Abstract

Female secondary school attendance has recently increased in sub-Saharan Africa, and so has the risk of becoming pregnant while attending school. We analyze the impact of teenage pregnancy on young women’s human capital using longitudinal data in Madagascar that capture the transition from adolescence to adulthood for a cohort aged 21–24 in 2012, first interviewed in 2004. We find that early childbearing increases the likelihood of dropping out of school and decreases the chances of completing secondary school. This pregnancy-related school dropout also has a detrimental impact on standardized test scores in math and French. We instrument early pregnancy with the young woman’s community-level access and her exposure to condoms since age 15 after controlling for pre-fertility socioeconomic conditions. Our results are robust to different specifications that address potential endogeneity of program placement and instrument validity.

Introduction

Fertility decisions can affect the welfare of women, men, and children through human capital formation. Particularly important is the evidence that delaying pregnancy among young women may contribute to higher school attainment, which might translate into better human capital outcomes for their children (Schultz 2007).1 The concern that adolescent pregnancy can have detrimental economic and social consequences is especially acute in developing countries, where early childbearing is associated not only with health risks such as maternal mortality and low birth weight2 but also with low school attainment and productivity, and consequently intergenerational transmission of poverty. A paucity of empirical evidence establishes a causal impact of early fertility on young women’s human capital formation in developing countries. This issue has come to greater prominence in sub-Saharan Africa as female secondary school enrollment has increased and so too has the risk of pregnancy-related school dropout (Lloyd and Mensch 2008). Girls in low-income countries face complex fertility and schooling decisions with the added constraints of low availability of information on safe sexual practices and limited access to reproductive health services (Chong et al. 2013).

We investigate whether early childbearing has a causal effect on young women’s school attainment and cognitive skills, measured by their math and French test scores, in Madagascar. This country offers an appropriate context for our research question: female progression to secondary school rapidly increased from 45 % to 69 % between 1998 and 2010 (World Bank 2013); however, 32 % of girls ages 15–19 have a child or are pregnant for the first time, and 48 % of women age 18 are mothers or pregnant (data from the 2009 Demographic and Health Surveys (DHS); Institut National de la Statistique (INSTAT); and ICF Macro 2010). Indeed, Madagascar is among the top 10 developing countries with teenage pregnancy rates above 20 % (Williamson 2013). This high teenage pregnancy rate occurs in a context of a high fertility rate, which remains at 4.8 children per woman. Moreover, the prevalence rate of modern family planning use is only 29 % among women aged 15–49, and there is no access to safe abortion (INSTAT and ICF Macro 2010).3,4

Empirical studies aiming to establish the causal effect of fertility on education are challenged by potential endogeneity. That is, education and fertility might be the result of joint decisions (Rindfuss et al. 1980); for example, adolescent girls may have strong preferences for education and labor market success and less preference for children. Furthermore, there might be reverse causality if adolescent girls who have lower academic performance are more likely to drop out of school to get pregnant, and also might view pregnancy as a path to marriage. These potential common unobservable characteristics in the error terms of the regression equations, for both early fertility and education, can bias the ordinary least squares (OLS) estimates of this causal effect.

Our analysis uses an instrumental variable (IV) approach to address this potential endogeneity. We use young women’s community-level access to family planning as a plausibly exogenous source of variation to predict early childbearing after controlling for other individual, household, and community characteristics, including pre-fertility socioeconomic conditions. We use a household panel survey in Madagascar that follows a cohort of young men and women aged 21–24 in 2012, first interviewed in 2004. We complement our surveys with 2001 and 2007 Malagasy social and economic infrastructure commune censuses.

We estimate two sets of models. First, we instrument a young woman’s early childbearing with “access to condoms,” defined as condom availability in the community where she resides. Second, using the year since condoms were available, we define the instrument as a young woman’s “exposure to condoms,” measured by the number of years she has had condom access since age 15. Using the same identification strategy, in the first stage, we estimate a Weibull hazard model for young women’s age at first birth, allowing us to show the impact on education outcomes of postponing the first pregnancy by one year.

The idea behind this identification strategy is that access (exposure) to condoms decreases young women’s fertility control costs and affects their schooling decisions through a reduction of early childbearing rather than through a direct effect. For our identification strategy to be valid, adolescent pregnancy does not need to be primarily determined by contraceptive access; we need only a setting where some nonzero proportion of teenage pregnancy is due to lacking access to contraceptives. To support this argument, we provide empirical evidence and conduct numerous robustness checks, including a placebo test in which we show that access to condoms does not have a statistically significant effect on young men’s school outcomes. Our IV for early childbearing is condoms rather than other family planning methods, such as pills or injectable contraceptives, because the latter are primarily used to space children within the family rather than to postpone the first birth in Madagascar (INSTAT and ICF Macro 2010; Raharinjatovo 2014).

The main concern with this identification strategy is the possibility of nonrandom program placement: condom programs are potentially located in communities where teen pregnancies are the highest or where the population is more inclined to use contraceptives (Molyneaux and Gertler 2000; Pitt et al. 1993; Pörtner et al. 2011). We lack detailed information on the administrative decision that resulted in whether and when our sample communities gained access to condoms. We, therefore, control in our models for an unusually complete set of social and economic infrastructure community variables, both contemporaneous and lagged over more than a decade. To further mitigate this concern, we show that 2006 community-level fertility variables, 2001 poverty and population, and preferences regarding family planning (FP) as measured by religion and ethnicity do not predict community-level access to condoms in 2012. We do not find any statistical evidence of non-random program placement in these and numerous robustness checks and placebo tests that we report on in detail in this article.

Our results show that early childbearing increases a young woman’s likelihood of dropping out of school and decreases her chances of completing lower secondary school. In addition to this effect on the extensive margin (i.e., being a school dropout), we also find an effect on the intensive margin: this early departure from school because of pregnancy has a detrimental impact on young women’s standardized scores in math and French. Consistently, we find that by delaying the first birth by a year, young women can have positive gains in school attainment and cognitive skills.

Unlike in developing countries, the socioeconomic effects of teenage pregnancy in the United States have been extensively researched. A series of empirical strategies have been used to identify causal effects and to address the systematic differences between mothers and nonmothers. These strategies include using sibling fixed effects to compare teen mothers with their childless sisters (Geronimus and Korenman 1992); quasi-experimental methods that use miscarriages as an instrument of early fertility (Ashcraft et al. 2013; Ashcraft and Lang 2006; Fletcher and Wolfe 2009; Hotz et al. 2005; Rindfuss et al. 1980); and other instruments for early fertility, such as age of menarche, abortion, and contraception rates (Klepinger et al. 1999; Ribar 1994). Propensity score matching methods have also been used to construct an appropriate counterfactual group for teenage mothers (Lee 2010; Levine and Painter 2003). Despite this extensive scholarly evidence, there is no consensus about whether teenage pregnancy has a causal effect on young women’s human capital in the United States. Indeed, Kane et al. (2013) found that the impact on schooling ranges from no effects to 2.6 fewer years among teen mothers, depending on the empirical methodology employed to address the selection into motherhood. Recently, Diaz and Fiel (2016) conciliated this debate by suggesting that each of these methodologies emphasizes different groups of women and that teenage pregnancy can have heterogeneous effects across populations.

In the context of developing countries, very few studies have rigorously established a causal effect of teenage fertility on young women’s human capital.5 Azevedo et al. (2012) used miscarriages as an instrument for the timing of pregnancy and found no adverse effects of a younger age at first birth on education in Mexico. In contrast, Arceo-Gomez and Campos-Vazquez (2014), using propensity score matching, found that teenage pregnancy decreased years of schooling. Using the same methodology, Ranchhod et al. (2011) found a reduced negative effect of teenage pregnancy on education in South Africa while controlling for pre-fertility characteristics; however, the effect remained large for African females, consistent with Ardington et al. (2015).

