Impact of Teenage Pregnancy on Education & Labor-Force of Urban Women in Cape Town, SA, Lecture notes of Religion

Download Impact of Teenage Pregnancy on Education & Labor-Force of Urban Women in Cape Town, SA and more Lecture notes Religion in PDF only on Docsity! The Effects of Teenage Pregnancy on Schooling and Labor Force Participation: Evidence from Urban South Africa Natalia Cantet, UIUC∗ Job Market Paper This version: November 10, 2019 Click here for the latest version Abstract Policy makers often express concerns over the lasting implications of teenage pregnancy, due to the observation that young mothers have worse health, less schooling, and poorer job market performance in adulthood. However, because there is selection into early motherhood, the causal impact of teenage pregnancy on human capital investments is difficult to estimate. Additionally, the majority of the literature has focused on high-income settings. I examine the impact of teenage pregnancy in Cape Town, South Africa, on educational outcomes and future labor-force participation using two main identification strategies. I use an instrumental variable strategy that relies on the number of fertile teenage years as an instrument for teenage pregnancy and exploit differences among a subsample of sisters in which one sister reported a teenage pregnancy and at least one did not. I find an increase of approximately 50 percentage points in the likelihood of failing a grade and an increase of 27% (10 percentage points) in the probability of dropping out of school. As for overall school attainment, teenagers who report a pregnancy are, on average, less educated by 1.8 fewer years. Finally, two specific South African characteristics mitigate the negative effects of teenage pregnancy. My findings suggest that strong familial networks, measured by the presence of the mother of the teenage mother, and attendance at a school with higher rates of grade repetition are associated with an attenuation effect of 0.5 and 0.4 years, respectively. Keywords: Teenage pregnancy; Health; Education; Labor Force Participation. JEL Classification: O1, I2, J1 ∗Department of Economics, University of Illinois at Urbana-Champaign. www.nataliacantet.com E-mail: cantet2@illinois.edu. Acknowledgements: I am very grateful to Rebecca Thornton and Marieke Kleemans for their advise during the course of the project. This paper also benefited from the helpful comments from seminar participants at the University of Illinois and conference participants at the Midwest Economic Association Conference, the LACEA’s Health Network Workshop, WashU EGSC 2019, Illinois Economics Association and the many comments received from several Professors and fellow students at UIUC. 1 Introduction Although women around the world are increasingly older when they begin having children, it is estimated that the absolute global number of teenage pregnancies will nonetheless continue to increase until at least 2030 (Monteiro et al., 2019).1 Policy makers, thus, have expressed concerns about the lasting implications of teenage pregnancy on human capital accumulation. The causal effects of teenage pregnancy on short- and long-term outcomes, however, are difficult to estimate. The main challenge is the potential for endogeneity around selection into early motherhood. Poorer human capital outcomes could be due to the causal impact of teenage pregnancy or because mothers who have teenage births are negatively selected into it, such that they would have had poorer outcomes regardless of their age at pregnancy. To overcome this concern, researchers who study teenage pregnancy in the United States and other high-income countries have taken several empirical approaches: propensity score matching (e.g., Levine and Painter, 2003 and Lee, 2010), with-in family differences (e.g., Herrera, Sahn, and Villa, 2019), and instrumental variables (Ashcraft, Fernández-Val, and Lang, 2013). However, the number of papers that utilize these methods to estimate the impact of teenage pregnancy in low- and middle-income countries is smaller.2 Importantly, the impact of teenage pregnancy in low- and middle-income settings might be different due to context-specific characteristics. In this paper, I examine the causal effects of teenage pregnancy on the educational attain- ment and labor-force participation of young urban women in Cape Town, South Africa. To identify these effects, I consider two main identification strategies: an instrumental variable 1The average global teenage birth rate (aged 15 to 19 years) decreased from 65 per 1000 girls in the mid-1990s to 49 in 2011 (WHO, 2018 and Sedgh, Finer, Bankole, Eilers, and Singh, 2015). Yet, because the teenage population is growing is growing, the absolute number will grow. 2Exceptions include Branson and Byker (2018), Ranchhod, Lam, Leibbrandt, and Marteleto (2011), Ardington, Menendez, and Mutevedzi (2014), Urdinola and Ospino, 2015, Azevedo, Lopez-Calva, and Perova (2012), Narita and Diaz (2016) and Herrera et al. (2019). 1 Consistent with the notion that teenage pregnancy has a negative effect on education, women who report a pregnancy in their teens are more likely to lag behind their peers in their education and have lower school attainment. The analysis of the school progression instrumental variable approach yields a 55-percentage-point increase in the likelihood of failure, a lag of 0.3 years and an increase of 27% in the risk of dropping out. The overall results suggest that women who report a pregnancy during their schooling years are, on average, 1.8 years less educated. Similarly, in the sibling comparisons estimation strategy, I find that women who report a pregnancy in their teens are 12.8% more likely to fail a grade and are 0.5 years behind their peers. The risk of dropping out increases by 49.4 percentage points. The school attainment effect is smaller in the sibling approach. Sisters who report a pregnancy are 0.774 years less educated, 44% (15.9 percentage points) more likely to sit for the matriculation exam, and 5 percentage points less likely to enroll in training or formal institution after the end of school. The second group of outcomes studied in this paper describes the labor-force participation following the probable end of school. I study whether reporting an early pregnancy has an impact on the labor-force participation at the ages of 19, 20, 21, and 22. Across both strategies, I find positive, though not statistically significant, effects. This result is indicative of substitution between postsecondary education and participation in the labor force. In this paper, the evidence on educational outcomes indicates a strong negative effect on the educational attainment outcomes. However, the opportunity cost of teenage pregnancy might be different in South Africa than in other high-income countries due to characteristics that are specific to setting. In particular, at the time of the study, the country had extremely high youth unemployment rates (Nattrass & Walker, 2005), strong familial networks (Duflo, 2003 and Magruder, 2010); and extremely high rates of grade repetition (Anderson, Case, & 4 Lam, 2001).5 Thus, using school and family composition data, I analyze whether there are attenuation effects of the second and third points in the instrumental variable identification strategy using a control function approach with interactions between teenage pregnancy and the attenuation variable of interest. In order to test the whether stronger familial ties attenuate the finding that teenage preg- nancy negatively impacts educational attainment, I first analyze whether the opportunity cost of early motherhood might be changed when the mother of the teenage mother is alive during her teenage years. A bigger network will change the cost of childcare and may change the relative cost of being enrolled in school. The findings in this paper provide suggestive evidence that having a mother mitigates the effect of teenage pregnancy by 0.5 years. Next, although female educational attainment was extremely high in South Africa, the country was also notable for its high rates of grade repetition. The class composition of grades is such that the spread in ages within grades is wider than in high-income countries (Anderson et al., 2001; Lam, Marteleto, and Ranchhod, 2013). Also, many teenagers have been found to come back to school after birth (Marteleto, Lam, & Ranchhod, 2008). In these types of classrooms, the social stigma of teenage pregnancy might decrease and encourage students to stay in school. I find that the interaction between teenage pregnancy and attending a school with an above-average grade failure rate leads to a positive mitigation effect of 0.41 additional years of education. Methodologically, this paper speaks to the literature of the effects of early pregnancy on short and medium term human capital investments in the United States, and more generally on high-income countries.6 Two papers, Klepinger et al., 1997 and Ribar, 1994, have used 5As for the labor force, women were found to be less likely to be in the labor force and more likely to be unemployed at the turn of the century. 6In Table B3 in the appendix, I present a comprehensive review of the literature in high-income countries. See Ribar, 1999; Hotz, McElroy, and Sanders, 2005; Ashcraft and Lang, 2006a; Diaz and Fiel, 2016 and Heiland, Korenman, and Smith, 2019 for reviews of the literature. 5 the age at menarche as an instrument for teenage pregnancy.7 In the United States, papers that use the strategy has rendered a lower impact on the educational attainment than those found in this paper. Furthermore, the second approach I undertake in this paper, the with-in family differences strategy, has also been explored by the literature in high-income countries. The main papers that use this strategy typically yield educational effects that are between 0 and 1 fewer years of education (Geronimus and Korenman, 1993; Ribar, 1999; Duncan, Lee, Rosales-Rueda, and Kalil, 2018 and Heiland et al., 2019). The magnitude of the effects is consistent with the magnitude found in this paper.8 This paper also adds to the smaller thread of literature that studies the impact of early pregnancy on education in low- and middle-income countries. Studies of Latin America have found that teenage mothers are less educated (Azevedo et al., 2012; Narita and Diaz, 2016; and Berthelon and Kruger, 2017). For Sub-Saharan Africa, Madhavan and Thomas (2005) and Marteleto et al. (2008) among others have documented an association between teenage pregnancy and lower educational attainment. Two papers, Almanza and Sahn (2018) and Branson and Byker (2018), consider birth expansion policies in Madagascar and South Africa, respectively. Their findings provide evidence that the delaying fertility increased human capital investments significantly. My identification strategy differs from these papers because I rely on individual teenagers making decisions about their own sexual behavior.9 Furthermore, my paper is closer to two papers that study the effects of teenage pregnancy in South Africa. Ranchhod et al. (2011) analyzed teenage pregnancy using a propensity-score matching strategy in the CAPS dataset. The evidence in this paper suggests a moderate 7Additional instruments have utilized miscarriages (Hotz et al., 2005; Ashcraft and Lang, 2006a; Fletcher and Wolfe, 2009; and Ashcraft et al., 2013), and abortion laws (Bitler & Zavodny, 2001). 