Deininger 2003; Grogan 2008; Kadzamira and Rose 2003). This is an important distinction because poor African school quality has proven to be a challenge for learning and cognitive outcomes (Pritchett 2013). This analysis considers only issues of school access and not school quality, a worthy topic for a separate investigation. It is also important to note that even in the absence of fees, a number of indirect costs remained—including materials, uniforms, and loss of labor in household livelihoods. Attempts in Malawi and Uganda to remove indirect costs—such as school materials—were not uniformly implemented; thus, removal of fees did not make schooling fully costless (Deininger 2003; Kadzamira and Rose 2003).

Empirical Strategy

Sample

The Demographic Health Surveys (DHS) are nationally representative standardized household-based surveys collected by ICF International in collaboration with the host country governments. Data for this analysis came from the most recent DHS: 2010 in Malawi, and 2011 in Uganda and Ethiopia. In this analysis, I compare women just below primary school exit age at time of UPE implementation (age 11–13) with women just above primary school exit age at time of UPE implementation (age 14–16).3 In Malawi, 2,520 women were exposed to UPE while of primary school age (treatment), and 2,143 women were not exposed to UPE while of primary school age (control) (Table 1). At survey in 2010, the sample ranged in age from 26 to 32. In Uganda, 979 women were exposed to UPE while of primary school age (treatment), and 862 women were not exposed to UPE while of primary school age (control) (Table 1). At survey in 2011, the sample ranged in age from 24 to 30. In Ethiopia, 1,424 women were exposed to UPE while of primary school age (treatment), and 1,231 women were not exposed to UPE while of primary school age (control) (Table 1).4 At survey in 2011, the sample ranged in age from 27 to 33.

Regression Discontinuity Design

In this study, I used a fuzzy regression discontinuity design (RDD) to assess the effect of schooling on desired fertility. The regression discontinuity empirical strategy offers a number of advantages over more conventional designs: for example, associations between years of schooling and desired fertility estimated using ordinary least squares (OLS) regression (Table 6 in the appendix). If there were unobserved characteristics—such as family wealth, cognitive ability, or family preferences—that predicted both schooling and desired fertility, then OLS estimates would misstate the schooling impact because schooling was partially proxying for these unobserved factors. For example, there might have been unobserved social norms that discouraged female schooling and encouraged high fertility, which would lead to bias toward 0 in the OLS estimates because schooling was partially proxying for unobserved norms that positively affected desired fertility independent of school. Alternatively, some females might have aspired to more schooling attainment and fewer children independent of school attendance. This would lead to bias away from 0 in OLS estimates because schooling was partially proxying for unobserved aspirations that negatively affected desired fertility independent of school.

The idea behind the RDD is that UPE implementation could be treated as a random event that allowed girls just below primary school exit age (age 11–13) at policy implementation to extend their schooling with zero fees, but not girls just above primary school exit age (age 14–16). In the classic RDD, exposure to UPE would be a deterministic and discontinuous function of birth cohort (Angrist and Pischke 2009). The RDD would be estimated by regressing Y, the outcome, on B, birth cohort, and D, treatment status, where Y is desired fertility at survey, and D is an indicator of exposure to UPE while aged 13 or younger (Eq. (1)). The RDD would control for the endogeneity of schooling because girls just above primary school exit age and just below primary school exit age at the year of policy implementation should be comparable on both observed and unobserved characteristics and differ only in their exposure to the UPE policy, which was randomly determined by age at policy implementation.
Yi=α0+α1Bi+α2Di...αkXk+vi.
(1)

However, in East Africa in the mid-1990s, grade repetition, long-term absenteeism, and late entry into school were all common. Thus, some girls who were beyond primary school age would have been exposed to UPE because they were still in primary school when UPE was introduced. Likewise, some girls who were primary school age would not have attended primary school even after the elimination of school fees because they had already entered the labor market or marriage market or because of indirect costs associated with schooling. This noncompliance with treatment assignment means that this was a “fuzzy”—as opposed to a clean—RDD and required a slightly different specification (Angrist and Pischke 2009).

The fuzzy RDD exploits discontinuities in the probability of exposure to UPE conditional on birth cohort. In a fuzzy RDD, the discontinuity became an instrumental variable for treatment status rather than deterministically indicating treatment or control status. Effects were estimated using a two-stage least squares (2SLS) estimation strategy. In the first stage, I regressed D, the treatment, on Z, the identifying instrument, where D is a continuous measure of number of years of school completed, and Z is a dichotomous indicator of exposure to UPE while age 13 or younger (Eq. (2)).
Di=α0+α1Zi+...αkXk+vi.
(2)
In the second stage, I regressed Y, the outcome, on the predicted value of D from the first stage, where Y is a continuous measure of desired fertility at time of DHS survey (Eq. (3)).
Yi=β0+β1Di+...βkXk+εi.
(3)

Controlling for relevant respondent pretreatment background characteristics increases the precision of estimates; thus, all models included controls for ethnolinguistic background, religion, and number of siblings. I controlled for ethnolinguistic background and religion because certain ethnic or religious groups may have historically prioritized schooling differently for political, geographic, or socioeconomic reasons. I controlled for number of siblings because families with more children might have benefited more from fee removal.