Our study contributes to the limited empirical evidence of the impact of early childbearing on socioeconomic outcomes in developing countries. In contrast to the aforementioned studies that have addressed this question in middle-income countries—where, for example, social attitudes to teenage pregnancy and institutions might help young women and their families cope with a teenage birth—we explore this issue for a low-income and high-fertility country in sub-Saharan Africa. Thus, we also contribute to the demographic literature that analyzes the timing of sexual initiation, school leaving, pregnancy, and marriage (Lloyd and Mensch 2008; Marteleto et al. 2008) during the transition from adolescence to adulthood in sub-Saharan Africa. Second, to our knowledge, there is no empirical evidence for the effect of teenage pregnancy on cognitive skills; we analyze the effect of teen pregnancy on both the extensive and the intensive margin of human capital formation. A few cross-sectional studies in South Africa have shown a negative association between test scores and fertility (Grant and Hallman 2008; Thomas 1999) or the initiation of sexual activity (Marteleto et al. 2008), but these studies have not established a causal effect of early pregnancy on cognitive skills. Third, using young women’s access to and exposure to condoms as an IV allows us to infer the potential economic consequences of policies that decrease young women’s costs to control fertility in low-income countries during adolescence, which represents a critical window for intervening to improve women’s economic opportunities (Bandiera et al. 2015).

Data Description

We use the 2012–2013 Madagascar Life Course Transition of Young Adults Survey, which reinterviewed a cohort of 1,749 young adults, 859 of them women, who were 21–24 years old at the time of the survey. This cohort was first surveyed in 2004 at ages 13–16. In the last round of the survey, 1,800 households were revisited in 73 communities across all regions in Madagascar.6 This panel tracked approximately 90 % of the cohort and therefore has a small attrition rate, especially given that eight years elapsed between survey rounds.7

The 2012 survey, designed to capture the transition from adolescence to adulthood, gathered detailed socioeconomic information of the cohort members and their households. The survey included detailed retrospective event histories on the cohort members’ schooling, fertility, marriage, and employment, as well as on a range of economic and life course events going back to 2004. The 2012 data include young adults’ cognitive tests measured through written math and French tests. The math tests also had an oral part so that the ability of illiterate members could be assessed; the Cronbach’s alpha test confirms the reliability of the tests to capture differences between more- and less-knowledgeable individuals. The tests were administered at the cohort members’ place of residence, even if they were not attending school at the time of the survey. To have a comparable measure of achievement, the tests were identical for all the individuals regardless of their school attainment, starting with basic questions and progressing to more complicated tasks. As in 2012, the 2004 survey contained tests for both math and French; the cognitive tests were administered to only 794 cohort members in 2004.

The 2012–2013 survey also questioned community leaders, teachers, and health personnel on the availability of social and economic infrastructure and services at the community level, including FP services, and the date of first availability of these services in the community. We complement this community-level information with the 2001 and 2007 commune censuses, which feature a wide range of information on basic public services and infrastructure in all communes in Madagascar.

Table 1 shows that 54 % of all female cohort members had given birth to at least one child by the time of the survey; we call this group ever-mothers. We refer to their female counterparts in the cohort who had not yet given birth as nonmothers. In our sample, the average age of first birth is 18, which is consistent with the 2009 DHS; nevertheless, 35 % of these women had their first child before age 18. Interestingly, 23 % of the ever-mothers reported having never been married at the time of the survey, plausibly suggesting that out-of-wedlock early pregnancy is not negligible in Madagascar.

We calculate the difference between the age of awareness of conception (measured as age of first birth minus 8 months of pregnancy) and the age of dropping out of school, and classify the young women of our sample according to the timing of their fertility and education decisions (Fig. 1). Almost 24 % of the sample, or 46 % of the young mothers, became pregnant while they were in school. In contrast, 30 % of the girls dropped out of school but had not become pregnant by the time of the survey. It is noteworthy that 27 % of the young women dropped out almost four years before their first birth (at median ages of 15 and 18.9, respectively), suggesting that among this group there is no overlap in the timing between their fertility and education decisions. This finding does not rule out the potential that women may drop out in light of their desire to have a child and that decisions are made jointly (Rindfuss et al. 1980). Finally, we observe that only 16 % of the girls were still attending school at the time of the survey and are nonmothers. A very negligible proportion of the sample (2 %) were ever-mothers and currently enrolled in school, suggesting that young women who desire to continue their education might find it difficult to do so after they have their first child.8 Given the potential endogeneity between education and fertility, we implement an IV strategy described in the next section.

Table 1 also shows substantial differences in schooling and cognitive performance between these two groups. Whereas 34 % of the nonmothers still attend school, only 3 % of the ever-mothers are enrolled. These patterns are consistent with the years of education completed among the two groups. The group of ever-mothers completed 6.2 years of schooling, compared with 9.2 years for nonmothers. Among the group of ever-mothers, only 5 % completed upper secondary, whereas this percentage is almost five times larger among the nonmothers. Also, 17.6 % of the women who have not yet had their first birth have some university education; this percentage among ever-mothers is less than 2 %. Additionally, nonmothers have better average 2012 math and French test scores, compared with the ever-mothers. This finding partly reflects the fact that the former group stays in school longer (Table 1).

Regarding the use of FP among the female cohort members, 31 % use at least one method of contraception (modern and/or traditional) at the time of the survey. Ever-mothers are more likely to be FP users than nonmothers (Table 1). Among the FP users in our survey, 37 % have primary school, 38 % have lower secondary, and 25 % have upper secondary or higher education. We find no evidence of a positive correlation between young women’s level of education and their FP use in our sample. Regarding access to FP services, defined as community-level availability, nonmothers have higher access to FP—specifically to oral contraceptives and condoms—than ever-mothers (Table 1).

Empirical Strategy

We estimate the following two-stage instrumental variable (IV) model:
First Stage:EverMotheri=μ+τAgei+δZi+γXi+φCi+vi.
1
Second Stage:Yi=a+βEverMotheri+πAgei+ρXi+θCi+ui.
2

In the first stage, we predict a young woman’s early childbearing (i.e., a dummy variable for being an ever-mother) using the community-level geographic and time variation of condom availability after controlling for household (Xi) and community (Ci) characteristics. Subsequently, we use this predicted fertility to explain her schooling in the second stage, where Yi represents four educational outcomes for a young woman i in 2012: (1) a dummy variable for current enrollment, (2) years of schooling, (3) a dummy variable for completing secondary school (i.e., having 9 or more years of education), and (4) standardized French and math test scores. To identify the effect of early childbearing on schooling outcomes in Eq. (2), our IV model is designed to exploit the variation in condom availability to isolate the variation in our endogenous variable (ever-mother) that is uncorrelated with the error term in Eq. (2) (Angrist and Pischke 2009; Stock and Watson 2007).