8A third commonly used strategy was the propensity-score matching identification. Generally, these papers find modest negative effects on educational attainment and life satisfaction (Levine and Painter, 2003; Lee, 2010; and Zito, 2018). 9Table B2 in the appendix includes a comprehensive review of the literature on low- and middle-income countries. 6 2.1 Sexual Debut, Early Childbearing and Marriage The relationship between teenage pregnancy and female outcomes has been at the center of the public concerns in many developing countries. The debate is usually focused on the perceived perverse incentives created by social benefits programs (Moultrie & Dorrington, 2004).11 Nevertheless, researchers have examined the determinants and consequences of teenage sexual initiation, and childbearing in developing countries (Lloyd and Mensch, 2008, and Marteleto et al., 2008) and have found that early sexual debut has several important implications for the likelihood of teenage pregnancy and other transitions to adulthood. An earlier age at the time of first sexual intercourse increases the likelihood of getting and transmitting sexually transmitted diseases (STDs) and HIV, as well as of reporting a pregnancy (Marteleto et al., 2008). In Sub-Saharan Africa, reporting of early entry into premarital sex by girls has been linked to the likelihood that they will drop out of school (Biddlecom, Gregory, Lloyd, and Mensch, 2008). In most low-income countries, women’s first sexual intercourse occurs largely within mar- riage. This is the case, for example, in many North Africa and some Asian countries (Singh, Wulf, Samara, & Cuca, 2000). However, the difference between the age at which women enter into sexual relationships and their age at marriage is expanding as more girls are reporting extramarital sexual activities in their teenagers. Personal characteristics and prior school experience have been linked to earlier the sexual and reproductive behavior (Marteleto et al., 2008). A report by the World Health Organi- zation (WHO) cites many barriers that stand in the way of teenagers practicing safe sex in low-income countries (WHO, 2018). First, access to contraception is sometimes restricted by 11In the particular case of South Africa, President Jacob Zuma’s 2008 election campaign included a proposal that teenage mothers be separated from their babies and forced to get an education (Ardington et al., 2014). 9 strict provision policies (Branson and Byker, 2018) or by health worker bias. Other barriers to consistent and correct use of contraception may also be linked to pressure to have chil- dren or the stigma surrounding non-marital sexual activity and lack of knowledge regarding correct use. Concerns over teenage pregnancy relate to its impacts on the health, well-being, and life course trajectories of the mothers and their infants. From a health perspective, teenage pregnancy has been linked to poor perinatal outcomes, low birth weight, and preterm birth. Policy makers have also noted that teenage mothers are often poorer, less educated, and less likely to be employed (WHO, 2018). Despite these concerns, teenage pregnancy and childbirth are in some settings planned events. Girls who get married early are often less able to effectively negotiate their sex practices (which facilitates sexually transmitted infections) and face pressure to have children (WHO, 2018). Unicef (2014) cites the example of Nepal, where women who married before the age of 15 are 33 percentage points more likely to have three or more children by the age of 24 as compared to the 1% of women who marry as adults. Alternatively, planned extramarital pregnancies have been linked to improvements in the social status of teenagers in the lower socioeconomic strata in Brazil (Heilborn and Cabral, 2011 and Faisal-Cury et al., 2017). 2.2 The South African Context 2.2.1 Sexual Debut and Early Childbearing Teenagers in South Africa are unique among developing nations in terms of their initiation into sexual behavior in that women become sexually active by age 18.12 The fact elicits 12The median age at first sexual intercourse in the Department of Health/South Africa and Macro Inter- national, 2002 was 17.8 years for women aged 20 to 24 years. 10 two distinctive conclusions regarding teen girls in South Africa. First, in this context young women usually become sexually active while they are still in school. Second, in contrast to many developing countries, sexual initiation occurs predominantly outside of marriage. The median age at marriage among women aged 25 to 49 was 24.2 years (DHS, 2002). Overall fertility levels in South Africa are low compared to other African countries (WHO, 2014). Teenage fertility rates, however, relatively high through the late 1990s and early 2000.13 The proportion of women aged 19 who had reported a pregnancy in the 1998 and 2003 South African DHS was 35.1% (as seen in Figure 1)and 27.1%, respectively 2007).14 As for the negative consequences of teenage pregnancy, another distinctive fact is that many South African mothers return to complete their schooling after giving birth. This is due to the support received from their families and fathers, who often recognize their children (Kaufman, De Wet, and Stadler, 2001 and Madhavan and Thomas, 2005). In my study’s particular setting, the Cape Area, teenage pregnancy was estimated to be approximately 22% in 2002. The average is similar to the national level, which is 25%. Compared to young adults in other regions, however, those who live in the Cape Area are 24 percentage points less likely to report a teenage birth than those in rural areas. 2.2.2 Schooling General education in South Africa is divided into three periods: primary, middle school and secondary school. Schooling is compulsory until grade 9, and spans 12 grades in total. 13Contrary to many other sub-Saharan countries, contraceptive usage in South Africa was high during the apartheid regime due to the government’s plan to control the non-white population (Cooper et al., 2004). Public clinics, hospitals and mobile services largely provided contraceptives for free. However, not everybody benefited from this access to family planning, as evidenced by the surge in unintended pregnancy rates among adolescents. The findings point to social barriers to access to family planning for teenagers. 14Abortion was legalized in the country in in 1996. However, even when public and private facilities in- creased progressively, teenagers have not reported using pregnancy termination services (Panday, Makiwane, Ranchod, & Letsoala, 2009). 11 before apartheid, such that there is no subsistence agriculture in rural settings. In more urban settings, the informal sector is relatively small compared to similar countries. Informality was outlawed previously but grew during the first years of the 2000s as the number of individuals who reported being self-employed or domestic workers increased from 19% in 1993 to 24% in 2004 (Magruder, 2010). As for the gender composition of unemployment, Banerjee et al. (2008) find that women are less likely to participate in the labor market and more likely than men to be unemployed. Female labor supply increased drastically in the early 2000s but the demand did not accom- pany the influx as employment in south Africa’s bigger industries, agriculture and mining employment steadily fell. 3 Data 3.1 Cape Area Panel Study To study the effects of teenage pregnancy on education, and labor supply, I use the Cape Area Panel Study (Lam et al., 2006). This is a longitudinal study, which follows young men and women who lived in the Cape Town Metropolitan Area in 2002.20 Individuals were sampled using a stratified two-stage sample of households, from sample clusters first and then through households within these clusters (Lam et al., 2006). Clusters were selected according to the breakdown of ethnic groups available in 1996 census, where white and black clusters were oversampled to achieve a representative sample (Lam et al., 20Cape Town is the second largest city in the country after Johannesburg, with 2,785,032 inhabitants, and it is the provincial capital of the Western Cape. It is located in the southwestern corner of South Africa. The population in Cape Town is 35% black, 44% coloured, 19% white, and less than 2% Asian (Statistics South Africa, 2007). The composition is different from the national population, which is 79% black, 9% coloured, 10% white, and 3% Asian. 14 2006). The design for the wave included young-adult and household questionnaires. It also administered a literacy and numeracy evaluation of the young adults in the sample. Five waves were conducted to create the CAPS data set: first in 2002, then in 2003–2004, in 2005, in 2006 and in 2009. The first wave, which was administered in August-December of 2002, surveyed 2,612 young women aged 14 to 22 in 2002 in 2,045 households. This paper draws heavily from the young-adult questionnaire in wave 1. The background individual information variables were created using the questions regarding the race, religion, place of birth, language utilized in the interview, and education of the parents from the young-adult questionnaire. Number of full siblings, household size and wealth and dwelling information were also created using questions from the household questionnaire. The pregnancy and fertility information is also captured in wave 1, which contained ques- tions on the complete pregnancy and birth histories, prenatal care, and partners’ information and was later updated in the subsequent surveys. The self-reported information on preg- nancy provides the basis for the early pregnancy history of the older cohorts, who are older than 18. Among the younger cohorts, 3.83% reported a pregnancy by 2002, so most of the inputs for the pregnancy variables are drawn from later waves. Next, the age at menarche was first asked in 2002. Almost all sampled women, 96.8%, provided the age in this wave. Wave 2 was conducted between July 2003 and December 2004 in two separate survey rounds: wave 2A in 2003 and wave 2B, in 2004. The total number of sampled women that were reached in the study was 2,140 (748 in 2003 and 1,410 in 2004). The main goal of wave 2 was to update the data collected in wave 1. There were some differences between waves A and B, as the 2003 survey added a module on HIV/AIDS stigma and the 2004 interview included modules on employment and school choice. Even though the pregnancy and birth history were not directly included in wave 2, the surveys inquire about reasons for not attending school. I am thus able to determine whether women reported pregnancy 15 during those years. Wave 3 was conducted between April and December 2005 to 1,911 young women now aged 17 to 25. The questions for young adults focused on schooling, employment, fertility, and personal health. Additional variables described residential and schooling history, in- tergenerational transfers, time allocation, and sexual partners. Particularly useful for this study was the update on age at menarche for those who did not answer the question in the first wave. The question extends the age at menarche to 103 additional women. Wave 4 was conducted in 2006 and accounted for 1,877 sampled women, now ages 18 to 26. There were three targeted populations: the young adults, their biological children, and older original residents (age 50 or over). It mainly consisted of follow-up information on the school, work, and childbearing histories of CAPS young adults. An important feature of wave 4 is that it includes a set of health outcomes for young adults and their parents. The fact that the module was included in wave 4 limits the sample size. Attrition was greater among the older, wealthier, and more educated young adults in the panel. Finally, wave 5 was administered in 2009 to 1,799 women, aged 21 to 29. It included a young-adult questionnaire, a young-adult telephonic questionnaire, and a young-adult proxy questionnaire. Field work was carried out in 2009 and respondents who were not successfully located in the field were contacted via telephone to update their basic information. The fifth wave of the CAPS data set updated educational outcomes and provided restrictive information for the fertility variables. 