Ethnolinguistic background was represented by indicator variables for native language (Table 7 in the appendix).5 In Malawi, this included Chichewa, Tumbuka, Yao, and other. In Uganda, this included Ateso-Karamojong, Luganda, Luo, Runyankole-Rukiga, and other. In Ethiopia, this included Amaringa, Oromigna, Tigrigna, and other. Religion was represented by indicator variables for (1) Christian (non-Catholic), (2) Catholic, (3) Muslim, (4) Orthodox, and (5) other religion. I controlled for respondent sibling size with indicator variables for (1) 0–2 siblings, (2) 3–5 siblings, (3) 6–8 siblings, and (4) more than 8 siblings.

Family wealth was potentially another important background variable because poorer families benefited more from removal of fees than wealthier families (Deininger 2003; World Bank 2009). However, I was unable to control for family wealth at time of UPE implementation in the mid-1990s because the DHS did not collect retrospective information of this type. I did not control for current household wealth at survey or any other post-treatment variables because they might be endogenous to the outcomes I analyze. For example, women with more schooling might have more current wealth because of improved labor market or marriage market outcomes; thus, it would be impossible to know whether schooling affected current wealth or whether current wealth affected schooling.

The RDD controls for unobserved characteristics, conditional on the assumption that Zi is uncorrelated with both εi and vi. In other words, I must assume that there is no additional pathway through which exposure to UPE could have affected desired fertility. If this were not the case, then it would be impossible to know whether schooling, as opposed to this alternative pathway, affected desired fertility. The validity of the RDD approach is predicated on the assumption of ignorability of the instrument: in other words, that the instrument—exposure to UPE—is random with regard to the right-side variables. The plausibility of this assumption depends on the interval of birth cohorts considered (e.g., ages 11–13 and 14–16). Correctly specifying that this interval is adequately narrow is important to ensure that treatment and control groups are comparable and differ only in their extent of exposure to UPE. At the same time, the interval must also be large enough to allow for adequate sample sizes.

A final methodological concern is the endogeneity of the implementation of UPE at the national level. Perhaps the sociohistorical conditions that led to UPE implementation also led to increases in schooling independently of UPE. In Malawi and Ethiopia, fee removal corresponded with the end of longstanding dictatorships and the introduction of multiparty democracy; in Uganda, fee elimination corresponded with the first direct presidential elections (Grogan 2008; World Bank 2009). Fee elimination was a central campaign promise of the winning presidential candidates in all three countries, thus indicating public support for increased investment in education. Following the 1990 Jomtien Conference for Education for All, elimination of school user fees also became a priority of donors—including the World Bank—who provided support for implementation of UPE in East Africa and elsewhere in the world (UNESCO 1990; World Bank 2009). Nonetheless, the large enrollment increases directly following UPE illustrate that fees were a major obstacle to enrollment for poor populations and that removal of fees probably led to increased school enrollment independent of concurrent social or political changes (Al-Samrrai and Zaman 2007; Deininger 2003; World Bank 2009).

Desired Fertility Outcomes

I analyzed one measure of desired fertility using the DHS question, “If you could go back to the time when you did not have any children and could choose exactly the number of children to have in your whole life, how many would that be?” However, differences in ideal family size between treatment and control groups could potentially be confounded by cohort trends in fertility and ex-post facto rationalization of current children (Table 1).6 Thus, I also ran analyses with a second measure: an indicator of whether the woman reported very high desired fertility, which was defined as an ideal family size of six or more children. The advantage of this second measure was that the vast majority of women (more than 95 %) in all three countries had no more than five surviving children at DHS survey (Table 1); thus, surviving children at survey would not have influenced desire for six or more children.

Pathways Outcomes

As a secondary analysis, I operationalized the four pathways through which schooling might have affected knowledge and attitudes that led to changes in desired fertility. I used the same RDD analytic strategy and substituted the pathway outcomes for desired fertility. This analysis provides important insight into the pathways through which schooling could have affected desired fertility across contexts; however, it had some limitations. First, this was not an exhaustive analysis as there were undoubtedly pathways or variables that I was unable to explore because of the data constraints of the DHS. Second, it was not possible to isolate pathway effects on knowledge and attitudes from each other. For example, more than one pathway might have affected knowledge about family planning. Finally, although this pathways analysis could establish a link between schooling and potential mediators, it could not explicitly link the mediators to desired fertility, the final outcome.

Pathway 1 explores the direct effect of school on knowledge about family planning and reproductive health, including: (1) a dichotomous indicator of whether the respondent correctly answered a question demonstrating understanding of the ovulatory cycle; (2) an index ranging from 0 to 5 about knowledge of HIV transmission, where 0 indicates poor knowledge, and 5 indicates good knowledge; and (3) a dichotomous indicator of whether the respondent knew where to obtain condoms.