We estimate two sets of models each using a different instrument, Zi. The first instrument is a dummy variable for whether a young woman has access to condoms in the community where she lives. Access is defined as the community-level availability of condoms reported by community leaders.9,10 Using the specific year that condoms became available in a particular community, the second instrument is exposure to condoms, defined as the number of years a young woman has had access to condoms since age 15 in her community.11 We select this age cutoff because the median age of sexual initiation among Malagasy women is 17.4 (INSTAT and ICF Macro 2010).12 In our sample communities, the median year during which condom distribution started is 2000, and the young women’s average exposure to condoms is 4.8 years.13 We estimate two-stage least squares (2SLS) models and IV probit models for binary dependent variable outcomes when instrumenting with exposure to condoms.14

In both stages, we control for young women cohort’s age dummy variables, Agei. We also include Xi, a set of young women’s socioeconomic variables: a dummy variable for whether the parents were alive at the time of the survey,15 parents’ education, and an asset index constructed from the 2004 round of the survey. These last two variables are important determinants of cognitive skills and school dropout in Madagascar (Glick et al. 2011, 2016). Including childhood socioeconomic characteristics in the model also allows us to isolate the effect of pregnancy on education from the effect of poverty conditions when the girls were aged 13–16. We restrict the estimation to those girls who leave school at age 13 or older, given that the median age at school dropout among girls in Madagascar in 2004 was 13 (Glick et al. 2016). This restriction, which ensures that the girls in the sample are attending school at the minimum age at which they might be at risk of pregnancy, excludes 10 % of the 859 sample women.16

Finally, Ci includes 2012 community-survey variables to control for access (defined as community-level availability) to lycée (upper secondary school), district hospital, CSB2 (community health center), electricity, piped water, weekly market, and paved roads. Furthermore, we control for time-varying community covariates by including access to lycée, electricity, and CSB2 when the young women were 10 years old. From the 2001 community census, a period when our cohort members were young children, we include a remoteness index created using factor analysis and information on community distances to the main social infrastructure services and transportation. Taken together, these control variables from multiple points in time of the young women’s life course allow us to uncover some of the community-level unobservable characteristics that might be related to access to condoms.17 Table 2 includes the summary statistics of the variables used in the IV models.

Condom Access and Early Childbearing

The literature offers conflicting evidence on the relative importance of contraceptive supply and demand factors in determining contraceptive use and fertility. Bongaarts (1994) found that increasing access to family planning programs can decrease long-term fertility because of increased use of contraception. In contrast, Pritchett (1994) argued that fertility decline is largely explained by demand factors—more specifically, the rising opportunity cost of childbearing that accompanies the process of economic development. There are good reasons to expect that these factors that affect fertility may be different for adolescents and adults (Branson and Byker 2018).

Conceptually, the effect of access to contraception on age at first birth and teen fertility is ambiguous (Buckles and Hungerman 2016). Greater contraceptive availability reduces fertility control costs and thus might reduce adolescent pregnancy; however, contraception access could also encourage sexual activity, thereby perhaps even increasing teen births (Paton 2002). Although the evidence is limited regarding the effects of FP on different age groups in developing countries, some empirical studies have shown that women’s exposure to FP access during adolescence can significantly reduce their total fertility (see Angeles et al. (2005) in Peru; Angeles et al. (1998) in Tanzania; and Miller (2010) in Colombia). The empirical evidence that specifically investigates the effects of condom access on teen fertility in Africa is scarce. In a recent review, Mmari and Sabherwal (2013) indicated that most of the studies in the region have focused on condom use among youth.

Teenage pregnancy can be considered an indicator of risky sexual behavior because it is usually a result of unprotected sex rather than a planned fertility decision (Dupas 2011; Duflo et al. 2015; Friedman 2015). Under such circumstances, it is plausible that supply-side interventions, such as easier and lower-cost access to condoms, may be a more effective and acceptable form of contraception for young women than alternatives, such as the pill and injectable contraceptives, that require planning and regular use over the long-term. Indeed, condoms are the method most commonly used by adolescents given that these contraceptives are readily accessible and inexpensive in developing countries (Chandra-Mouli et al. 2014). Using the 2009 Madagascar DHS, we find that while the pill, injectable contraceptives, and condoms are widely known to women of all ages, the first two are primarily used to space children within the family rather than to postpone the first birth. In fact, 38 % of women use FP for the first time only after they have at least one child. Consistent with this finding, in Madagascar, condoms are more commonly used by single than married women (Glick et al. 2009) and are particularly popular among unmarried, urban, and young out-of-school women (Raharinjatovo 2014).

In contrast to the pill and injectable contraceptives, condoms are not perceived to have negative secondary health effects. Thus, using condoms as an IV avoids accounting for misconceptions about contraceptive use.

Given that we lack information on the specific channels of condom distribution within our sample communities, we use the 2009 DHS information regarding women’s supply sources of the last contraception method used. We find that the primary distribution channel for condoms is through stores (40 %), followed by pharmacies (20 %). Condoms are not distributed in schools. These distribution channels strengthen our argument about why condom availability in the young women’s community of residence can decrease early childbearing. As expressed in personal communications with FP stakeholders during our field work visits in Madagascar in 2012, school girls face the stigma of going to FP clinics to get injections or birth control pills in their community of residence, an observation that is consistent with evidence from other developing countries suggesting that adolescents are unwilling to visit FP clinics because they view them as unwelcoming (Chandra-Mouli et al. 2014). In contrast, these stakeholders suggest that potential stigma effects are far less important at stores and pharmacies, where an adolescent presence is less scrutinized than at FP clinics. Also, condoms are free or heavily subsidized by the government or NGOs. Indeed, in Madagascar, only 0.2 % of 15- to 49-year-old women who are non-FP users indicate price as a reason for not using modern contraception (INSTAT and ICF Macro 2010).18

Furthermore, according to the 2009 DHS in Madagascar, 90 % of young women aged 15–24 will ask their husband or partner to use condoms if they suspect that he has a sexually transmitted infection (STI). This is a strong indication that women have leverage in negotiating condom use as opposed to other sub-Saharan African countries, where women have limited negotiating power (Pinchoff et al. 2017). Consistently, Meekers et al. (2006) showed that young males who speak with their partners about STIs and FP are more likely to use condoms.

More broadly, condoms are considered a key policy instrument used to prevent STIs and unintended pregnancy among young women (Chong et al. 2013). The Malagasy government and NGOs have attempted to increase condom access and use among young people to reduce Madagascar’s prevalence of STIs, one of the highest in sub-Saharan Africa (Glick et al. 2009).19

Finally, to validate these arguments regarding condoms as a contraceptive method to postpone the first birth in our sample of young women, we show that community-level access (availability) to the pill, unlike condoms, does not have a statistically significant effect on early childbearing (see Online Resource 1, Table A.4).

Nonrandom Program Placement

The possibility of nonrandom program placement can threaten our identification strategy. FP programs might be placed where fertility is higher or where contraception is more socially acceptable (Molyneaux and Gertler 2000; Pitt et al. 1993; Pörtner et al. 2011), which might bias the results of contraception access on fertility. To mitigate this concern, we estimate a linear probability model of 2012 access to condoms on 2006 community-level fertility variables: number of births and number of women who died during and/or immediately after delivery, a proxy for adolescent pregnancy (Williamson 2013), and a set of 2012 community covariates.20 We also estimate the same model but control for the size of the population and poverty rates, rather than fertility measures, from the 2001 commune census.21 In Online Resource 1, Table A.1, we show that each of the coefficients for these community-level 2006 fertility and 2001 population and poverty variables, in their respective models, does not statistically affect the 2012 probability of having access to condoms.

We also investigate whether 2012 access to condoms is determined by the following community-level indicators of cultural and/or social norms and preferences regarding FP: (1) participation of the major ethnic groups (Merina, Betsileo, and Betsimisaraka) from the 2001 community census, and (2) 2012 participation of the main religious groups (Catholic, Protestant, and traditional). None of these variables are statistically significant in our models (see Online Resource 1, Table A.1), providing no statistical evidence that these variables affect the commune-level condom program placement among our communes. Because of the small number of commune-level observations and given our interest in testing whether these variables plausibly affect the timing and availability of condoms and thus the IV models, we show that our main IV results are robust to the inclusion of these fertility, religion, and ethnicity variables (see Online Resource 1, Tables B.4-B.6; see section A.1 for additional nonrandom placement robustness checks).