3.2 Variable Construction This paper takes advantage of the data available in the various waves and retrospective information questions asked in the first CAPS wave to conduct the analysis. Using the 16 or passed each year. Thus, by combining grade data with year data, I study two important sets of outcomes: (I) the sampled women’s progression through their stages of schooling and, (II) the sampled women’s overall educational attainment. I begin by analyzing schooling progression in the panel structure. Conditionally on being enrolled in a particular year, I consider the likelihood of failing the grade. The variable “Failureit”, for individual i observed in year t, is an indicator variable that is equal to one if a sampled woman reported having failed a grade or having dropped out midyear, and zero otherwise. Second, grade for age captures the lag in education. I follow (Glynn et al., 2018) by setting a measure of how much behind in school and censoring the lags up to two grades behind. I only consider grade failure and grade-for-age measures until grade 12 in order to consider the impact of pregnancy during years of schooling. The third variable considered is a dropout measure created using enrollment levels and the passing grade questions. For sampled woman i, “Drop Outit” is then defined as one for those who report not being enrolled in schooling in year t, and zero if they were. In the static analysis, conducted in the collapsed panel, I study educational attainment using three outcome variables: completed years of education, whether they ever took the ma- triculation examination, and whether they continued their education after high school. First, “Years of educationi” is a continuous variable measuring years of education completed during the last observed period. Next, “Took Matrici” (or NSC exam) is an indicator variable of whether the individual sat for the NSC at some point. Finally, “Post secondary Educationi” is equal to one if the individual kept studying after high school, and zero otherwise. 3.2.3 Labor Force Participation Outcomes In order to study what happens to these women after the age of 18, I study the effects of teenage pregnancy on their labor-force participation after the age at which they should have 19 finished high school. Specifically, I consider the labor-force participation for ages 19, 20, 21, and 22.23 3.3 Samples In this paper, the analytical sample is comprised of the full sample of 1,741 women. The analysis is conducted on two sample formats: a panel and a static samples. The former is constructed using a combination of the retrospective self-reported information available in the first wave, and updated using the questions asked in the following waves. The resulting data follows the sampled women from birth to 2009 such that the unit of observation is at the woman-year level. Since the goal of the paper is to study the effects of teenage pregnancy, the analysis is conducted between the ages of 10 and 20. Next, for the static analysis, I collapse the information into a “collapsed panel” where each observation is one sampled woman. Selection of the women for the present study is based on three main criteria: age at menarche, health module availability, and sampling location. Specifically, I limit my study to women who have undergone menarche between the ages of 10 and 17, have health infor- mation, and live in clusters where there is at least one other person in the cluster. First, medical researchers have established that menarche is delayed when girls report its onset two standard deviations (years) after girls of similar background (Hillard, 2013). Among the women in the full sample, the average age at menarche for those who report the onset was 13.433 (1.667 SD). I thus analyze women who have undergone menarche until the age of 17. The percentage of women reporting reaching menarche after that age, however, 23Also, in the appendix Table A13, I examine the extensive and intensive margins of women in the sample for the year 2006. The information available in wave 4 allows me to study both margins because the round includes questions about the number of hours the respondent devotes to working and the number of hours she dedicates to studying. If she reported that she attends school, I sum the number of hours. In this table, I also examine the willingness of the sampled women to accept positions as domestic workers and security guards that pay R900 and R1300, respectively. 20 is small: only 1.32%.it is chosen as the lower bound for the age at menarche in this study. The sampled women interviewed in the first wave who report reaching menarche before this age is small (0.85%).24 Two additional requirements relate to health and sampling clusters. The health module was only asked in wave 4, to a total of 1,790 out of the 1,851 women who were surveyed in 2006 with in the chosen menarcheal age range. Finally, the sampling location limitation requires women living in the same sampling cluster with at least one other woman in it. The last criteria censors an additional 1.8% of the women sampled in wave 1. The second sample of women is comprised of women who share a parent, report living in the same household, and have differing teenage pregnancy reports. Given the criteria, the sample of siblings is small, 418 women, and they come from bigger households than the composition in the full sample. 3.4 Summary Statistics In Table 2, I provide summary statistics for the sampled woman in the full sample and in the women-with-sisters sub-sample. The average age in 2002 of the all of the women studied in this paper is 17.723. The sisters in the comparison data set are slightly younger, their average age is 17.902. As for racial composition, 46.1% of the individuals are coloured and 48.6% are black; whereas in the sibling sample, the percentages are 40.9 and 56.7, respectively. The education of the mothers of the women is lower in the sister sample, as they are on average, 0.3 years more educated than in the sister sample (8.271 vs. 8.773 years). In contrast, as can be seen in Table 2, because the sample of siblings belong to households where there are at least two young adults in 2002, the average household size is greater by 0.54 individuals (5.808 versus 6.348). Similarly, the number of full siblings is 2.312 in the full sample, but it 24The percentage of women who fell outside the bound amounts to 2.17%. 21 possible by considering factors that affect the likelihood of teenage pregnancy but are not directly linked to educational outcomes.25 In the presence of heterogeneous treatment effects, the instrumental variable estimation coefficients identify the local average treatment effect (LATE), capturing the average effect for those induced to report a pregnancy in their teens by reaching menarche earlier. That is, the instrument captures the effects of teenage pregnancy through the compliers.26 4.2.1 Panel Dataset I estimate the following two-stage instrumental variable model: First Stage: Pregnanticst = σ1 + σ2Post Menarcheicst + β3Xicst + ςt + λi + εicst (2) Second Stage: Outcomeicst = β1 + β2Pregnanticst ∧ + β3Xicst + ςt + λi + εicst (3) In the first-stage equation, 2, Pregnanticst is an indicator variable that denotes whether the sampled woman i of cohort c from the sampling cluster s reported a pregnancy in year t. Post Menarcheicst indicates whether individual i was fertile in that year. In this equation, ϑt, and λc denote time and individual fixed effects, respectively. The Xicst is a set of time varying controls such as the age in year t. Finally, εitc is the error term. In the second-stage equation, Outcomeitxc is the educational, labor-supply, or fertility outcome estimated for the sampled woman i of cohort c from the sampling cluster s, who 25The previous literature has used a variety of instruments to identify the causal effect, including mis- carriages, the age at menarche, and abortion or family-planning policy changes (Ribar, 1999; Bitler and Zavodny, 2001; Ashcraft and Lang, 2006b; Fletcher and Wolfe, 2009; Azevedo et al., 2012, Ardington et al., 2014 and Almanza and Sahn, 2018). 26The instrumental variable methodology has some important caveats. It is often difficult to find a strong instrument and that the resulting estimate can be imprecise. Duncan et al. (2018), for example, chose a sibling-and-cousin fixed-effects methodology for the impact of maternal age on child development in the United States. Using the same data, this paper also finds that the age at menarche is a weak instrument for maternal age at first birth. Additionally, the authors suggest that miscarriages (Hotz et al., 2005) and state abortion laws (Bitler and Zavodny, 2001)are stronger predictors of teenage pregnancy, but resulted in standard errors that were too large and unable detect significant effects. 24 reported a pregnancy in year t. The coefficient of interest, β2, describes the relationship between the chosen outputs and teenage pregnancy. The controls and fixed effects are the same as in the first-stage equation. The error term is represented by εicst. Hazard Estimation To estimate the likelihood of dropping out, I utilize the same instrumental variable iden- tification strategy in a hazard model which can be described in the following equation: Drop Outicst = β1Pregnanticst + β3Xicst + ςt + λi + εicst (4) In Equation 4, Dropouticst is coded as an indicator variable for whether individual i dropped out in that period. The reminder variables in Equation 4 are defined in the same way as in Equation 3. Also in this equation, ho represents the baseline risk, defined as defined as ho(t) = ptp−1. The Weibull model estimates an expected survival time. Because I am interested in teenage pregnancy, I limit the data to women between the age of 10 and 20. Hence, women who drop out before the age of 10 are left censored and women who drop out at the age of 20 are right censored. To address endogeneity in the context of the hazard function, I take a control-function approach in which the first stage is estimated linearly (Wooldridge, 2015). The variance- covariance matrix is also corrected to accurately estimate the standard errors. The first stage allows me to generate a predicted likelihood of teen pregnancy that is used to tease out the confounding factors of teenage pregnancy. 