Pathway 2 evaluates Caldwell’s hypothesis that schooling socialized children into different gender and family norms, including: (1) an index ranging from 0 to 4 of the number of household decisions in which the respondent participated (either alone or jointly with her partner); (2) a dichotomous indicator of whether the respondent participated in decisions about how to spend her earnings (either alone or jointly with her partner); and (3) a dichotomous indicator of whether the respondent believed that a husband is never justified in beating his wife.

Pathway 3 explores the effect of school on engagement with mass media, including: (1) a dichotomous indicator of whether the respondent reads a newspaper; (2) a dichotomous indicator of whether the respondent watches television; and (3) a dichotomous indicator of whether the respondent listens to radio.7

Pathway 4 evaluates the effect of school on partner characteristics, including: (1) the total years of schooling completed by the respondent’s current partner; (2) a dichotomous indicator of whether the respondent’s current partner was considerably older (10 years or greater) than the respondent; and (3) a dichotomous indicator of whether the respondent’s current partner completed more years of schooling than the respondent. The DHS collected no information on social networks; thus, I was unable to assess the role of school on peers and other social networks.

First-Stage Results

In the first stage, I regressed total years of schooling on exposure to UPE. In Malawi, girls exposed to UPE at age 13 had an average of 0.46 more years of schooling than girls not exposed to UPE at age 13 (p < .001) (Table 2, column 1). In Uganda, girls exposed to UPE at age 13 had an average of 1.06 more years of schooling than girls not exposed to UPE at age 13 (p < .001) (Table 2, column 2). In Ethiopia, girls exposed to UPE at age 13 had an average of 0.61 more years of schooling than girls not exposed to UPE at age 13 (p < .001) (Table 2, column 3). Gains in schooling ranging from one-half to a full year may seem small. However, overall levels of schooling were quite low for all three countries: the average years of school for the control (no exposure) group was 5.3 in Malawi, 5.6 in Uganda, and 2.2 in Ethiopia (Table 1). Thus, in these countries the aforementioned increases in schooling were sizable.

In order to use UPE exposure as an instrument for years of schooling, UPE must have actually increased primary school attendance: in other words, cov(Zi, Di) > 0. For a single instrument and single endogenous regressor, the instrument could be considered relevant if the t value for the instrument was larger than 3.2, or the corresponding p value was below .0016, or the F for the excluded instrument was greater than 10 (Stock et al. 2002). This was the case for all three countries. A plot of average years of schooling by birth cohort shows an upward trend in schooling over time in each country but also a discontinuous jump in average years of schooling for birth cohorts directly affected by UPE (Fig. 1). Figure 1 also shows country-level variation in the extent of fuzziness in the regression discontinuity. Uganda had the sharpest discontinuity, and Ethiopia had the fuzziest, with Malawi somewhere in between. This could be a result of the different implementation strategies. Uganda adopted a “big bang” approach from the onset, whereas Ethiopia adopted a gradual approach, and Malawi adopted a gradual approach that was replaced by a “big bang” approach. In spite of these differences in fuzziness, the strength of the instrument is strong (p < .001) for each of the three countries.

Second-Stage Results of the Effect of School on Ideal Family Size and Very High Desired Fertility

In the second stage, I regress ideal family size on the predictions from the first stage. I found that a one-year increase in schooling for a woman in Malawi led to a 0.34 reduction in ideal family size, compared with if she had not attended school an extra year (p < .001 level) (Table 3, column 1). In Uganda, a one-year increase in a woman’s schooling led to a 0.11 reduction in ideal family size, compared with if she had not attended school an extra year (p < .05) (Table 3, column 2). In Ethiopia, a one-year increase in a woman’s schooling led to a 0.34 reduction in ideal family size, compared with if she had not attended school an extra year (p < .01) (Table 3, column 3).

To account for the possibility that responses to ideal family size were biased by cohort trends in fertility and ex-post facto rationalization of current children, I reran all analyses with an outcome variable indicating very high desired fertility (six or more children), which is a measure not subject to these concerns because more than 95 % of the sample had five or fewer surviving children. In Malawi, a one-year increase in a woman’s schooling led to a .09 percentage point reduction in the probability of very high desired fertility compared with if she had not attended school an extra year (p < .01 level) (Table 3, column 4). In Uganda, a one-year increase in a woman’s schooling led to a .05 percentage point reduction in the probability of very high desired fertility compared with if she had not attended school an extra year (p < .01) (Table 3, column 5). In Ethiopia, a one-year increase in a woman’s schooling led to a .11 percentage point reduction in the probability of very high desired fertility compared with if she had not attended school an extra year (p < .01) (Table 3, column 6).

The largest limitation of the analysis was the possibility of a secular downward trend in desired fertility over time and that the discontinuity was just representing this trend. As a robustness check, I ran an alternative specification of the model that controlled for time (birth cohort) in the first stage; results were substantively the same (see Table 5 in the appendix).8 The robustness of results to this alternative model specification suggests that results were not driven by a secular time trend. Further insight was gained from the Hansen J test statistic, which tested the null hypothesis that the identifying instruments were uncorrelated with the error term (Hansen 1982). A rejection of the null hypothesis of the Hansen J test statistic would cast doubt on the validity of the instruments. In the alternative specification of the model controlling for time (birth cohort) in the first stage, the Hansen J test statistic failed to reject the null hypothesis in all three countries. If my results were driven by a secular time trend, I would expect the Hansen J test statistic to reject the null hypothesis because the error term in the second stage would have been correlated with the time trend in the first stage. The fact that this was not the case provides further support that results were not being driven by a secular time trend in the second stage.