Beyond these statistical tests, we use our survey data to construct condom availability maps. We find no particular patterns in availability over geographic regions and implementation years in our sample communities. For example, we observe some communities without access to condoms in the most (center of the country) and the least (south and north) economically developed regions (Fig. A.1, Online Resource 1). Furthermore, we observe that there are no temporal patterns in the timing of arrival of these condom programs (Fig. A.2, Online Resource 1). Although the main entities that procure condoms at the national level in Madagascar (the U.S. Agency for International Development, the United Nations Population Fund, and the Global Fund) use different channels for condom distribution, they do not coordinate targeting efforts over geographic regions. Based on this information and that none of the agencies we spoke with could articulate precise criteria that would explain condom placement in our sample communities, we infer that little (if any) purposeful decision-making underlies the access and timing of condom programs among our sample communities.

The evidence provided in this section suggests that nonrandom program placement is not biasing our results, but we recognize that we did not and cannot completely rule out this possibility. Therefore, we still perform a series of robustness checks on the results to further strengthen the basis for our empirical strategy.

First-Stage Results

The identification of the IV model requires a strong correlation between access (or exposure) to condoms and the endogenous variable ever-mother. Having community-level access to condoms decreases the probability of being a mother by 18 percentage points, a finding that is statistically significant at the 1 % level (Table 3). Compared with the sample mean (46 %), this effect is meaningful, sizable, and consistent with the results of Meekers et al. (2006), who showed that access to condoms increases condom use among young women in Taomina, Madagascar.22 Similarly, having one extra year of exposure to condoms at the community level since age 15 decreases the probability of having children by 2.3 percentage points at the 1 % significance level (Table 3). These results add to the limited empirical evidence on FP access effects on teen fertility in Africa.

The F statistics for the first stage are 11.3 and 11.8 when using access and exposure to condoms, respectively (Table 3). These values are above the critical threshold of 10, suggesting that each of these instruments has a strong correlation with the endogenous variable and that there is no indication of a weak instrument problem (Staiger and Stock 1997); for further tests supporting that our instruments are not weak, see section A.2, Online Resource 1. These F statistics values for the first stage increase to 19 for access and exposure to condoms when we do not control for community characteristics (Tables A.2 and A.3, Online Resource 1).23 We also validate the relevance condition of our instruments by analyzing their reduced-form effect on the outcomes (Angrist and Pischke 2009). We find that access to condoms has sizable and statistically significant effects on most school and cognitive outcomes; the exception is for years of schooling, with a significance level below standard thresholds (Table 7). In the results shown in Table 3, we do not cluster the standard errors at the community level. In section A.3 of Online Resource 1, we explain the implications of clustering the standard errors given our sample size and show that our IV results are robust to this specification (Tables A.7 and A.8, Online Resource 1).

Finally, merging information on condom availability from our sample with data on condom use from a nationally representative household survey in 2012–2013 (see more details in section A.5, Online Resource 1), we find that community-level access to condoms statistically significantly increases condom use among sexually active women aged 15–49 by 7 percentage points compared with the sample mean of 9 % (Table A.10, Online Resource 1). Although this finding is consistent with evidence from Madagascar (Meekers et al. 2006) and from other developing countries (Meekers et al. 2005; Sweat et al. 2012), the widespread availability of condoms at the community level may also affect social norms and, in particular, women’s attitudes and risk assessments about sex. Thus, there may be indirect impacts of condom availability on fertility, which could explain the findings of the systematic review by Free et al. (2011), who concluded that there is little consistent evidence on the effectiveness of condom promotion interventions on condom use.

Hazard Models

Using the same identification strategy, we estimate in the first stage the following Weibull hazard model for age at first birth:
hjt=hotexpδAgej+βZj+αXj+ρCj.
3

The hazard rate h(t) is the probability of having the first birth at time (or age) t, conditional on not having a child until t; the other variables are the same as in Eq. (1). The term ho is the baseline hazard defined as ho(t) = ptp − 1. This Weibull model allows us to calculate an expected predicted survival time, a “predicted age of first birth” (PredAFB), which we use in the second stage to explain the school outcomes:

Yi=a+βPredAFBi+πAgei+ρXi+θCi+ui,
4
where Yi corresponds to the different school outcomes previously analyzed. We report in Table A.11, Online Resource 1, the hazard ratios for the main covariates using exposure to condoms.24 We obtain the same qualitative results as when modeling the probability of ever-mother. Figure 2 shows the predicted hazard function after estimating the Weibull model (Eq. (3)) that controls for access to condoms. Young women in communities with access to condoms have a lower risk of being pregnant, which confirms the validity of our identification strategy (see Online Resource 1, section A.6, for robustness checks for the Weibull hazard model).

Results

Table 4 reports the early childbearing effects on school attainment. The OLS results indicate that being a teen mother decreases the probability of current school enrollment by 27 percentage points. This estimate increases to 42 percentage points in the IV models and is statistically significant at the 5 % and 1 % levels, respectively, using access to condoms and exposure to condoms as the instrument. The difference between the OLS and IV estimates is statistically significant only when using exposure to condoms as an instrument.25

Adolescent motherhood decreases by two the number of years of schooling under the OLS model and between 2.1 and 2.4 years in the case of the IV specifications; however, the IV point estimates are not statistically significant (Table 4). As described earlier, the difference in years of schooling between nonmothers and ever-mothers is larger when girls are going through the secondary school cycle. In fact, early childbearing decreases the probability of completing lower secondary school by 25 percentage points under the OLS estimation, 44 percentage points when exposure to condoms is the instrument (p < .01), and 48 percentage points when the instrument is access to condoms (p < .05). We plot in Fig. 3 the impact of early childbearing during the progression through secondary school (i.e., from having at least five years to having at least 13 years of schooling) as estimated in the IV probit models. The figure shows that the most adverse effect of early childbearing occurs when young women are in the lower secondary school cycle (i.e., having 7–9 years of education).

The OLS estimations underestimate the effect of teenage pregnancy on school outcomes.26 If OLS and IV estimates have a causal interpretation, then the IV and OLS models do not estimate the same parameter. In particular, if the response to treatment (in this case, early fertility) is heterogeneous, then OLS captures a variance-weighted response and the IV captures the response for those young women whose treatment status was affected by the instrument (i.e., local average treatment effect, or LATE; Imbens and Angrist 1994). Thus, the higher IV estimates capture the marginal impact of early childbearing on schooling outcomes for those young women in our sample whose fertility decisions have been affected by the variation in the access (or exposure) to condoms.

Larger IV estimates might suggest that those young mothers on the margin—that is, those girls who face higher costs of condom access and who would have avoided early childbearing if these costs had been lower—experience larger human capital losses than the average young mother. One potential explanation is that young women who are more likely to protect themselves from unintended pregnancy are those who are more likely to lose the most from dropping out of school. For instance, it is possible that if all girls are at risk of engaging in casual sex, those who are more motivated to be in school and thus have higher opportunity costs of dropping out from school are also more likely to use condoms. This hypothesis is consistent with the results of Diaz and Fiel (2016) in the United States, which suggest that those young women who have higher economic prospects and are less likely to be mothers are the ones who experience higher opportunity costs of teenage pregnancy. Unfortunately, our sample size prevents us from testing this explanation, which would require estimating heterogeneous effects and characterizing female subgroups of the sample that are more affected by access (exposure) to condoms.