4.2.2 Collapsed Panel In the Collapsed Panel sample, the two-stage instrumental variable is estimated as follows: First Stage: Pregnant ≤ 18ics = σ1 + σ2Fertile Y earsitc + σ3Xi + ϑs + λc + εics (5) Second Stage: Outcomeics = ϕ1 + β2Pregnant ≤ 18ics ∧ + β3Xi + ϑs + λc + υics (6) 25 In Equation 5, Fertile Y earsics is defined as a continuous measure of the number of fertile years between the age of menarche and the age of 17 for a sampled woman i of cohort c who lives in the sampling locations.27 Pregnant ≤ 18ics is equal to one if the same individual reported a pregnancy before the age of 18 and zero otherwise. Additionally, Xi is the set of individual controls including height, literacy status of the mother of the teenage mother, language spoken at home, race, whether the woman was born outside the Western Cape, her self-reported religion, the normalized literacy exams, number of full siblings and household, and asset characteristics index. Additionally, is the set of individual controls including height, literacy status of the mother of teenage mother, language spoken at home, race, whether the woman was born outside the Western Cape, her self-reported religion, the normalized literacy exams, number of full siblings and household, and asset characteristics index. The ϑs and λc are sampling location and cohort fixed effects. Finally, εitc is the error term. In Equation 6, Outcomeics is the educational or labor outcome estimated. The coefficient of interest, β2, describes the relationship between the chosen outputs and teenage pregnancy. The controls and fixed effects are the same as in the first stage equation. The error term is represented by εics. The inclusion of the sampling location fixed effects (λc) accounts for the adverse events that may have occurred at the year of birth of the sampled women that will also be linked to both the age at menarche and the outcomes of interest. They are included in order to account for any geographical condition that may affect the age at menarche. Finally, I control for the women’s adult height, which has been found to be linked to childhood nutrition as a proxy for childhood nutritional shocks. 27Mathematically, I define this continuous variable as 17 minus the age at which she had her first men- struation. 26 In Tables A6 and A7 in the appendix, I present two variations of the identification strategy for the collapsed panel instrumental variable approach. First, in Table A6 I analyze whether changing the thresholds for teenage pregnancy between the ages of 16 and 21. Younger ages are associated with greater coefficients and stronger F-statistics. In Table A7, I present three variations in the construction of the instrument. Instead of utilizing 17 minus the age at puberty, I change the upper bound to 18, 19 and 20. The coefficients are similar in size and F-statistic is greater than 10. The results across all specifications describe a positive relationship between the number of fertile teenage years and teenage pregnancy. Each additional fertile year during a woman’s teenage years increases the likelihood that a sampled woman will report a teenage pregnancy. Table 4 clearly describe the positive sign described by the coefficients of interest in both samples.30 In the next section, I turn to issues associated with the validity of the instrument. 4.2.4 Validity Issues 4.2.4.1 Validity of the Instrument Since the identification strategy relies heavily on the correlation between age at menarche and likelihood of teenage pregnancy, I examine the instrument’s validity. Importantly, for the instrument to successfully tease out endogeneity, the exclusion restriction requires that the relationship between teenage fertility and adult outcomes is fully mediated by changes in the likelihood of teen pregnancy.31 There are, however, two important concerns regarding the exclusion restriction of the age at menarche as an instrument for the likelihood of teen 30The pattern is consistent with previous findings (for example, Ribar, 1994 and Klepinger et al., 1997). 31It is worth mentioning the monotonicity condition. Given the set-up of the study, girls who experience a late menarche, even when using the chosen conservative definition, are very unlikely to experience a pregnancy during their schooling years. The decrease is largely due to the biological impossibility of having a child without undergoing menarche. It is thus possible that the setting rules out the existence of potential defiers. The monotonicity property of the age at menarche as an instrument for teenage pregnancy is therefore satisfied, if we account for the health shocks. 29 pregnancy. The first relates to the exogeneity of the instrument, and the second concerns the social consequences of the onset of menarche. Exogeneity of the Menarcheal Age The medical debate over whether the age at menarche is an exogenous event is volumi- nous. There is a large body of evidence suggesting that long-run health is determined by shocks that happen in the prenatal and perinatal period (Almond, Chay, and Lee, 2005; Alderman, Hoddinott, and Kinsey, 2006 and Almond and Mazumder, 2011) and in early life (Akresh, Bhalotra, Leone, and Osili, 2012; Mahmud, Shah, and Becker, 2012 and Almond, 2006). The debate also has implications for whether early childhood nutrition determines the age at menarche. There is a strand of the literature that argues that a random genetic component explains the age at first menstruation (Mpora et al., 2014; Jahanfar, Lye, and Krishnarajah, 2013, Sørensen et al., 2013, Adair, 2001), which would suggest that the menar- cheal age is quasi-random. In contrast, a group of authors have found that early childhood nutrition and socioeconomic status are linked to the timing of puberty among girls (Kara- panou and Papadimitriou, 2010; Dahiya and Rathi, 2010; and Rah et al., 2009). In this case, childhood nutrition may affect long-run well-being through human capital development or through its impact of the timing of menarche.32 To address the debate over the exogeneity of the age at menarche, I proceed to examine the question in two ways. My first approach is to examine an indicator of adult health that has been linked to prepubescent health status and nutrition: adult height (Martorell and Habicht, 1986; Fogel, 1993; Silventoinen, 2003). Since in 2006 the younger cohort sampled in the CAPS Data set was 18 years of age, this height variable is the best available proxy of 32More recently, Khanna (2019)’s main corollary is that the age at menarche is a poor instrument for the age at marriage in India because it does not satisfy the exclusion restriction. Using data from the Young Lives panel, reaching menarche early (defined as before the age of twelve) decreases school enrollment by 13%. It has been argued that nutrition is linked to the timing of menarche among girls. 30 degree of stunting caused by poor nutrition or health issues in childhood. Figure 6 displays the relationship between age at menarche and adult height. As shown, the relationship is weakly positive but not statistically significant. Furthermore, following Field and Ambrus (2008), Sekhri and Debnath (2014) and Chari et al. (2017), I include the height of the sampled teenage women among the vectors of controls to account for any remaining within-sampling location variation in environmental conditions. A further test of the exogeneity can be found in Table A3. In this table, I present some descriptive characteristics for girls who have reached menarche by the age of 14, and those who report reaching it later. As seen, most of the demographic characteristics remain balanced across groups. The race composition and the education of the mothers, however, are different among those who reached menarche in different age groups. 33 My second approach to the endogeneity question is to examine another threat to the exogeneity of the instrument, the fact that environmental factors, such as toxins that are specific to a location, may cause delays in menarche.34 The CAPS data is mostly homoge- neous: 80% of the population report having lived most of their lives in a formal or informal urban setting. Nevertheless, I control for sampling location fixed effects in the static model and test whether the sampling clusters faced different environmental factors. I also compare the poverty level per sampling location and the height in Figure 7. This relationship is not statistically significant. Third, since menarche is an event that may have occurred a few years earlier, an additional feasible concern would be the possibility of recall bias. The age at menarche is established using a self-reported account of the year in which the sampled women reached menarche. In cases of misremembrance, the first-stage coefficients would not correctly estimate the 33In Table A9 in the appendix, I present the estimates for the largest race group in my sample: black women. This group is driving the results. 34This concern is particularly important because girls raised in urban environments have been found to display earlier menarcheal ages than those who grew up in rural environments. 31 In columns (1) and (2), the main regressor is the age at menarche. In column (1) the estimation does not include controls and in column (2) the estimation includes indicator variables for race, cohort, religion, household size, the literacy level of the mother of the teenage mother, the teenage mother’s place of birth, and the household size. Column (2) also includes sampling cluster fixed effects. The sample includes all women who have reported an age at menarche. Columns (1) and (2) provide evidence that attrition is not statistically associated with the age at menarche. Both coefficients are negative, close to zero and not statistically significant. In columns (3) through (6), the regression includes an indicator variable for pregnancy before the age of 18. As with the age at menarche, I analyze the relationship between attrition and teenage pregnancy with and without controls and sampling cluster fixed effects. The sample in columns (3) and (4) is composed of all of the women surveyed in the first wave, whereas the regressions presented in columns (5) and (6) limit the sample to those who were already 18 years of age in the first wave (i.e., their teenage pregnancy status was already defined in 2002). The teenage pregnancy coefficient in column (3) is equal to -0.088 (statistically significant at the one-percent level). However, when controls and fixed effects are added, the association between teenage pregnancy and attrition diminishes to 0.044 and is no longer statistically significant. Similarly, when the sample is restricted to women who were older than 18 in 2002, between the Eastern Cape and municipalities in the Western Cape driven by labor demand. During the period of interest, urbanization rates were high and migration was mostly urban-to-urban, but homelands were also affected (Posel, 2004). Historically, migration was highly regulated by the government. In this context, mostly male migrants moved for labor reasons and migration was circular or temporary. Once the policies were lifted, although the expectation was that it would became permanent, temporary migration remained high. Migrants from the East Cape to the West Cape were more likely to be male, young (25 to 29 years), and unmarried. They were generally from low income households unemployed or not economically active. The inflow of migrants to the cities brought additional housing, health, education and sanitation challenges to the local authorities. Jacobs and Du Plessis, 2016 estimated that, once these migrants arrived in Cape Town, 31.3% lived in informal dwellings in backyards or informal settlements. 