Second-Stage Results From Analysis of Pathways Through Which Schooling Affects Desired Fertility

As a secondary exploratory analysis, I empirically investigated some of the pathways through which schooling could have affected desired fertility. First, I evaluated the effect of school on knowledge about family planning and reproductive health (Pathway 1). In Ethiopia, increased schooling had a significant positive effect on knowledge about HIV transmission and on the probability of knowing where to access a condom (Table 4, columns 8–9). In Malawi and Uganda, schooling had no significant effect on knowledge about ovulation, knowledge about HIV transmission, or knowledge about where to access a condom (Table 4, columns 1–6).

Next, I explored the effect of schooling on gender and family norms (Pathway 2). In Malawi, schooling had a significant positive effect on women’s participation in household decision-making (Table 4, column 1). In Uganda and Ethiopia, schooling had no effect on women’s participation in household decision-making, participation in decisions about women’s earnings, or opinions about the appropriateness of domestic violence (Table 4, columns 4–9).

Then, I looked at the effect of schooling on engagement with mass media (Pathway 3). In Uganda and Ethiopia, increased schooling positively affected women’s probability of reading a newspaper (Table 4, columns 4 and 7). In Ethiopia, schooling also positively affected women’s probability of watching television and listening to the radio (Table 4, columns 8–9). In Malawi, schooling had no effect on women’s probability of reading a newspaper, watching television, or listening to the radio (Table 4, columns 1–3).

Finally, I explored the effect of schooling on partner characteristics (Pathway 4). In all three countries, increased women’s schooling positively affected the years of schooling of the women’s current partner (Table 4, columns 1, 4, and 7). In Ethiopia, increased schooling also negatively affected the probability of a respondent having a considerably older (10 or more years) current partner (Table 4, column 8). In Uganda, increased schooling significantly negatively affected the probability that the respondent’s current partner had completed more years of school than the respondent (Table 4, column 6).

Discussion

In this article, I used a fuzzy regression discontinuity approach to assess the effect of schooling on women’s desired fertility across three East African countries. I found that a one-year increase in schooling decreased a woman’s ideal family size by 0.34 in Malawi, 0.11 in Uganda, and 0.34 in Ethiopia. Although these effects may appear small, long-term cumulative effects could be sizable when the increases in schooling that occurs over the course of a generation are taken into account. For example, in Malawi, the female survival rate to the end of primary school was 17 % in 1994, the year when UPE was implemented, and more than 50 % in 2012 (UIS 2014). In Ethiopia, the female survival rate to the end of primary school was 28 % in 1994, the year when UPE was implemented, and 40 % in 2011 (UIS 2014). Furthermore, the preceding estimates represent population averages that do not capture the fact that effects were likely larger among women of low SES, who historically had low schooling and high fertility.

A naïve OLS model also indicated a significant negative association between years of schooling and ideal family size in Malawi (–.10), Uganda (–.10), and Ethiopia (–.07) (p < .001) (Table 6 in the appendix). Nonetheless, the point estimates in the OLS model were smaller than those generated by the RDD in Malawi and Ethiopia, perhaps because of unobserved factors that were positively associated with desired fertility and negatively associated with schooling. For example, there might have been unobserved social norms that discouraged female schooling and encouraged high fertility, leading to bias toward 0 in OLS estimates because schooling was partially proxying for unobserved norms that positively affected desired fertility independent of schooling. This possibility highlights the need for future research on fertility behaviors and preferences that better addresses the endogeneity of schooling.

The analysis also indicated that increased schooling had a significant negative effect on women’s probability of very high desired fertility (six or more children), a measure not subject to concerns about cohort trends in fertility and ex-post facto rationalization of current children because more than 95 % of the sample had five or fewer surviving children. Nonetheless, it is worth noting that ideal family size remained high—around four children—among women exposed to UPE across all three countries in this study (Table 1).

A number of country-level factors could have contributed to the differential impacts of schooling on desired fertility across the three countries. Of the three countries, Uganda had the highest average years of schooling prior to policy implementation (5.6 years) (Table 1). Malawi had marginally lower average years of schooling (5.3 years), and Ethiopia had considerably lower average years of schooling (2.4 years) (Table 1). Furthermore, policy implementation took different forms in each of the three countries. Uganda adopted a “big bang” approach to UPE implementation, whereas Malawi used a gradual approach that was replaced by a “big bang” approach, and Ethiopia relied upon a gradual approach because of lags in implementation (World Bank 2009). These differences in implementation could have implications for school quality—including student-to-teacher ratios and availability of resources—that also influenced knowledge and attitudes related to fertility. The Ethiopian program also eliminated fees for two years of secondary school, whereas the Malawian and Ugandan policies eliminated primary school fees only. The returns to schooling may have been higher in Ethiopia because girls were allowed to access higher levels of schooling without fees.