Impact on Cognitive Skills

Table 5 shows that under the OLS specification, early childbearing is associated with a loss of 0.37 and 0.43 standard deviations (SD) in the math and French standardized test scores, respectively (p < .01). When we account for endogeneity and instrument fertility with access to condoms, this effect increases to 1.13 and 1.14 SD, respectively, for math and French (p < .05). Using exposure to condoms as an instrument, adolescent motherhood decreases by 1.49 and 1.56 SD the math and French test scores, respectively (p < .01). The large and statistically significant differences between the OLS and IV results confirm the expected endogeneity between adolescent fertility and test score achievement.27

The reduction in cognitive skills that results from teenage pregnancy plausibly depends on how long girls have been in school. We estimate OLS models of school attainment on the standardized test scores in math and French including all the cohort members and controlling for the same IV model covariates. We acknowledge the potential endogeneity of school attainment and cognitive skills given that there might be some unobservable characteristics, such as parental preferences, that simultaneously affect grade completion and cognitive skills. Nevertheless, we perform this exercise to compare the magnitude of the average effect of school attainment on the test scores with our estimates of early childbearing. Having completed 9 years of schooling increases the standardized test scores in math and French between 0.9 and 1.25 at the 1 % significance level. This estimate increases to 1.5 SD when individuals complete secondary school (i.e., 12 years of schooling). The longer the stay in school, the larger this effect (Table B.1, Online Resource 1).

We also explore whether the effect of early childbearing on 2012 test scores is partly explained by young women’s earlier test score performance. Accounting for other socioeconomic factors that affect young women’s education and pregnancy decisions, girls who have lower test scores during adolescence might be more likely to drop out of school and engage in risky behaviors or enter into relationships that may result in teen pregnancy. To test this hypothesis, following value-added models of test scores (Todd and Wolpin 2003), we estimate the effect of the 2004 test scores on the 2012 standardized test scores with and without including ever-mother and controlling for the same IV model covariates. We have 2004 test scores for only half the women in the sample; therefore, we cannot use access or exposure to condoms to instrument fertility. We find that the effect of the 2004 tests scores on the 2012 math and French test scores does not change significantly when we included ever-mother (Table B.2, Online Resource 1). Although we acknowledge the potential endogeneity of fertility in these estimations, the results suggest that early childbearing has an impact on these 2012 test scores independent of the 2004 test performance.

Age at First Birth and Schooling

Table 6 summarizes the effect of the predicted age of first birth on schooling and cognitive skills. We observe that a one-year delay in the first birth increases the probability of current enrollment by 5.6 percentage points and the probability of completing lower secondary school by 8.4 percentage points at the 5 % significance level. Postponing the first birth by one year increases the standardized test scores in math and French by 0.19 and 0.21 SD, respectively (p < .01).

Robustness Checks

We conduct placebo tests to analyze further whether the potential nonrandom placement of condom programs represents a challenge for identification. First, we examine the effect of access to condoms on young male school outcomes to explore whether there are alternative channels that are not captured in our community controls through which commune-level condom availability might affect schooling. If communes with access to condoms have better schooling conditions that are unobservable, such as teacher’s motivation, then young women and men’s school outcomes might be positively affected by access to condoms in these communes. We observe that access to condoms does not have a statistically significant effect on any of the young men’s outcomes but has a statistically significant effect on all female outcomes except for years of schooling (Table 7). These findings suggest that the main channel through which access to condoms affects young women’s schooling is the reduction of their early childbearing risk.

Second, we conduct a falsification exercise using young women’s height as the outcome, instead of their school attainment and tests scores. Although our IV models control for the 2001 remoteness index and some conditions when young women were 10 years old, it is plausible that some community socioeconomic conditions and programs, such as health and nutrition interventions, at a very young age and even in utero, might affect later-life school and cognitive outcomes. We would expect such conditions and programs to have helped improve the young women’s health, proxied by height (Strauss and Thomas 2007), as well as their later performance in school and cognition. We find, however, that the coefficients of access and exposure to condoms on young women’s height are close to zero and are not statistically significant, in contrast to their schooling and cognitive outcomes (Table B.3, Online Resource 1).

Finally, we conduct a sensitivity analysis of our IV estimates following the local-to-zero approximation method (Conley et al. 2012). We find that our IV results on schooling and test scores are statistically significant to mild and moderate deviations from the exclusion restriction (section B.1, Online Resource 1).

Discussion and Conclusions

Empirical evidence on the economic consequences of adolescent pregnancy is scarce in developing countries, particularly in sub-Saharan Africa. We contribute to this literature by addressing whether early childbearing causally affects school dropout and cognitive skills among young women in Madagascar, a low-income and high-fertility country. Using panel survey data combined with community censuses, we address endogeneity between fertility and education decisions by instrumenting a young woman’s early childbearing using her community-level access and exposure to condoms since age 15. We control for an extensive set of current and historical community-level covariates to account for the potential endogeneity of program placement and present numerous robustness checks that strengthen our conclusions.

Our findings show that during the transition from adolescence to adulthood, reducing early childbearing or delaying the age of first birth generates substantial gains in female human capital. Young women’s early childbearing increases their likelihood of dropping out of school by 42 percentage points, suggesting that schooling and pregnancy might not be compatible, which is consistent with experimental evidence from Kenya (Duflo et al. 2015) and Malawi (Baird et al. 2011). We also find that adolescent motherhood decreases the chances of completing lower secondary school by 44 percentage points. These findings are in line with Ranchhod et al. (2011) in South Africa; however, those researchers found that teenage pregnancy increases school dropout initially, but the effect on high school graduation at age 22 is negligible, suggesting that mothers can “catch up” on their education. Their result reflects, in contrast to the situation in Madagascar, the policies that facilitate teen mother’s return to school in South Africa (Grant and Hallman 2008; Marteleto et al. 2008).

This shortened stay in school due to pregnancy reduces the standardized test scores in math and French by 1.1 to 1.5 SD—a finding that contributes to the empirical literature in developing countries. The magnitude of this effect is comparable to the effect of secondary school attainment on test scores in our cohort sample, consistent with previous evidence showing a positive impact of secondary school on test scores in Madagascar (Glick et al. 2011) and in Kenya (Ozier 2018). Furthermore, our results suggest that not only the prevention but also the delay of the first birth increases young women’s human capital. Delaying the first birth by a year increases the probability of current enrollment by 5 percentage points and increases the standardized test scores by 0.2 SD.

Bear in mind that our estimations of teen fertility on schooling and cognitive skills depend on the instrumental variable used to establish causality, which can affect a specific subgroup of young women in Madagascar (Diaz and Fiel 2016; Schultz 2007). Thus, these findings are applicable to the sample of young women whose childbearing decisions are affected by the access and exposure to condoms, and consequently may not represent the average young schoolgirl in Madagascar.

Our results underline the potential role for policies that can prevent early childbearing and those that allow teen mothers to “catch up” on their education to protect young women’s human capital investments. In fact, McQueston et al. (2013) highlighted cash transfers, health promotion, and school enrollment as effective policies to decrease adolescent childbearing in developing countries. Our IV results, in particular, suggest that reproductive health and FP policies that lower the costs of postponing the first birth among young women can have human capital gains and prevent poor-health pregnancy outcomes. This evidence is consistent with findings from a large FP program in Colombia that enabled young women to postpone their first birth, thus allowing them to increase their years of education and labor participation in the formal sector (Miller 2010).

Ongoing debate centers on the effectiveness of reproductive health policies in developing countries, particularly concerning whether access to FP policies reduce total fertility (Bongaarts 1994; Pritchett 1994) and improve socioeconomic outcomes (Canning and Schultz 2012; Joshi and Schultz 2013; Miller and Babiarz 2016). Our findings suggest that regardless of any effect on total fertility, the timing of postponing the first birth is crucial to increasing women’s educational attainment and cognitive skills.