34 the coefficients are smaller but are only significant when additional variables are included. Even though there is no statistically significant relationship between teenage pregnancy and attrition with the inclusion of controls, I will include sampling-cluster fixed effects, as specified in the empirical approach section. I also present, in Table A12 in the appendix, an alternative specification that includes inverse probability weights. Additionally, given the concerns over the relationship between pregnancy and attrition, the current paper should be considered as the upper bound of the effects of teenage pregnancy on schooling and labor-force participation in South Africa. Another important consideration in this study is the overall demographic characteristics of individuals who drop from the sample. The attrition rates differ significantly by race. The black population attrition rate is 20%. Marteleto et al. (2008) explain that this attrition level is due to return migration to the rural Eastern Cape province. The colored popula- tion attrition level is 10%. Finally, attrition is higher among older cohort groups, which is indicative of a positive association between age and attrition in the CAPS data. Panel Dataset As specified in the data section, the panel data set is constructed using the women included in the static analysis. However, because of the data structure and the rate of attrition, the panel is not balanced. The sample is missing 12.86% of the observations it would have included. First, as seen in Table B1 in the appendix, the final wave of the CAPS data set does not ask any information regarding pregnancy in the year 2008, so I exclude that year from the sample. The fact affects the youngest cohort only. Additionally, the data on attrition is the largest in the last round of the data, so the schooling data for 2007 is limited. Only 77.8% of the sampled women provide data for that year. A further point that limits the balance of 35 the table is the lack of availability of information for a small number of years and sampled women. For example, the missing schooling data affects 1.11% of the year-women sample in 1999, 2% in 2000, and 0.4% in 2003. 4.3 Sibling differences Approach The second empirical approach utilized in this study to reduce selection bias by comparing outcomes among family members. The key assumption is that there are factors that are shared by members of the same family but are unobservable to the researcher. The inclusion of sibling fixed effects thus allows me to control for common socioeconomic factors (such as genetics, school quality, and economic resources). I then test whether women who share the same background but have different pregnancy outcomes perform differently in their educational paths. If this variation is conditionally independent of unmeasured within- household differences, this would also affect the outcomes, and thus the estimate is unbiased. 4.3.1 Panel Data Estimation Equation 7 describes the econometric specification for the panel dataset: Outcomeicht = ϕ+ ϕ2Pregnanticht + ϕ3Xi + ψh + ςc + εt + υitjh (7) In equation 7, Outcomeicht represents the outcome of the sampled woman i of cohort c, who lives in household h and is observed in year t. Also, Pregnanticht indicates whether that same woman got pregnant in year t. Xi is a set of individual-level adult health controls including individual i ’s height and normalized grade on the literacy exam. Additionally, ψh is the sibling fixed effect, ςj is the birth year fixed effect and εt are time fixed effects. Finally, υi is the error term. In Equation 7, ϕ2 represents the coefficient of 36 educated households. Hence, the selection may explain smaller effects. Second, it is also worth noting that most of the non-teenage mothers in the sample were sharing a household at the time of the pregnancy or birth. Educational attainment might thus be hindered by the presence of the young child in the home. In this case, the analysis of the impact of pregnancy would find smaller effects than the one found on the full sample. 5 Educational Results 5.1 Instrumental Variable Approach 5.1.1 Schooling Progression Table 3 reports the effects of teenage pregnancy on schooling progression. The top panel displays instrumental variable estimates for the effect of early pregnancy on the likelihood of failing a grade, the age-for-grade measure and the dropout risk; the bottom panel shows the reduced-form estimation.38 In the first place, a reported teenage pregnancy increases the probability of grade failure by 55.7 (0.028 se) percentage points and of lagging behind their non-pregnant teens by 0.284 years (0.010 se). Both coefficients are statistically significant at the one-percent level. The dropout risk is equal to a 0.097% additional likelihood of grade failure.39 The reduced-form estimation of the association between teenage fertility and the grade failure is equal to 55.7 (0.028 se). The age-for-grade measure is estimated to be 0.315 (0.009 38In Table A8 in the appendix, I extend the sample until the age of 24. 39Table A6 in the appendix contains the OLS results for the same variables. A teenage pregnancy, in this specification, is associated with an additional probability of grade failure by 25 percentage points. Next, the age for grade is 0.09. Both coefficients are significant at the one-percent level. Although small, the coefficient represents a 10% lag in the schooling progression of teenage mothers. 39 se). Finally, the dropout risk coefficient is equal to 0.423 (0.201 se). All coefficients are statistically significant at the one-percent level. 5.1.2 Schooling Attainment In Table 3, I present the effects of reporting a pregnancy before the age of 18 on the school attainment of the sampled women using a static instrumental variables identification strat- egy. Teenage pregnancy decreases the number of years of completed schooling by 1.8 years (p<0.05) in the instrumental-variable specifications. Teenage mothers are not less likely to sit for the matriculation exam but are not reporting that they continue to post-secondary school education, as the coefficients are 0.040 and -0.259, respectively. None of these coeffi- cients is statistically significant.40 The bottom panel of Table 3 presents the reduced form estimation coefficients. The estimates are equal to 0.057, 0.001, and -0.008 for the years of completed years of education, taking the matric exam, and the post secondary education variables respectively. However, only the first coefficient is statistically significant. Across the paper, I have presented the estimates of women who report a pregnancy before the age of 18. However, the CAPS data set allows me to identify which pregnancy resulted in a birth. Hence, in Appendix Table A11 I present the estimation of the effects of teenage birth on the schooling attainment of the sampled women. The results are consistent with those found in Table 3. 40Table A5 in the appendix contains the OLS results for the educational attainment variables. Teenage pregnancy decreases the number of years of completed schooling by 1.05, likelihood of sitting for the matric- ulation exam decreases the number by 0.233 percentage points, and postsecondary education decreases it by 4.3 points. 40 5.2 Sibling Comparison 5.2.1 Schooling Progression The effects of self-reported teenage pregnancy between sisters are described in Table 4. The report of a teenage pregnancy increases the probability of grade failure by 12.8 percentage points in the OLS estimation and 13.2 percentage points in the siblings approach. Both co- efficients are significant at the one-percent level (p<0.01). Teenage pregnancy also increases the age-for-grade measure by approximately half a year in both specifications: 0.501 and 0.521 (p<0.01), in the OLS and sister comparisons approaches, respectively. Next, among sisters in the panel, teenage pregnancy increases the likelihood of dropping out by 0.583 percentage points (p<0.01). The effects decreases to 0.494 (p <0.01) in the sisters approach. 5.2.2 Schooling Attainment As shown in Table 4, teenage pregnancy decreases completed schooling by 0.78 years in the OLS model and 0.79 years in the instrumental-variable specification. Both coefficients are significant at the one-percent level. Additionally, teenage mothers are not less likely to sit for the matriculation exam but do not report that they continue to post secondary school education, as the coefficients are 0.211 and -0.100, respectively. 6 Labor Force Participation I now present the effects of teenage pregnancy on the labor force participation of the full and the sister samples. The coefficients presented in Table 5 represent the extensive margin in the labor market at ages 19, 20, 21 and 22.41 41Table A13 in the appendix, shows the results for the intensive margins 41 percent level). The interaction coefficient, although positive and equal to 0.017 (0.073 se), is not statistically significant. The coefficients for teenage pregnancy, living grandmother, and the interaction are -0.165 (0.142 se), 0.023 (0.028 se) and -0.015 (0.054 se), respectively.44 7.2 Schools with High Levels of Grade Repetition To examine whether there is any reduction in the stigma in an environment where grade repetition is high, I take advantage of the school data available in wave 1 of the CAPS data set for all of the young adults, men and women.45 For each reported institution, I estimate the percentage of young adults who reported failing a grade in 2002. I then create an indicator variable that takes the value of one if the sampled woman attended a school with high grade-repetition rate and zero otherwise. The coefficients for these regressions are presented in Table 6.46 In column (3), I present the impact of teenage pregnancy on the total completed years of education, including the interaction between teenage pregnancy and attendance at a school with a high rate of grade repetition. The coefficient for teenage pregnancy is 1.889 fewer years of completed education (0.095 se, p<0.05). Attending a school with a high level of grade repetition leads to an increase of 0.143 (0.104 se) years of education. However, the interaction coefficient is positive 0.410 (0.225 se, p<0.1). The coefficients for the regression for when the left-hand side variable is an indicator vari- able for whether the individual sat for the matriculation exam are 0.123 (0.333 se) for teenage pregnancy and 0.110 (0.040 se, p<0.1) for attendance at a school with high grade repetition. The interaction estimated coefficient is -0.006 (0.052). As seen in column (6), when the 44The results for the presence of the grandfather can be found in found in Appendix Table A14. As shown, there are no statistically significant effects when the father is alive. 45They reported the last institution they have attended. 46The correlation between attending schools with high grade repetition and teenage pregnancy in the sample is -0.0415. 44 outcome variable is post secondary education, the coefficients for teenage pregnancy, living grandmother, and the interaction are all negative: 0.255 (0.216 se), 0.004 (0.029 se), and 0.016 (0.029 se), respectively. Neither coefficient is statistically significant. 