The negative effect of school on desired fertility across the three countries supports the theoretical views of Caldwell (1980) and empirical evidence provided by Lloyd et al. (2000), Bongaarts (2010), and others about the important role that mass schooling plays in changing fertility preferences across contexts. My exploratory analysis of the pathways through which schooling affected desired fertility suggests some commonalities in pathways across contexts. For example, women’s schooling positively affected partner’s schooling in all three countries. However, other pathways were country-specific. Malawi was the only country where schooling affected women’s participation in household decision-making, and Ethiopia was the only country where schooling affected women’s knowledge about reproductive health. In this analysis, I was unable to assess whether certain pathways were more important than others or whether more important pathways tended to be prevalent across multiple contexts or instead were context-specific. Furthermore, I could not assess some variables—for example, the effect of school on peers and social networks—at all because of the constraints of the DHS. Future research should examine these issues in more detail using in-depth country-specific data.

The question remains as to what changes in desired fertility mean for changing African fertility. On one hand, a number of factors—including lack of women’s economic opportunities and high unmet need for family planning—may continue to contribute to slow rates of fertility decline. Nonetheless, the fertility transition is underway in Africa, and the results of this analysis suggest that the continued expansion in school enrollment over the last two decades could have important implications for changing fertility—in part, through changing desired fertility—in Malawi, Uganda, and Ethiopia as well as in other African countries that have expanded enrollment through UPE or other mass schooling policies.

Acknowledgments

Background support for this study was provided by the grant Team 1000+ Saving Brains: Economic Impact of Poverty-Related Risk Factors for Cognitive Development and Human Capital “0072-03” provided to the Grantee, The Trustees of the University of Pennsylvania by Grand Challenges Canada. I am grateful to Jere Behrman, Lawrence Wu, Delia Baldassarri, Jennifer Jennings, Amber Peterman, Florencia Torche, Paula England, Jennifer Hill, and three anonymous reviewers for helpful comments on earlier versions of this article.

Appendix

Notes

1

Primary schooling had a significant negative effect on pregnancy and marriage in Kenya (Duflo et al. 2012; Dupas 2011; Ferre 2009) and Nigeria (Osili and Long 2008), and a negative effect on total fertility in Malawi (Zanin et al. 2015). Secondary schooling had a negative effect on pregnancy and marriage in Malawi (Baird et al. 2010) and Kenya (Ozier 2010), and a negative effect on sexual debut in Uganda (Alsan and Cutler 2013). The literature has focused primarily on adolescent fertility outcomes rather than total fertility, likely because of the long time span needed to observe total fertility. Nonetheless, it is well documented that delays to fertility almost universally result in lower total fertility (Bongaarts 2002).

2

The primary school gross enrollment rate is defined as total enrollment in primary school divided by the primary age school population. This figure can exceed 100 % if children over primary school age are still in primary school.

3

I was unable to empirically investigate whether school affected fertility preferences of men because the association between exposure to UPE and years of schooling (the first stage) is not statistically significant for males in Malawi and Uganda. This is likely due to the greater benefit to girls than to boys from the removal of fees (World Bank 2009).

4

Dates in the Ethiopian data are converted to the Gregorian calendar to ensure comparability with other countries.

5

Major ethnolinguistic groups were those accounting for at least 10 % of the sample. I focused on ethnolinguistic background rather than ethnic group because of the large number of ethnic groups in these countries. Ethiopia and Uganda had approximately 80 and 40 ethnic groups listed in the DHS, respectively; thus, controlling for individual ethnic groups was problematic because of the small number of respondents in each group. Individual ethnic groups can be classified into broader ethnolinguistic groups sharing common linguistic and sociohistorical roots. Ethnolinguistic background captured ethnic diversity at a higher level of aggregation more suitable to this analysis.

6

None of the women in the sample gave nonnumeric responses to ideal family size. This was consistent with Bachan and Frye’s (2013) finding that nonnumeric responses to ideal family size have decreased over time in Africa.

7

I reran analyses using continuous variables that measured the frequency of watching television, reading a newspaper, and listening to radio. Results were substantively unchanged.

8

In my final model, I did not include the control for time (birth cohort) in the first stage because this caused the birth cohort variable and exposure to UPE variable to be imprecise in the first stage in two of the three countries. This was not surprising given that the exposure to UPE variable was constructed using birth cohort (the simple correlation between exposure to UPE and birth cohort was .83 in Malawi, .82 in Uganda, and .84 in Ethiopia). Nonetheless, in the alternative model specification that controlled for time (birth cohort) in the first stage, exposure to UPE and birth cohort were jointly highly significant in the first stage in all three countries (p < .001), and the second-stage estimates were substantively unchanged (Table 5 in the appendix).