Future research should analyze whether this reduction in teen fertility translates into reduced fertility over a woman’s lifetime as well as into improvements in her family’s long-term economic outcomes. For instance, improved human capital for young women who delay their first birth might translate into better health and education outcomes for their children (Ardington et al. 2015; Urdinola and Ospino 2015), breaking the cycle of intergenerational poverty transmission. Therefore, policies aiming to reduce teenage pregnancy may affect not only young women’s economic opportunities but also those of their children.

Furthermore, our results commend continued research on the effectiveness and impact of reproductive health and FP policies in sub-Saharan African countries, such as Madagascar, where the demographic transition is contributing to a bulge in the number of young people, and where there may be a unique opportunity to reap the benefits of enhancing young women’s human capital.

Acknowledgments

The authors thank Günther Fink, George Jakubson, Ravi Kanbur, David Lam, Paul Schultz, anonymous reviewers and seminar participants at the 2013 Cornell Economics Seminar, the 2013 Population Association of America Conference, the 2013 Northeast Universities Development Consortium Conference-NEUDC, the 2014 Population and Reproductive Health Conference, and the 2015 Harvard Population Center seminar series for helpful comments and discussions. This study was funded by the IZA/DFID GLM | LIC Program under Grant Agreement GA-C1-RA4-067. This document is an output from a project funded by the UK Department for International Development (DFID) and the Institute for the Study of Labor (IZA) for the benefit of developing countries. The views expressed are not necessarily those of DFID or IZA. Herrera is very grateful for the support from the Hewlett Foundation/(IIE) Doctoral Dissertation Fellowship. The authors declare that they have no conflict of interest. Any errors are solely the responsibility of the authors.

Notes

1

In developing countries, empirical evidence has shown a positive association between maternal education and children’s health (Strauss and Thomas 1995); however, very few studies have established causality (Behrman et al. 2009; and Breierova and Duflo 2004; Güneş 2015).

2

Complications during pregnancy and childbirth are the leading cause of death for girls aged 15–19 in low- and middle-income countries (World Health Organization 2011).

3

Modern methods include oral contraceptives (“the pill”), female and male sterilization, IUD, injectable contraceptives, implants, male and female condoms, diaphragm, and emergency contraception (INSTAT and ICF Macro 2010).

4

Abortion is illegal, and estimates put abortion rates at 1 per 10 live births. Abortion complications are one of the major contributors to maternal death in Madagascar (Sharp and Kruse 2011).

5

In a related study, Field and Ambrus (2008) found a negative effect of adolescent marriage on schooling in Bangladesh. In this context, female schooling is restricted by marriage; pregnancy comes after marriage. This is different from African countries, where out-of-wedlock pregnancy is common.

6

The 2004 survey defined a community as the catchment area for a primary school. These communities were chosen from a national school-based sampling frame (see Glick et al. 2009). The data are not strictly nationally representative of the entire population, but they closely reflect the main demographic characteristics of our cohort members.

7

Comparisons between female attritors and nonattritors in 2004 on socioeconomic characteristics used to predict early childbearing show that attrition is not a source of bias in our IV results. Results are available upon request.

8

In spite of the activities of the 1996 National Action Plan to encourage single mothers to resume education, government school rules in Madagascar stipulate that pregnant girls be expelled and not allowed back to school after childbirth (United Nations Economic Commission for Africa 2009). This evidence is consistent with our interviews with community-level various stakeholders, who acknowledged that school girls who get pregnant are socially pressured, often by the school principal, to leave the school to reduce reputational costs for the school.

9

The exact question is, “Can the residents obtain condoms in the community? Since when (year) were these available?”

10

In all of the communities with condom availability, community leaders report that condoms are available “at all times,” suggesting that stockouts might not be an issue.

11

In the event of misreporting in the instruments, a classical measurement error does not bias our IV coefficients of early childbearing on schooling (see Online Resource 1, section A.4).

12

We lack information on the age of sexual initiation in our surveys.

13

As a robustness check, we estimate our IV models with different measures of exposure: since age 10, and since the young woman’s birth year. We find that these instruments have lower correlation with ever-mother. The F statistic of exposure to condoms since age 15 is twice as large as the F statistic of these instruments.

14

The 2 SLS IV models that use exposure to condoms are available upon request. Results are similar to the IV probit models.

15

Using a dummy variable for whether the parents were alive when a young woman was 15 does not change the results.

16

Our IV results are robust to the estimation of the models including the young women who dropped out before age 13. Results using the full sample are available upon request.

17

The stable unit treatment value assumption (SUTVA) in our IV model would imply that there are no plausible social network or spillover effects resulting from community-level condom access. To the extent that young women who live in areas without condom access do benefit from condom availability, our first stage might be weakened. However, we do not expect that such spillovers would bias our main results because our IV estimates explore only differences in community-level access to condoms, which will still be positive even in the presence of such spillovers; it is very unlikely that condom distribution points benefit equally the communities that have no access to condoms.

18

Distance is indicated as a separate reason for not using modern contraception (INSTAT and ICF Macro 2010).

19

Despite the low HIV/AIDS prevalence in Madagascar, it remains a serious public health concern (Sharp and Kruse 2011).

20

From the 2007 commune census, information on the number of births and number of women who died during or after delivery in 2006 is available for only 68 and 66 of our 73 communities, respectively.

21

Only 71 of our 73 sample communities were included in the 2001 commune census.

22

Meekers et al. (2006) showed that young women aged 15–24 who self-reported condom access (defined as knowing a condom source within 10 min walking) are 1.8 times more likely than others to have ever used condoms.

23

Our IV results are qualitatively the same when we exclude the community-level controls. Results are available upon request.

24

Results using access to condoms are qualitatively similar. We keep the specifications with exposure to condoms.

25

We fail to reject the null hypothesis of exogeneity under the Haussmann and Durbin Watson test using access to condoms as an instrument.

26

Ashcraft et al. (2013), Ashcraft and Lang (2006), and Klepinger et al. (1999) found that IV teenage pregnancy effects on education were larger than the OLS estimates in the United States.

27

Based on the Haussmann and Durbin Watson test, we reject the null hypothesis of exogeneity for both math and French standardized tests scores at the 5 % significance level, using access and exposure to condoms as IVs.