8 Conclusion Evidence of the causal effects of teen pregnancy is scarce in low-income countries, and partic- ularly in sub-Saharan Africa. The paper provides empirical evidence by addressing whether early pregnancy causally affects the education and labor outcomes of young women in Cape Town, South Africa. Estimating the economic consequences of teen pregnancy is difficult because it requires disentangling the causal effects of the confounding selection into early motherhood. Using a rich data set of South Africa, I explore the issue using two identification strategies: instrumenting teenage pregnancy using teenage fertility and a sibling comparison. My findings suggest that teen pregnancy decreases the pace at which girls progress in their schooling. In particular, I find that teenage mothers are more likely to fail a grade, lag behind in their education, and drop out of school. In the static model, my findings indicate that women who report a pregnancy are less educated and less likely to continue to post secondary education. The results are mostly consistent between identification strategies. Taken together, the results contribute to existing evidence that delayed pregnancy leads to beneficial outcomes for teenage girls in the short run. Nevertheless, I find two factors that attenuate the negative effects of teenage pregnancy. First, the presence of the teen’s mother during the teenage years attenuates the negative effects of early pregnancy on schooling. Second, attending a school with above-average failure rates also mitigates the lag in education effect. Estimating these specific attenuating characteristics is an important step towards designing policies can be successfully assist 45 women continue their education. In South Africa, where a third of the women report a birth by the age of 19, the evidence presented in this paper suggests that the timing of the first birth is important for the edu- cational outcomes of young women. This paper speaks to the debate on reproductive health policies in low-income countries (Herrera et al., 2019). My findings imply that postponing the age at first birth can have important consequences for women’s educational attainment and may delay entrance into the labor force or improve overall well-being (Ardington et al. (2014) and Urdinola and Ospino, 2015). The mitigating factors found in this paper would also inform these policies. Additionally, policies that aim at assisting teenage mothers should focus on the factors that can alleviate the duties related to childcare and the stigma attached to it.47 References Adair, L. S. (2001). Size at birth predicts age at menarche. Pediatrics, 107 (4), E59. Adebara, I. & Munir’deen, A. I. (2012). Relationship of biosocial factors and hemoglobin genotype with age of menarche among school girls in ilorin, nigeria. Nepal Journal of Obstetrics and Gynaecology, 7 (1), 19–24. Akresh, R., Bhalotra, S., Leone, M., & Osili, U. O. (2012). War and stature: Growing up during the nigerian civil war. American Economic Review, 102 (3), 273–77. Alderman, H., Hoddinott, J., & Kinsey, B. (2006). Long term consequences of early childhood malnutrition. Oxford economic papers, 58 (3), 450–474. Almanza, C. H. & Sahn, D. E. (2018). Early childbearing, school attainment, and cognitive skills: Evidence from madagascar. Demography, 55 (2), 643–668. Almond, D. (2006). Is the 1918 influenza pandemic over? long-term effects of in utero in- fluenza exposure in the post-1940 us population. Journal of political Economy, 114 (4), 672–712. Almond, D., Chay, K. Y., & Lee, D. S. (2005). The costs of low birth weight. The Quarterly Journal of Economics, 120 (3), 1031–1083. 47In future research, I will study the overall fertility decisions of South African women. A secondary result, presented in Table A15 in the appendix, suggests that women who report a teenage pregnancy are not more likely to have more children and are more likely to report a smaller subsequent fertility. These results hint at a degree of substitution between teenage fertility and subsequent fertility that should be explored in more detail 46 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 (3), 683–715. Jacobs, W. & Du Plessis, D. J. (2016). A spatial perspective of the patterns and charac- teristics of main-and substream migration to the western cape, south africa. In Urban forum (Vol. 27, 2, pp. 167–185). Springer. Jahanfar, S., Lye, M.-S., & Krishnarajah, I. S. (2013). Genetic and environmental effects on age at menarche, and its relationship with reproductive health in twins. Indian journal of human genetics, 19 (2), 245. Karapanou, O. & Papadimitriou, A. (2010). Determinants of menarche. Reproductive Biology and Endocrinology, 8 (1), 115. Kaufman, C. E., De Wet, T., & Stadler, J. (2001). Adolescent pregnancy and parenthood in south africa. Studies in family planning, 32 (2), 147–160. Khanna, M. (2019). The precocious period: The impact of early menarche on schooling in india. Available at SSRN 3419041. Kim, Y. & Je, Y. (2019). Early menarche and risk of metabolic syndrome: A systematic review and meta-analysis. Journal of Women’s Health, 28 (1), 77–86. Klepinger, D. H., Lundberg, S., Plotnick, R., et al. (1997). How does adolescent fertility affect the human capital and wages of young women? Institute for Research on Poverty, University of Wisconsin. Lam, D., Ardington, C., & Leibbrandt, M. (2011). Schooling as a lottery: Racial differences in school advancement in urban south africa. Journal of development economics, 95 (2), 121–136. Lam, D., Marteleto, L. J., & Ranchhod, V. (2013). The influence of older classmates on adolescent sexual behavior in cape town, south africa. Studies in family planning, 44 (2), 147–167. Lam, D., Seekings, J., & Sparks, M. (2006). The cape area panel study (caps): Overview and technical documentation for waves 1–2–3. Cape Town, South Africa: CAPS, University of Cape Town. Lancaster, J. B. & Hamburg, B. A. (1986). School-age pregnancy and parenthood: Bisocial dimensions. Transaction Publishers. Lawn, R. B., Lawlor, D. A., & Fraser, A. (2017). Associations between maternal prepregnancy body mass index and gestational weight gain and daughter’s age at menarche: The avon longitudinal study of parents and children. American journal of epidemiology, 187 (4), 677–686. Lee, D. (2010). The early socioeconomic effects of teenage childbearing: A propensity score matching approach. Demographic Research, 23, 697–736. Levine, D. I. & Painter, G. (2003). The schooling costs of teenage out-of-wedlock child- bearing: Analysis with a within-school propensity-score-matching estimator. Review of Economics and Statistics, 85 (4), 884–900. 49 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), 1– 13. Madhavan, S. & Thomas, K. J. (2005). Childbearing and schooling: New evidence from south africa. Comparative Education Review, 49 (4), 452–467. Magadla, S. S., Leibbrandt, M., & Mlatsheni, C. (2019). Does a motherhood penalty exist in the post-apartheid south african labour market? Magruder, J. R. (2010). Intergenerational networks, unemployment, and persistent inequality in south africa. American Economic Journal: Applied Economics, 2 (1), 62–85. Mahmud, S., Shah, N. M., & Becker, S. (2012). Measurement of women’s empowerment in rural bangladesh. World development, 40 (3), 610–619. Marteleto, L., Lam, D., & Ranchhod, V. (2008). Sexual behavior, pregnancy, and schooling among young people in urban south africa. Studies in family planning, 39 (4), 351–368. Martorell, R. & Habicht, J.-P. (1986). Growth in early childhood in developing countries. Monteiro, D. L. M., Martins, J. A. F. d. S., Rodrigues, N. C. P., Miranda, F. R. D. d., Lacerda, I. M. S., Souza, F. M. d., . . . Trajano, A. J. B. (2019). Adolescent pregnancy trends in the last decade. Revista da Associação Médica Brasileira, 65 (9), 1209–1215. Moultrie, T. A. & Dorrington, R. (2004). Estimation of fertility from the 2001 south africa census data. Centre for Actuarial Research, University of Cape Town Cape Town. Mpora, B. O., Piloya, T., Awor, S., Ngwiri, T., Laigong, P., Mworozi, E. A., & Hochberg, Z. (2014). Age at menarche in relation to nutritional status and critical life events among rural and urban secondary school girls in post-conflict northern uganda. BMC women’s health, 14 (1), 66. Narita, R. & Diaz, M. D. M. (2016). Teenage motherhood, education, and labor market outcomes of the mother: Evidence from brazilian data. Economia, 17 (2), 238–252. Nattrass, N. & Walker, R. (2005). Unemployment and reservation wages in working-class cape town. South African Journal of Economics, 73 (3), 498–509. Padez, C. (2003). Age at menarche of schoolgirls in maputo, mozambique. Annals of human biology, 30 (4), 487–495. Panday, S., Makiwane, M., Ranchod, C., & Letsoala, T. (2009). Teenage pregnancy in south africa: With a specific focus on school-going learners. Posel, D. (2004). Have migration patterns in post-apartheid south africa changed? Journal of Interdisciplinary Economics, 15 (3-4), 277–292. Prentice, S., Fulford, A. J., Jarjou, L. M. A., Goldberg, G. R., & Prentice, A. (2010). Evidence for a downward secular trend in age of menarche in a rural gambian population. Annals of Human Biology, 37 (5), 717–721. PMID: 20465526. doi:10.3109/03014461003727606. eprint: https://doi.org/10.3109/03014461003727606 Rah, J. H., Shamim, A. A., Arju, U. T., Labrique, A. B., Rashid, M., & Christian, P. (2009). Age of onset, nutritional determinants, and seasonal variations in menarche in rural bangladesh. Journal of health, population, and nutrition, 27 (6), 802. 50 Ranchhod, V., Lam, D., Leibbrandt, M., & Marteleto, L. (2011). Estimating the effect of adolescent fertility on educational attainment in cape town using a propensity score weighted regression. Ribar, D. (1994). Teenage fertility and high school completion. The Review of Economics and Statistics, 76(3):413–424. Ribar, D. C. (1999). The socioeconomic consequences of young women’s childbearing: Rec- onciling disparate evidence. Journal of Population Economics, 12 (4), 547–565. Schultz, T. P. & Mwabu, G. (1998). Labor unions and the distribution of wages and employ- ment in south africa. ILR Review, 51 (4), 680–703. Sedgh, G., Finer, L. B., Bankole, A., Eilers, M. A., & Singh, S. (2015). Adolescent preg- nancy, birth, and abortion rates across countries: Levels and recent trends. Journal of Adolescent Health, 56 (2), 223–230. Sekhri, S. & Debnath, S. (2014). Intergenerational consequences of early age marriages of girls: Effect on children’s human capital. The Journal of Development Studies, 50 (12), 1670–1686. Silventoinen, K. (2003). Determinants of variation in adult body height. Journal of biosocial science, 35 (2), 263–285. Singh, S., Wulf, D., Samara, R., & Cuca, Y. P. (2000). Gender differences in the timing of first intercourse: Data from 14 countries. International Family Planning Perspectives, 26 (1), 21–28. Sørensen, K., Juul, A., Christensen, K., Skytthe, A., Scheike, T., & Kold Jensen, T. (2013). Birth size and age at menarche: A twin perspective. Human Reproduction, 28 (10), 2865–2871. Sunder, N. (2018). Marriage age, social status and intergenerational effects in uganda. Social Status and Intergenerational Effects in Uganda (March 1, 2018). Unicef. (2014). Ending child marriage: Progress and prospects. UNICEF. Urdinola, B. P. & Ospino, C. (2015). Long-term consequences of adolescent fertility: The colombian case. Demographic Research, 32, 1487–1518. Van der Berg, S., Louw, M. et al. (2007). Lessons learnt from sacmeqii: South african student performance in regional context. University of Stellenbosch, Department of Economics and Bureau for Economic Research Working Paper, 16 (07). WHO. (2018). World health organization. https://www.who.int/news-room/fact-sheets/ detail/adolescent-pregnancy. Wooldridge, J. M. (2015). Control function methods in applied econometrics. Journal of Human Resources, 50 (2), 420–445. World Health Organization. (2014). Trends in maternal mortality: 1990 to 2013: Estimates by who, unicef, unfpa, the world bank and the united nations population division. Zito, R. C. (2018). Children as saviors? a propensity score analysis of the impact of teenage motherhood on personal transformation. Youth & Society, 50 (8), 1100–1122. 