References

Ainsworth, M., Beegle, K., & Nyamette, A. (
1996
).
The impact of women’s schooling on fertility and contraceptive use: A study of fourteen sub-Saharan African countries
.
The World Bank Economic Review
,
10
,
85
122
. 10.1093/wber/10.1.85
Al Samarrai, S., & Zaman, H. (
2007
).
Abolishing school fees in Malawi: The impact on education access and equity
.
Education Economics
,
15
,
359
375
. 10.1080/09645290701273632
Alsan, M. M., & Cutler, D. M. (
2013
).
Girls’ education and HIV risk: Evidence from Uganda
.
Journal of Health Economics
,
32
,
863
872
. 10.1016/j.jhealeco.2013.06.002
Angrist, J. D., & Pischke, J-S (
2009
).
Mostly harmless econometrics: An empiricist’s companion
.
Princeton, NJ
:
Princeton University Press
.
Avenstrup, R., Liang, X., & Nellemann, S. (
2004
).
Kenya, Lesotho, Malawi and Uganda: Universal primary education and poverty reduction
(Research report).
Washington, DC
:
The World Bank
.
Bachan, L., & Frye, M. (
2013
).
The decline in non-numeric ideal family size: A cross-regional analysis
. Unpublished manuscript,
The Pennsylvania State University
,
State College, PA
and
University of California at Berkeley
,
Berkeley, CA
.
Baird, S., Chirwa, E., McIntosh, C., & Ozler, B. (
2010
).
The short-term impacts of a schooling conditional cash transfer program on the sexual behavior of young women
.
Health Economics
,
19
,
55
68
. 10.1002/hec.1569
Baird, S., Garfein, R. S., McIntosh, C., & Ozler, B. (
2012
).
Effect of a cash transfer programme for schooling on prevalence of HIV and herpes simplex type 2 in Malawi: A cluster randomised trial
.
Lancet
,
379
,
1320
1329
. 10.1016/S0140-6736(11)61709-1
Bankole, A., Ahmed, F. H., Neema, S., Ouedraogo, C., & Konyani, S. (
2007
).
Knowledge of correct condom use and consistency of use among adolescents in four countries in sub-Saharan Africa
.
African Journal of Reproductive Health
,
11
,
197
220
. 10.2307/25549740
Bankole, A., & Singh, S. (
1998
).
Couples’ fertility and contraceptive decision-making in developing countries: Hearing the man’s voice
.
International Family Planning Perspectives
,
24
,
15
24
. 10.2307/2991915
Becker, G. S. (
1981
).
Treatise on the family
.
Cambridge, MA
:
Harvard University Press
.
Becker, G. S., & Lewis, H. G. (
1973
).
On the interaction between the quantity and quality of children
.
Journal of Political Economy
,
81
,
S279
S288
. 10.1086/260166
Bentaouet Kattan, R. (
2006
).
Implementation of free basic education policy
(Education Working Paper Series No. 7).
Washington, DC
:
The World Bank
.
Bentaouet Kattan, R., & Burnett, N. (
2004
).
User fees in primary education
(Education for All Working Paper).
Washington, DC
:
The World Bank
.
Bongaarts, J. (
1994
). The impact of population
policies
:
Comment
.
Population and Development Review
,
20
,
616
620
.
Bongaarts, J. (
2002
).
The end of the fertility transition in the developed world
.
Population and Development Review
,
28
,
419
443
. 10.1111/j.1728-4457.2002.00419.x
Bongaarts, J. (
2010
).
The causes of educational differences in fertility in sub-Saharan Africa
(Poverty, Gender, and Youth Working Paper No. 20).
New York, NY
:
The Population Council
.
Bongaarts, J., & Watkins, S. C. (
1996
).
Social interactions and contemporary fertility transitions
.
Population and Development Review
,
22
,
639
682
. 10.2307/2137804
Bulatao, R. A. (
1981
).
Values and disvalues of children in successive childbearing decisions
.
Demography
,
18
,
1
25
. 10.2307/2061046
Caldwell, J. C. (
1976
).
Toward a restatement of the demographic transition theory
.
Population and Development Review
,
2
,
321
366
. 10.2307/1971615
Caldwell, J. C. (
1980
).
Mass education as a determinant of the timing of fertility decline
.
Population and Development Review
,
6
,
225
255
. 10.2307/1972729
Castro Martin, T. (
1995
).
Women’s education and fertility: Results from 26 Demographic and Health Surveys
.
Studies in Family Planning
,
26
,
187
202
. 10.2307/2137845
Cleland, J., & Wilson, C. (
1987
).
Demand theories of the fertility transition: An iconoclastic view
.
Population Studies
,
41
,
5
30
. 10.1080/0032472031000142516
Coale, A. J. (
1973
).
The demographic transition reconsidered
. In
Proceedings of the International Population Conference
(Vol.
1
, pp.
52
72
).
Liege, Belgium
:
IUSSP
.
Cochrane, S. H. (
1979
).
Fertility and education: What do we really know?
(World Bank Staff Occasional Papers No. 26).
Baltimore, MD and London, UK
:
The Johns Hopkins University Press for the World Bank
.
Davies, M., & Macdowall, W. (