References

Angeles, G., Guilkey, D. K., & Mroz, T. A. (
1998
).
Purposive program placement and the estimation of family planning program effects in Tanzania
.
Journal of the American Statistical Association
,
93
,
884
899
. 10.1080/01621459.1998.10473745
Angeles, G., Guilkey, D. K., & Mroz, T. A. (
2005
).
The determinants of fertility in rural Peru: Program effects in the early years of the National Family Planning Program
.
Journal of Population Economics
,
18
,
367
389
. 10.1007/s00148-005-0226-5
Angrist, J. D., & Pischke, J. S. (
2009
).
Mostly harmless econometrics: An empiricist’s companion
.
Princeton, NJ
:
Princeton University Press
.
Arceo-Gomez, E. O., & Campos-Vazquez, R. M. (
2014
).
Teenage pregnancy in Mexico: Evolution and consequences
.
Latin American Journal of Economics
,
51
,
109
146
. 10.7764/LAJE.51.1.109
Ardington, C., Menendez, A., & Mutevedzi, T. (
2015
).
Early childbearing, human capital attainment and mortality risk: Evidence from a longitudinal demographic surveillance area in rural KwaZulu-Natal, South Africa
.
Economic Development and Cultural Change
,
63
,
281
317
. 10.1086/678983
Ashcraft, A., Fernández-Val, I., & Lang, K. (
2013
).
The consequences of teenage childbearing: Consistent estimates when abortion makes miscarriage non-random
.
Economic Journal
,
123
,
875
905
. 10.1111/ecoj.12005
Ashcraft, A., & Lang, K. (
2006
).
The consequences of teenage childbearing
(NBER Working Paper No. 12485).
Cambridge, MA
:
National Bureau of Economic Research
.
Azevedo, J. P., Lopez-Calva, L. F., & Perova, E. (
2012
).
Is the baby to blame? An inquiry into the consequences of early childbearing
(World Bank Policy Research Working Paper No. 6074).
Washington, DC
:
World Bank Group
.
Baird, S., McIntosh, C., & Özler, B. (
2011
).
Cash or condition? Evidence from a cash transfer experiment
.
Quarterly Journal of Economics
,
126
,
1709
1753
. 10.1093/qje/qjr032
Bandiera, O., Buehren, N., Burgess, R., Goldstein, M., Gulesci, S., Rasul, I., & Sulaiman, M. (
2015
).
Women’s empowerment in action: Evidence from a randomized control trial in Africa
. Retrieved from https://www.povertyactionlab.org/sites/default/files/publications/111%20Womens%20Empowerment%20June%202015.pdf
Behrman, J. R., Murphy, A., Quisumbing, A. R., & Yount, K. M. (
2009
).
Are returns to mothers’ human capital realized in the next generation? The impact of mothers’ intellectual human capital and long-run nutritional status on children’s human capital in Guatemala
(IFPRI Discussion Paper No. 850).
Washington, DC
:
International Food Policy Research Institute
.
Bongaarts, J. (
1994
).
The impact of population policies: Comment
.
Population and Development Review
,
20
,
616
620
. 10.2307/2137604
Branson, N., & Byker, T. (
2018
).
Causes and consequences of teen childbearing: Evidence from a reproductive health intervention in South Africa
.
Journal of Health Economics
,
57
,
221
235
. 10.1016/j.jhealeco.2017.11.006
Breierova, L., & Duflo, E. (
2004
).
The impact of education on fertility and child mortality: Do fathers really matter less than mothers?
(NBER Working Paper No. 10513).
Cambridge, MA
:
National Bureau of Economic Research
.
Buckles, K. S., & Hungerman, D. M. (
2016
).
The incidental fertility effects of school condom distribution programs (NBER Working Paper No. w22322)
.
Cambridge, MA
:
National Bureau of Economic Research
.
Canning, D., & Schultz, T. P. (
2012
).
The economic consequences of reproductive health and family planning
.
Lancet
,
380
,
165
171
. 10.1016/S0140-6736(12)60827-7
Chandra-Mouli, V., McCarraher, D. R., Phillips, S. J., Williamson, N. E., & Hainsworth, G. (
2014
).
Contraception for adolescents in low and middle income countries: Needs, barriers, and access
.
Reproductive Health
,
11
. https://doi.org/10.1186/1742-4755-11-1
Chong, A., Gonzalez-Navarro, M., Karlan, D., & Valdivia, M. (
2013
).
Effectiveness and spillovers of online sex education: Evidence from a randomized evaluation in Colombian public schools (NBER Working Paper No. w18776)
.
Cambridge, MA
:
National Bureau of Economic Research
.
Conley, T. G., Hansen, C. B., & Rossi, P. E. (
2012
).
Plausibly exogenous
.
Review of Economics and Statistics
,
94
,
260
272
. 10.1162/REST_a_00139
Diaz, C. J., & Fiel, J. E. (
2016
).
The effect(s) of teen pregnancy: Reconciling theory, methods, and findings
.
Demography
,
53
,
85
116
. 10.1007/s13524-015-0446-6
Duflo, E., Dupas, P., & Kremer, M. (
2015
).
Education, HIV, and early fertility: Experimental evidence from Kenya
.
American Economic Review
,
105
,
2757
2797
. 10.1257/aer.20121607
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
34
.
Field, E., & Ambrus, A. (
2008
).
Early marriage, age of menarche, and female schooling attainment in Bangladesh
.
Journal of Political Economy
,
116
,
881
930
. 10.1086/593333
Fletcher, J. M., & Wolfe, B. L. (
2009
).
Education and labor market consequences of teenage childbearing evidence using the timing of pregnancy outcomes and community fixed effects
.
Journal of Human Resources
,
44
,
303
325
. 10.1353/jhr.2009.0026
Free, C., Roberts, I. G., Abramsky, T., Fitzgerald, M., & Wensley, F. (
2011
).
A systematic review of randomised controlled trials of interventions promoting effective condom use
.
Journal of Epidemiology & Community Health
,
65
,
100
110
. 10.1136/jech.2008.085456
Friedman, W. (
2015
,
May
).
Antiretroviral drug access and behavior change
. Paper presented at the 7th Annual Meeting on the Economics of Risky Behaviors,
Balçova, İzmir, Turkey
.
Geronimus, A. T., & Korenman, S. (
1992
).
The socioeconomic consequences of teen childbearing reconsidered
.
Quarterly Journal of Economics
,
107
,
1187
1214
. 10.2307/2118385
Glick, P., Randriamamonjy, J., & Sahn, D. E. (
2009
).
Determinants of HIV knowledge and condom use among women in Madagascar: An analysis using matched household and community data
.
African Development Review
,
21
,
147
179
. 10.1111/j.1467-8268.2009.00206.x
Glick, P., Randrianarisoa, J. C., & Sahn, D. E. (
2011
).
Family background, school characteristics, and children’s cognitive achievement in Madagascar
.
Education Economics
,
19
,
363
396
. 10.1080/09645292.2010.488476
Glick, P. J., Sahn, D. E., & Walker, T. F. (
2016
).
Household shocks and education investments in Madagascar
.
Oxford Bulletin of Economics and Statistics
,
78
,
792
813
. 10.1111/obes.12129
Grant, M. J., & Hallman, K. K. (
2008
).
Pregnancy-related school dropout and prior school performance in KwaZulu-Natal, South Africa
.
Studies in Family Planning
,
39
,
369
382
. 10.1111/j.1728-4465.2008.00181.x
Güneş, P. M. (
2015
).
The role of maternal education in child health: Evidence from a compulsory schooling law
.
Economics of Education Review
,
47
,
1
16
. 10.1016/j.econedurev.2015.02.008
Hotz, V. J., McElroy, S. W., & Sanders, S. G. (
2005
).
Teenage childbearing and its life cycle consequences exploiting a natural experiment
.
Journal of Human Resources
,
40
,
683
715
. 10.3368/jhr.XL.3.683
Imbens, G. W., & Angrist, J. D. (
1994
).
Identification and estimation of local average treatment effects
.
Econometrica
,
62
,
467
475
. 10.2307/2951620
Institut National de la Statistique (INSTAT), & ICF Macro
. (
2010
)
Demographic and Health Surveys of Madagascar
(pp.
2008
2009
). (
2010
).
Antananarivo, Madagascar
:
INSTAT and ICF Macro
.
Joshi, S., & Schultz, T. P. (
2013
).