51 Figure 4: Completed Years of Education Source: waves 1-5 of the CAPS data set. The sample includes women who reached menarche between the ages of 10 and 17 with complete information ob all the outcomes and the control variables. 54 Figure 5: First Stage: Teenage Pregnancy and Fertility Panel A: Relationship between an indicator for teenage fertility and pregnancy in the panel dataset. Panel B: Relationship between the number of Fertile years and Teenage Pregnancy. Source: CAPS waves 1-5. The sample includes women who reached menarche between the ages of 10 and 17. 55 Figure 6: Adult Height and Age at Menarche Source: CAPS wave 4. The sample includes women who reached menarche between the ages of 10 and 17 with complete outcomes and health variables. Figure 7: Height by Percentage of the District’s Population below the poverty line Observations=1,741. Source: CAPS waves 1 and 4. The sample includes women who reached menarche between the ages of 10 and 17 with complete outcomes and health 56 Table 1: Summary Statistics - by Sample Full Sample Sibling Sample Difference (1) (2) (3) Demographic Characteristics Age in 2002 17.723 17.902 0.179 (2.445) (2.556) % Coloured 0.461 0.409 -0.052* (0.499) (0.492) % Black 0.486 0.565 0.070*** (0.500) (0.496) Asset index -0.145 -0.560 -0.415*** (1.981) (2.012) Adult Height - cm 157.979 158.355 0.376 (8.131) (8.014) Mother’s Education 8.276 7.973 -0.303* (3.136) (2.906) Not Born in WC 0.283 0.378 0.095*** (0.451) (0.485) # Full Siblings 2.317 2.557 0.240** (1.756) (1.829) Sexual Activity Age 1st Partner 19.779 19.656 -0.123 (3.209) (3.347) Age 1st Active 17.392 17.016 -0.376*** (2.289) (2.036) Observations 1,741 418 Notes: “Adult Height”is measured in 2006. “% Coloured” and “% Black” report the percentage of the population who identify with that race (the third category is white). “Asset index” is pca index of the assets of women’s household. “ Adult Height - cm” is the women’s height measured in centimeters. “Mother’s Education” describe the to- tal number of years completed by mother and “Not Born in WC” is an indicator vari- able for individuals born outside the Western Cape. “# Full Siblings” refers to the number of siblings. Finally, “Age of 1st. Partner” and “Age 1st Active” describe the ages of the first sexual partner and the age in which women in the sample be- came active respectively. The Diff column is the difference in means of Columns (1) and (2), where a T-test where the hypothesis is that the coefficient is equal to zero. *** p≤0.01, ** p≤0.05, * p≤0.1 59 Table 2: First Stage Estimation Panel Analysis Collapsed Panel Pregnantit Pregnancy ≤ 18i (1) (2) (3) (4) (5) (6) Fertilet 0.023*** 0.024*** (0.002) (0.002) Fertile yearsi 0.016*** 0.032*** (0.006) (0.008) Menarche ≤ 14i 0.054*** 0.090*** (0.018) (0.022) Observations 15,170 15,170 1,741 1,741 1,741 1,741 R-squared 0.009 0.010 0.004 0.011 0.005 0.011 First-stage F 202 171.2 6.914 17.61 8.935 17.09 Sampling Location FE No Yes No Yes No Yes Controls No Yes No Yes No Yes Time FE No Yes Notes: “Pregnanticst” is defined 1 for individual i of cohort c, who lives in sampling cluster s and observed in year t if she got pregnant and zero; if she did not get pregnant during that year. In the same Data set, “Fertilet” is a dichotomous variable equal to zero if the sample individual had not reached menarche in year t and equal to one if she had in that same year. “Pregnancy ≤ 18” in the collapsed panel Data-set is defined one for those who got pregnant before 18 and zero for those who did not. Also, the “Number of Fertile Years” indicates the number of years passed since menarche until 17. Controls include the race, an asset index, height, number of siblings, the log of the household level and indicator variables for religion, language spoken by her family, place of birth and residence. Cohort and sampling-location fixed effects are included. In the panel data set, Pregnant in year t is an indicator variable for whether the individual got pregnant in year t and Fertile is an indicator variable of whether a woman was fertile in a specific year. In both data sets, standard errors in parentheses, clustered at the sampling location level.*** p ≤ 0.01, ** p ≤ 0.05, * p≤0.1 60 T ab le 3: E st im at io n of th e E ff ec ts of T ee n ag e P re gn an cy - S ch o ol in g P ro gr es si on P a n e l C o ll a p se d P a n e l H az ar d Y ea rs of T o ok P os t S ec on d ar y F ai le d gr ad e A ge fo r G ra d e D ro p O u t E d u ca ti on M at ri c E x am S ch o ol in g (1 ) (2 ) (3 ) (4 ) (5 ) (6 ) P a n e l A : IV E st im a ti o n P re gn an t i cs t 0. 55 7* ** 0. 28 4* ** 0. 09 7* ** (0 .0 28 ) (0 .0 10 ) (0 .0 09 ) P re gn an cy ≤ 18 i -1 .8 20 ** 0. 04 4 -0 .2 59 (0 .9 22 ) (0 .2 89 ) (0 .1 99 ) O b se rv at io n s 15 ,1 70 15 ,1 70 14 ,3 54 1, 74 1 1, 74 1 1, 74 1 R -s q u ar ed 0. 01 0 0. 02 3 0. 17 2 0. 20 8 0. 01 9 F ir st st ag e- F -s ta t 17 1. 2 17 1. 2 17 1. 2 17 .6 1 17 .6 1 17 .6 1 P a n e l B : R e d u ce d F o rm E st im a ti o n P os t M en ar ch e i tc s 0. 61 7* ** 0. 31 5* ** 0. 42 3* (0 .0 33 ) (0 .0 09 ) (0 .2 01 ) F er ti le Y ea rs i -0 .0 57 * 0. 00 1 -0 .0 08 (0 .0 30 ) (0 .0 09 ) (0 .0 06 ) O b se rv at io n s 15 ,1 70 15 ,1 70 14 ,3 54 1, 74 1 1, 74 1 1, 74 1 R -s q u ar ed 0. 01 0 0. 02 3 0. 46 2 0. 45 7 0. 40 5 M ea n d ep en d en t va r. 0. 11 3 1. 09 1 0. 35 9 11 .0 5 0. 43 6 0. 14 6 N ot es : “P re gn an t i cs t” is eq u al to 1 fo r in d iv id u al i of co h or t c, w h o li ve s in sa m p li n g cl u st er s an d re p or te d a p re gn an cy in ye ar t an d 0 ot h er w is e. S im il ar ly , “P re gn an t≤ 18 i” ta ke s th e va lu e of 1 if w om en i re p or te d a p re gn an cy b ef or e th e ag e of 18 an d 0 ot h er w is e. C ol u m n s (1 ) - (6 ) co n ta in th e IV an d re d u ce d es ti m at es of th e eff ec ts on th e li ke li h o o d of fa il in g a gr ad e in ye ar t an d th e d eg re e to w h ic h a sa m p le d w om en la gg ed b eh in d h er se co n d ar y ed u ca ti on in th at sa m e ye ar . In p an el A , en d og en ei ty is ac co u n te d fo r u si n g a li n ea r fi rs t st ag e of th e eff ec ts of te en ag e fe rt il it y on p re gn an cy . T h e F ir st S ta ge st at is ti c co m es fr om th e a fi rs t st ag e re gr es si on w h er e p re gn an cy is in st ru m en te d u si n g ei th er a p os t m en ar ch e in d ic at or or th e n u m b er of fe rt il e ye ar s. In th e th e st at ic an al y si s, co n tr ol s in cl u d e th e ra ce , an as se t in d ex , h ei gh t, n u m b er of si b li n gs , an d th e in d ic at or va ri ab le s fo r re li gi on , la n gu ag e sp ok en b y h er fa m il y. S ta n d ar d er ro rs in p ar en th es es , cl u st er ed at th e sm al le st sa m p li n g lo ca ti on le ve l. ** * p ≤ 0. 01 , ** p ≤ 0. 05 , * p ≤ 0. 1 61 T ab le 6: S ib li n g D iff er en ce s E st im at io n of th e E ff ec ts of T ee n ag e P re gn an cy Y ea rs of T o ok P os t S ec on d ar y Y ea rs of T o ok P os t S ec on d ar y E d u ca ti on M at ri c E x am S ch o ol in g E d u ca ti on M at ri c E x am S ch o ol in g (1 ) (2 ) (3 ) (4 ) (5 ) (6 ) P re gn an cy ≤ 18 -2 .1 38 ** * -0 .0 59 -0 .1 65 -1 .8 89 ** 0. 12 3 -0 .2 55 (0 .6 92 ) (0 .2 11 ) (0 .1 42 ) (0 .9 55 ) (0 .3 33 ) (0 .2 16 ) G ra n d m ot h er 0. 28 5* 0. 12 4* * 0. 02 3 (0 .1 57 ) (0 .0 57 ) (0 .0 28 ) P re gn an cy ≤ 18 x G ra n d m ot h er 0. 52 1* 0. 01 7 -0 .0 15 (0 .3 16 ) (0 .0 73 ) (0 .0 54 ) H ig h F ai lu re S ch o ol 0. 14 3 0. 11 0* ** -0 .0 04 (0 .1 04 ) (0 .0 40 ) (0 .0 29 ) P re gn an cy ≤ 18 x H ig h F ai lu re S ch o ol 0. 41 0* -0 .0 06 -0 .0 16 (0 .2 25 ) (0 .0 52 ) (0 .0 29 ) O b se rv at io n s 1, 74 1 1, 74 1 1, 74 1 1, 74 1 1, 74 1 1, 74 1 R -s q u ar ed 0. 16 9 0. 17 5 0. 05 8 0. 18 4 0. 17 7 0. 05 7 F ir st S ta ge F -S ta ti st ic 17 .6 7 17 .6 7 17 .6 7 16 .7 6 16 .7 6 16 .7 6 M ea n D ep en d en t V ar 10 .9 4 0. 52 4 0. 14 7 10 .9 8 0. 52 4 0. 15 7 N o te s: P re gn an cy ≤ 18 is d efi n ed on e fo r th os e w h o g o t p re g n a n t b ef o re 1 8 a n d ze ro fo r th o se w h o d id n o t. ” G ra n d m o th er ” is a n in d ic a to r va ri a b le fo r w h et h er th e te en ag er ’s m ot h er w as al iv e in h er te en ye a rs , a n d ze ro o th er w is e. “ H ig h F a il u re S ch o o l” is ea iq l T h e va ri a b le w a s co n st ru ct ed u si n g th e C A P S d a ta se t. “Y ea rs of E d u ca ti on ” in d ic at es th e Y ea rs of ed u ca ti o n co m p le te d . “ T o o k M a tr ic ” is a n in d ic a to r va ri a b le o f w h et h er th e in d iv id u a l to o k th e M a tr ic ex a m a n d “P os t S ec on d ar y E d u ca ti on ” is eq u al to on e if th e sa m p le d w o m a n ke p t st u d y in g a ft er h ig h sc h o o l, a n d ze ro o th er w is e. T h e “ F ir st S ta g e F -s ta t” co m es fr o m th e a fi rs t st ag e re gr es si on w h er e “P re gn an cy ¡ 1 8 ” is in st ru m en te d w it h th e n u m b er o f te en a g e fe rt il e ye a rs . T h e M ea n D ep en d en t V a r in d ic a te s th e av er a g e o f th e O u tc om e if “P re gn an cy ¡ 18 ” is eq u al to ze ro . C o n tr o ls in cl u d e th e ra ce , a n a ss et in d ex , h ei g h t, n u m b er o f si b li n g s, a n d th e in d ic a to r va ri a b le s fo r re li g io n , la n gu ag e sp ok en b y h er fa m il y. C oh or t an d sa m p li n g -l o ca ti o n fi x ed eff ec ts a re in cl u d ed . S ta n d a rd er ro rs in p a re n th es es , cl u st er ed a t th e sa m p li n g lo ca ti o n le ve l. ** * p ≤ 0. 01 , ** p ≤ 0. 05 , * p ≤ 0. 1 64 A Appendix Tables Table A1: Samples of Women across waves 2002 2003-2004 2005 2006 2009-2010 Wave 1 Wave 2 Wave 3 Wave 4 Wave 5 (1) (2) (3) (4) (5) Panel A: All Women Interviewed Per Wave Observations 2,612 2,140 1,911 1,881 1,799 Percentage 100% 81.93% 73.16% 72.01% 68.87% Panel B: Analytical Sample Observations 2,455 2,020 1,815 1,741 1,707 Percentage 100% 82.28% 73.93% 70.96% 69.53% Source: CAPS Data waves 1-5 Notes: Panel A includes all sampled women interviewed in each wave of the CAPS data set. Panel B includes sampled women that where interviewed and also had reached menarche between the ages of 10 and 17 and lived in sampling clusters with at least one other sampled women. 