2006
).
Health promotion theory
.
Berkshire, UK
:
Open University Press
.
Deininger, K. (
2003
).
Does cost of schooling affect enrollment by the poor? Universal primary education in Uganda
.
Economics of Education Review
,
22
,
291
305
. 10.1016/S0272-7757(02)00053-5
Dewi, R. K., Suryadarma, D., & Suryahadi, A. (
2013
).
The impact of expansion of television coverage on fertility: Evidence from Indonesia
(SMERU Working paper).
Jakarta, Indonesia
:
SMERU Research Institute
.
Duflo, E., Dupas, P., & Kremer, M. (
2012
).
Education, HIV, and early fertility: Experimental evidence from Kenya
. Unpublished manuscript,
Economics Department, Massachusetts Institute of Technology
,
Cambridge, MA
,
Economics Department, Stanford University
,
Stanford, CA
, and
Department of Economics
,
Harvard University Cambridge, MA
.
Dupas, P. (
2011
).
Do teenagers respond to HIV risk information? Evidence from a field experiment in Kenya
.
American Economic Journal: Applied Economics
,
3
(
1
),
1
36
.
Easterlin, R. A. (
1975
).
An economic framework for fertility analysis
.
Studies in Family Planning
,
6
(
3
),
54
63
. 10.2307/1964934
Easterlin, R. A., & Crimmins, E. (
1985
).
The fertility revolution
.
Chicago, IL
:
University of Chicago Press
.
Ferré, C. (
2009
).
Age at first child: Does education delay fertility timing? The Case of Kenya
(Policy Research Working Paper No. 4833).
Washington, DC
:
The World Bank
.
Gallant, M., & Maticka-Tyndale, E. (
2004
).
School-based HIV prevention programmes for African youth
.
Social Science & Medicine
,
58
,
1337
1351
. 10.1016/S0277-9536(03)00331-9
Grogan, L. (
2008
).
Universal primary education and school entry in Uganda
.
Journal of African Economies
,
18
,
183
211
. 10.1093/jae/ejn015
Hansen, L. P. (
1982
).
Large sample properties of generalized method of moments estimators
.
Econometrica
,
50
,
1029
1054
. 10.2307/1912775
Hayford, S. R. (
2009
).
The evolution of fertility expectations over the life course
.
Demography
,
46
,
765
783
. 10.1353/dem.0.0073
Hayford, S. R., & Agadjanian, V. (
2011
).
Uncertain future, non-numeric preferences and the fertility transition: A case study of rural Mozambique
.
African Population Studies
,
25
,
419
439
. 10.11564/25-2-239
Heiland, F., Prskawetz, A., & Sanderson, W. C. (
2008
).
Are individuals’ desired family sizes stable? Evidence from West German panel data
.
European Journal of Population
,
24
,
129
156
. 10.1007/s10680-008-9162-x
Hirschman, C. (
1994
).
Why fertility changes
.
Annual Review of Sociology
,
20
,
203
233
. 10.1146/annurev.so.20.080194.001223
Jensen, R., & Oster, E. (
2009
).
The power of TV: Cable television and women’s status in India
.
Quarterly Journal of Economics
,
124
,
1057
1094
. 10.1162/qjec.2009.124.3.1057
Johnson-Hanks, J. (
2007
).
Natural intentions: Fertility decline in the African Demographic and Health Surveys
.
American Journal of Sociology
,
112
,
1008
1043
. 10.1086/508791
Kadzamira, E., & Rose, P. (
2003
).
Can free primary education meet the needs of the poor? Evidence from Malawi
.
International Journal of Educational Development
,
23
,
501
516
. 10.1016/S0738-0593(03)00026-9
Knodel, J., & van de Walle, E. (
1986
).
Lessons from the past: Policy implications of historical fertility studies
. In Coale, A. J., & Watkins, S. C. (Eds.),
The decline of fertility in Europe: The revised proceedings of a conference on the Princeton European Fertility Project
(pp.
390
419
).
Princeton, NJ
:
Princeton University Press
.
Kravdal, Ø (
2002
).
Education and fertility in sub-Saharan Africa: Individual and community effects
.
Demography
,
39
,
233
250
.
La Ferrara, E., Chong, A., & Duryea, S. (
2012
).
Soap operas and fertility: Evidence from Brazil
.
American Economic Journal: Applied Economics
,
4
,
1
31
.
Lee, R. D. (
1980
).
Aiming at a moving target: Period fertility and changing reproductive goals
.
Population Studies
,
34
,
205
226
. 10.1080/00324728.1980.10410385
Liefbroer, A. C. (
2009
).
Changes in family size intentions across young adulthood: A life-course perspective
.
European Journal of Population
,
25
,
363
386
. 10.1007/s10680-008-9173-7
Lloyd, C. B., Kauffman, C. E., & Hewett, P. (
2000
).
The spread of primary schooling in sub-Saharan Africa: Implications for fertility change
.
Population and Development Review
,
26
,
483
515
. 10.1111/j.1728-4457.2000.00483.x
Mare, R. D. (
1991
).
Five decades of educational assortative mating
.
American Sociological Review
,
56
,
15
32
. 10.2307/2095670
Mare, R. D., & Maralani, V. (
2006
).