Family planning and women’s and children’s health: Long-term consequences of an outreach program in Matlab, Bangladesh
.
Demography
,
50
,
149
180
. 10.1007/s13524-012-0172-2
Kane, J. B., Morgan, S. P., Harris, K. M., & Guilkey, D. K. (
2013
).
The educational consequences of teen childbearing
.
Demography
,
50
,
2129
2150
. 10.1007/s13524-013-0238-9
Klepinger, D., Lundberg, S., & Plotnick, R. (
1999
).
How does adolescent fertility affect the human capital and wages of young women?
.
Journal of Human Resources
,
34
,
421
448
. 10.2307/146375
Lee, D. (
2010
).
The early socioeconomic effects of teenage childbearing: A propensity score matching approach
.
Demographic Research
,
23
(
article 25
),
697
736
. 10.4054/DemRes.2010.23.25
Levine, D. I., & Painter, G. (
2003
).
The schooling costs of teenage out-of-wedlock childbearing: Analysis with a within-school propensity-score-matching estimator
.
Review of Economics and Statistics
,
85
,
884
900
. 10.1162/003465303772815790
Lloyd, C. B., & Mensch, B. S. (
2008
).
Marriage and childbirth as factors in dropping out from school: An analysis of DHS data from sub-Saharan Africa
.
Population Studies
,
62
,
1
13
. 10.1080/00324720701810840
Marteleto, L., Lam, D., & Ranchhod, V. (
2008
).
Sexual behavior, pregnancy, and schooling among young people in urban South Africa
.
Studies in Family Planning
,
39
,
351
368
. 10.1111/j.1728-4465.2008.00180.x
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
388
. 10.1111/j.1728-4465.2013.00365.x
Meekers, D., Agha, S., & Klein, M. (
2005
).
The impact on condom use of the “100% Jeune” social marketing program in Cameroon
.
Journal of Adolescent Health
,
36
,
529
530
.
Meekers, D., Silva, M., & Klein, M. (
2006
).
Determinants of condom use among youth in Madagascar
.
Journal of Biosocial Science
,
38
,
365
380
.
Miller, G. (
2010
).
Contraception as development? New evidence from family planning in Colombia
.
Economic Journal
,
120
,
709
736
. 10.1111/j.1468-0297.2009.02306.x
Miller, G., & Babiarz, K. S. (
2016
).
Family planning program effects: Evidence from Microdata
.
Population and Development Review
,
42
,
7
26
. 10.1111/j.1728-4457.2016.00109.x
Mmari, K., & Sabherwal, S. (
2013
).
A review of risk and protective factors for adolescent sexual and reproductive health in developing countries: An update
.
Journal of Adolescent Health
,
53
,
562
572
. 10.1016/j.jadohealth.2013.07.018
Molyneaux, J. W., & Gertler, P. J. (
2000
).
The impact of targeted family planning programs in Indonesia
.
Population and Development Review
,
26
(
Suppl
),
61
85
.
Ozier, O. (
2018
).
The impact of secondary schooling in Kenya: A regression discontinuity analysis
.
Journal of Human Resources
,
53
,
157
188
. 10.3368/jhr.53.1.0915-7407R
Paton, D. (
2002
).
The economics of family planning and underage conceptions
.
Journal of Health Economics
,
21
,
207
225
. 10.1016/S0167-6296(01)00115-1
Pinchoff, J., Boyer, C. B., Mutombo, N., Chowdhuri, R. N., & Ngo, T. D. (
2017
).
Why don’t urban youth in Zambia use condoms? The influence of gender and marriage on non-use of male condoms among young adults
.
PLoS One
,
12
(
3
). https://doi.org/10.1371/journal.pone.0172062
Pitt, M. M., Rosenzweig, M. R., & Gibbons, D. M. (
1993
).
The determinants and consequences of the placement of government programs in Indonesia
.
World Bank Economic Review
,
7
,
319
348
. 10.1093/wber/7.3.319
Pörtner, C. C., Beegle, K., & Christiaensen, L. (
2011
).
Family planning and fertility: Estimating program effects using cross-sectional data (World Bank Policy Research Working Paper No. WPS5812)
.
Washington, DC
:
World Bank Group
.
Pritchett, L. H. (
1994
).
Desired fertility and the impact of population policies
.
Population and Development Review
,
20
,
1
55
. 10.2307/2137629
Raharinjatovo, J. A. (
2014
).
Madagascar (2008): Family planning TRaC study evaluating the condom use and pill use as contraceptive methods among young females 15–24 years—Round three
(PSI TRAC Summary report).
Washington, DC
:
Population Services International
. https://doi.org/10.7910/DVN/27620
Ranchhod, V., Lam, D., Leibbrandt, M., & Marteleto, L. (
2011
,
January
).
Estimating the effect of adolescent fertility on educational attainment in Cape Town using a propensity score weighted regression
. Paper presented at the fifth annual PopPov Conference on Population, Reproductive Health, and Economic Development,
Marseille, France
.
Ribar, D. C. (
1994
).
Teenage fertility and high school completion
.
Review of Economics and Statistics
,
76
,
413
424
. 10.2307/2109967
Rindfuss, R. R., Bumpass, L., & St. John, C. (
1980
).
Education and fertility: Implications for the roles women occupy
.
American Sociological Review
,
45
,
431
447
. 10.2307/2095176
Schultz, T. P. (
2007
).
Population policies, fertility, women’s human capital, and child quality
.
Handbook of Development Economics
,
4
,
3249
3303
. 10.1016/S1573-4471(07)04052-1
Sharp, M., & Kruse, I. (
2011
).
Health, nutrition, and population in Madagascar, 2000–09
(World Bank Working Paper No. 216).
Washington, DC
:
World Bank Publications
.
Staiger, D., & Stock, J. (
1997
).
Instrumental variables regression with weak instruments
.
Econometrica
,
65
,
557
586
. 10.2307/2171753
Stock, J. H., & Watson, M. W. (
2007
).
Introduction to econometrics
(2nd ed.).
London, UK
:
Pearson
.
Strauss, J., & Thomas, D. (
1995
).
Human resources: Empirical modeling of household and family decisions
.
Handbook of Development Economics
,
3
,
1883
2023
. 10.1016/S1573-4471(05)80006-3
Strauss, J., & Thomas, D. (
2007
).
Health over the life course
. In Schultz, T. P., & Strauss, J. (Eds.),
Handbook of development economics
(pp.
3375
3474
).
Amsterdam, the Netherlands
:
Elsevier
.
Sweat, M. D., Denison, J., Kennedy, C., Tedrow, V., & O’Reilly, K. (
2012
).
Effects of condom social marketing on condom use in developing countries: A systematic review and meta-analysis, 1990–2010
.
Bulletin of the World Health Organization
,
90
,
613
622A
. 10.2471/BLT.11.094268
Thomas, D. (
1999
).
Fertility, education and resources in South Africa
. In Bledsoe, C. H., Casterline, J. B., Johnson-Kuhn, J. A., & Haaga, J. G. (Eds.),
Critical perspectives on schooling and fertility in the developing world
(pp.
138
180
).
Washington, DC
:
National Academy Press
.
Todd, P. E., & Wolpin, K. I. (
2003
).
On the specification and estimation of the production function for cognitive achievement
.
Economic Journal
,
113
(
485
),
F3
F33
.
United Nations Economic Commission for Africa
. (
2009
).
African women’s report: Measuring gender inequality in Africa: Experiences and lessons from the African Gender and Development Index
. (
2009
).
Addis Ababa, Ethiopia
:
United Nations Economic Commission for Africa
.
Urdinola, B. P., & Ospino, C. G. (
2015
).
Long-term consequences of adolescent fertility: The Colombian case
.
Demographic Research
,
32
(
article 55
),
487
1518
.
Williamson, N. E. (
2013
).
Motherhood in childhood: Facing the challenge of adolescent pregnancy
.
New York, NY
:
Information and External Relations Division of the United Nations Population Fund
.
World Bank
. (
2013
).
World Development Indicators
. Retrieved from http://data.worldbank.org/data-catalog/world-development-indicators
World Health Organization (WHO)
. (
2011
).
WHO guidelines on preventing early pregnancy and poor reproductive outcomes among adolescents in developing countries
. (
2011
).
Geneva, Switzerland
:
WHO
.

Supplementary data