The intergenerational effects of changes in women's educational attainments
.
American Sociological Review
,
71
,
542
564
. 10.1177/000312240607100402
Mason, K. O. (
1997
).
Explaining fertility transitions
.
Demography
,
34
,
443
454
. 10.2307/3038299
McCarthy, J., & Oni, G. A. (
1987
).
Desired family size and its determinants among urban Nigerian women: A two-stage analysis
.
Demography
,
24
,
279
290
. 10.2307/2061635
McQueston, K., Silverman, R., & Glassman, A. (
2013
).
The efficacy of interventions to reduce adolescent childbearing in low and middle income countries: A systematic review
.
Studies in Family Planning
,
44
,
369
387
.
Method, F., Ayele, T., Bonner, C., Horn, N., Meshesha, A., & Talore Abiche, T. (
2010
).
Impact assessment of USAID’s education program in Ethiopia 1994–2009
(USAID Research Report).
Washington, DC
:
United States Agency for International Development
.
Moultrie, T. A., & Timaeus, I. M. (
2014
,
May
).
Rethinking African fertility: The state in, and of, the future sub-Saharan African fertility decline
. Paper presented at the annual meeting of the Population Association of America, Boston, MA.
Mouw, T. (
2006
).
Estimating the causal effect of social capital: A review of the recent research
.
Annual Review of Sociology
,
32
,
79
102
. 10.1146/annurev.soc.32.061604.123150
Olusanya, P. O. (
1971
).
Status differentials in the fertility attitudes of married women in two communities in Western Nigeria
.
Economic Development and Cultural Change
,
19
,
641
651
. 10.1086/450517
Osili, U. O., & Long, B. T. (
2008
).
Does female schooling reduce fertility? Evidence from Nigeria
.
Journal of Development Economics
,
87
,
57
75
. 10.1016/j.jdeveco.2007.10.003
Ozier, O. (
2010
).
The impact of secondary schooling in Kenya: A regression discontinuity analysis
. Unpublished manuscript,
Sloan School of Management, Massachusetts Institute of Technology
,
Cambridge, MA
.
Pollak, R. A., & Watkins, S. C. (
1993
).
Cultural and economic approaches to fertility: Proper marriage or mesalliance?
.
Population and Development Review
,
19
,
467
496
. 10.2307/2938463
Pritchett, L. (
1994
).
Desired fertility and the impact of population policies
.
Population and Development Review
,
20
,
1
55
. 10.2307/2137629
Pritchett, L. (
2013
).
The rebirth of education: Schooling ain’t learning
.
Washington, DC
:
Center for Global Development
.
Reed, H., Briere, R., & Casterline, J. (Eds.). (
1999
).
The role of diffusion processes in fertility change in developing countries
.
Washington, DC
:
National Academies Press
.
Riley, A. P., Hermalin, A. I., & Rosero-Bixby, L. (
1993
).
A new look at the determinants of nonnumeric response to desired family size: The case of Costa Rica
.
Demography
,
30
,
159
174
. 10.2307/2061835
Ryder, N. B. (
1973
).
A critique of the National Fertility Study
.
Demography
,
10
,
495
506
. 10.2307/2060877
Ryder, N. B., & Westoff, C. F. (
1967
).
The trend of expected parity in the United States: 1955, 1960, 1965
.
Population Index
,
33
,
153
168
. 10.2307/2733064
Stock, J. H., Wright, J. H., & Yogo, M. (
2002
).
A survey of weak instruments and weak identification in generalized method of moments
.
Journal of Business and Economic Statistics
,
20
,
518
529
. 10.1198/073500102288618658
Thomson, E. (
1997
).
Couple childbearing desires, intentions, and births
.
Demography
,
34
,
343
354
. 10.2307/3038288
Udry, J. R. (
1983
).
Do couples make fertility plans one birth at a time?
.
Demography
,
20
,
117
128
. 10.2307/2061230
UNESCO
. (
1990
).
World declaration on education for all and framework for action to meet basic learning needs
.
Paris, France
:
UNESCO
.
UNESCO Institute for Statistics (UIS)
. (
2014
).
Primary school survival statistics
[Database]. Retrieved from http://data.uis.unesco.org/
Watkins, S. C. (
1991
).
From provinces into nations: Demographic integration in Western Europe, 1870–1960
.
Princeton, NJ
:
Princeton University Press
.
World Bank
. (
2009
).
Abolishing school fees in Africa: Lessons from Ethiopia, Ghana, Kenya, Malawi, and Mozambique
(Development Practice in Education report).
Washington DC
:
The World Bank and UNICEF
.
Yeatman, S., Sennott, C., & Culpepper, S. (
2013
).
Young women’s dynamic family size preferences in the context of transitioning fertility
.
Demography
,
50
,
1715
1737
. 10.1007/s13524-013-0214-4
Zanin, L., Radice, R., & Marra, G. (
2015
).
Modelling the impact of women’s education on fertility in Malawi
.
Journal of Population Economics
,
28
,
89
111
. 10.1007/s00148-013-0502-8