65 Table A2: Summary Statistics by Self-reported Early Pregnancy No Early Pregnancy Pregnancy≤18 Difference (1) (2) (3) Demographic Characteristics Age in 2002 17.689 17.865 0.176 (2.445) (2.448) % Coloured 0.438 0.555 0.117*** (0.496) (0.498) % Black 0.497 0.443 -0.055* (0.500) (0.497) Adult Height 158.271 156.808 -1.463*** (7.987) (8.597) Mother’s Education 8.447 7.559 -0.888*** (3.191) (2.785) # Full Siblings 2.286 2.416 0.162 (1.739) (1.813) Household Size 5.691 6.256 0.565*** (2.528) (2.755) Not Born in WC 0.288 0.264 -0.024 (0.453) (0.442) Sexual Activity Age 1st Partner 19.923 19.378 -0.546*** (3.128) (3.395) Age 1st Active 17.811 15.928 -1.883*** (2.326) (1.374) Observations 1394 348 Notes: Pregnancy early is equal to 1 if the individual reported a pregnancy before the age of 17, and 0 otherwise. “Adult Height”is the height of the sampled women, measured in 2006. “Not Born in WC” is an indicator variable for individuals born outside the Western Cape, and “Mother’s Education” is the total number of years completed by mother. “# Full Siblings” is number of siblings. Finally “Age of 1st. Partner” and “Age 1st Active” describe the ages of the first sexual partner and the age in which women in the sample became active respectively. The Diff column is the difference in means of Columns (1) and (2), where a T-test where the hypothesis is that the coefficient is equal to zero. *** p≤0.01, ** p≤0.05, * p≤0.1 66 Table A5: OLS Estimation of the Effects of Teenage Pregnancy Panel A: School Progression Hazard Failed grade Age for Grade Drop Out (1) (2) (3) Pregnanticst 0.242*** 0.687*** 0.017*** (0.031) (0.051) (0.001) Observations 1,741 1,741 1,741 R-squared 0.160 0.178 Panel B: Schooling Attainment Years of Took Post Secondary Education Matric Exam Schooling (1) (2) (3) Pregnancy ≤ 18 -1.034*** -0.233*** -0.043*** (0.115) (0.026) (0.014) Observations 15,170 15,170 14,354 R-squared 0.462 0.457 0.405 Panel C: Labor Force Participation Active at age 19 age 20 age 21 age 22 (1) (2) (3) (4) Pregnancy ≤ 18 0.115*** 0.059* 0.048 0.078*** (0.030) (0.033) (0.033) (0.029) Observations 1,741 1,741 1,741 1,741 R-squared 0.071 0.037 0.061 0.083 Notes: Birth ≤ 18 is defined one for those who gave before 18 and zero for those who did not. Controls include the race, an asset index, height, number of siblings, and the indicator variables for religion, language spoken by her family. “Pregnanticst” is defined as 1 for individual i of cohort c, who lives in sampling cluster s and observed in year t if she got pregnant and zero; if she did not get pregnant during that year.“Years of Education” indicates the Years of education completed. “took Matric” is an indicator variable of whether the individual took the Matric exam and “Post Secondary Education” is equal to one if the sampled woman kept studying after high school, and zero otherwise. Active at each specific age is an indicator variable for whether the sampled women reported being in the labor force at that age. 69 Table A6: First stage - Different Thresholds for Early Pregnancy Pregnancy≤16 Pregnancy≤17 Pregnancy≤18 (1) (2) (3) Fertile Years 0.023*** 0.031*** 0.032*** (0.005) (0.006) (0.008) Observations 1,741 1,741 1,741 R-squared 0.014 0.016 0.009 First stage F 20.67 25.53 17.61 Notes: “Pregnancy ≤ 16”, “Pregnancy ≤ 17” and “Pregnancy ≤ 18” are defined one for those who got pregnant before the indicated age and zero for those who did not. Also, the “Number of Fertile Years” indicates the number of years passed since menarche until 17. Controls include the race, an asset index, height, number of siblings, the log of the household level and indicator variables for religion, language spoken by her family, place of birth and residence. Cohort and sampling-location fixed effects are included. In the panel data set, Pregnant in year t is an indicator variable for whether the individual got pregnant in year t and Fertile is an indicator variable of whether a woman was fertile in a specific year. In both data sets, standard errors in parentheses, clustered at the sampling location level.*** p ≤ 0.01, ** p ≤ 0.05, * p≤0.1 70 Table A7: First stage - First stage with different top values in the instrument Pregnancy≤18 (1) (2) (3) (4) (5) (6) Fertile Years18 0.016** 0.031*** (0.006) (0.008) Fertile Years19 0.016** 0.031*** (0.006) (0.008) Fertile Years20 0.016** 0.031*** (0.006) (0.008) Observations 1,741 1,741 1,741 1,741 1,741 1,741 R-squared 0.003 0.011 0.003 0.011 0.003 0.011 First stage F 6.626 17.57 6.626 17.57 6.626 17.57 Notes: “Pregnancy ≤ 18” is defined one for those who got pregnant before 18 and zero for those who did not. Also, the “Fertile Years” indicates the number of years passed since menarche until 18, 19 and 20 in each Row. Controls include the race, an asset index, height, number of siblings, the log of the household level and indicator variables for religion, language spoken by her family, place of birth and residence. Cohort and sampling-location fixed effects are included. In the panel data set, Pregnant in year t is an indicator variable for whether the individual got pregnant in year t and Fertile is an indicator variable of whether a woman was fertile in a specific year. In both data sets, standard errors in parentheses, clustered at the sampling location level.*** p ≤ 0.01, ** p ≤ 0.05, * p≤0.1 71 Table A10: Inverse Probability Weights Estimation Panel A: School Attainment Years Education Took Matric Post Secondary Educ. (1) (2) (3) Pregnancy ≤18i -1.275** -0.096 -0.122 (0.629) (0.168) (0.106) Observations 1,735 1,735 1,735 Mean Dependent Var 10.94 0.524 0.147 First Stage F-stat 22.85 22.85 22.85 Panel B: Labor force participation At age 19 At age 20 At age 21 At age 22 (1) (2) (3) Pregnancy ≤18i 0.176 0.314* 0.042 0.074 (0.188) (0.173) (0.175) (0.169) Observations 1,735 1,735 1,735 1,735 First Stage F-stat 22.85 22.85 22.85 22.85 Mean Dependent Var 0.566 0.461 0.421 0.394 Notes: Pregnancy ≤ 18 is defined one for those who got pregnant before 18 and zero for those who did not. Controls include height and cohort fixed effects. “Years of Education” indicates the Years of education completed. “Took Matric” is an indicator variable of whether the individual sat for Matric exam and “Post Secondary Education” is equal to one if the sampled woman kept studying after high school, and zero otherwise. The Mean Dependent Var indicates the average of the Outcome if “Pregnancy ≤ 18” is equal to zero. Cohort and sampling-location fixed effects are included. *** p ≤ 0.01, ** p ≤ 0.05, * p≤0.1 74 Table A11: IV Estimation of the Effects of Teenage Birth - Schooling Attainment Years of Took Post Secondary Education Matric Exam Schooling (1) (2) (3) Panel A: OLS Estimation Birth ≤ 18i -1.087*** -0.224*** -0.041*** (0.117) (0.027) (0.014) Observations 1,741 1,741 1,741 R-squared 0.167 0.166 0.055 Panel B: IV Estimation Birth ≤ 18 -2.380** 0.056 -0.337 (1.196) (0.378) (0.267) Observations 1,741 1,741 1,741 R-squared 0.082 0.117 -0.092 First Stage F-stat 10.64 10.64 10.64 Mean Dependent Var 10.94 0.524 0.147 Notes: Birth ≤ 18 is defined one for those who gave before 18 and zero for those who did not. Controls include the race, an asset index, height, number of siblings, and the indicator variables for religion, language spoken by her family. “Years of Education” indicates the Years of education completed. “Took Matric” is an indicator variable of whether the individual sat for Matric exam and “Post Secondary Education” is equal to one if the sampled woman kept studying after high school, and zero otherwise. The “First Stage F-stat” comes from the a first stage regression where “Pregnancy ≤ 18” is instrumented with the number of teenage fertile years. The Mean Dependent Var indicates the average of the Outcome if “Pregnancy ≤ 18” is equal to zero. Cohort and sampling-location fixed effects are included. Standard errors in parentheses, clustered at the sampling location level. *** p ≤ 0.01, ** p ≤ 0.05, * p≤0.1 75 Table A12: Estimation of the Effects of Teenage Pregnancy - Schooling Attainment Inverse probability Weights Years of Took Post Secondary Education Matric Exam Schooling (1) (2) (3) Panel A: IV Estimation Pregnancy≤ 18i -1.275** -0.096 -0.122 (0.629) (0.168) (0.106) Observations 1,735 1,735 1,735 R-squared 0.162 0.169 0.048 First Stage F-stat 22.85 22.85 22.85 Panel B: Reduced Form Estimation Fertile Yearsi -0.075* -0.006 -0.007 (0.041) (0.011) (0.007) Observations 1,735 1,735 1,735 R-squared 0.429 0.395 0.396 Comparison Mean Var 10.98 0.524 0.157 Notes: Pregnancy ≤ 18 is defined one for those who got pregnant before 18 and zero for those who did not. The Number of Children born > the age 19 is equal to the number of pregnancies each women reported between the age of 19 and 2009. Additionally, the Total Number of Children indicates the number of children each women in the sample has has before 2009, regardless of the age she gave birth. Controls include the race, an asset index, height, number of siblings, and the indicator variables for religion, language spoken by her family. The “First Stage F-stat” comes from the a first stage regression where “Pregnancy ≤ 18” is instrumented with the number of teenage fertile years. The Mean Dependent Var indicates the average of the Outcome if “Pregnancy ≤ 18” is equal to zero. Cohort and sampling-location fixed effects are included. Standard errors in parentheses, clustered at the sampling location level. *** p ≤ 0.01, ** p ≤ 0.05, * p≤0.1 76 Table A15: IV Estimation of the Effects of Teenage Pregnancy - Marriage and Fertility Married Ever Subsequent Total Fertility Fertility (1) (2) (3) Panel A: IV Estimation Pregnancy≤ 18i 0.480 0.517 -0.744** (0.303) (0.360) (0.363) Observations 1,741 1,741 1,741 R-squared -0.014 0.252 0.197 First Stage F-stat 17.59 17.59 17.59 Panel B: Reduced Form Estimation Fertile Yearsi 0.015 0.016 -0.023* (0.009) (0.012) (0.013) Observations 1,741 1,741 1,741 R-squared 0.082 0.127 0.150 Mean Dependent Var 0.354 0.536 0.536 Notes: Pregnancy ≤ 18 is defined one for those who got pregnant before 18 and zero for those who did not. Subsequent Fertility is equal to the number of pregnancies each women reported between the age of 19 and 2009. Additionally, Total Fertility indicates the number of children each women in the sample has has before 2009. Controls include the race, an asset index, height, number of siblings, and the indicator variables for religion, language spoken by her family. Married ever is defined as one for those who got married at some point before 2009, and 0 otherwise. The “First Stage F-stat” comes from the a first stage regression where “Pregnancy ≤ 18” is instrumented with the number of teenage fertile years. The Mean Dependent Var indicates the average of the Outcome if “Pregnancy ≤ 18” is equal to zero. Cohort and sampling-location fixed effects are included. Standard errors in parentheses, clustered at the sampling location level. *** p ≤ 0.01, ** p ≤ 0.05, * p≤0.1 79 T ab le B 1: V ar ia b le C on st ru ct io n b y W av e 20 02 20 03 -2 00 4 20 05 20 06 20 09 -2 01 0 W av e 1 W av e 2 W av e 3 W av e 4 W av e 5 (1 ) (2 ) (3 ) (4 ) (5 ) H e a lt h , F e rt il it y & M a rr ia g e A ge at M en ar ch e - fo r th os e w /o d at a in w av e 1 - - P re gn an cy R et ro sp ec ti ve ye ar ly fo r 19 79 -2 00 2 - B ir th s R et ro sp ec ti ve ye ar ly fo r 19 79 -2 00 2 - M ar ri ag e R et ro sp ec ti ve ye ar ly fo r 19 79 -2 00 2 - R et ro sp ec ti ve ye ar ly fo r 20 03 -2 00 5 R et ro sp ec ti ve ye ar ly fo r 20 07 -2 00 9 A d u lt H ei gh t - - - - E d u ca ti o n L it er ac y E x am - - - - Y ea rs of E d u ca ti on R et ro sp ec ti ve ye ar ly fo r 19 79 -2 00 2 R et ro sp ec ti ve ye ar ly fo r 20 07 -2 00 9 G ra d e P ro gr es s R et ro sp ec ti ve ye ar ly fo r 19 79 -2 00 2 R et ro sp ec ti ve ye ar ly fo r 20 07 -2 00 9 M at ri cu la ti on R et ro sp ec ti ve ye ar ly fo r 19 79 -2 00 2 R et ro sp ec ti ve ye ar ly fo r 20 07 -2 00 9 E m p lo y m e n t E m p lo y m en t E m p lo y m en t C h ar ac t. C o n tr o l V a ri a b le s B ac k gr ou n d C h ar ac . - - - - C h il d h o o d In fo - - - - P ar en ts D em og ra p h ic s - - H ea lt h - P ar en ts D ea th R et ro sp ec ti ve ye ar ly fo r 19 79 -2 00 2 - R et ro sp ec ti ve ye ar ly fo r 20 03 -2 00 5 R et ro sp ec ti ve ye ar ly fo r 20 07 -2 00 9 H ou se h ol d C h ar ac . - - - - S ou rc e: C A P S D at a w av es 1- 5 T h e in d ic at es th at th e in fo rm at io n is av ai la b le in th a t sp ec ifi c w av e. “ R et ro sp ec ti ve y ea rl y fo r 1 9 7 9 -2 0 0 2 ” in d ic a te s th a t q u es ti o n s a b o u t th e sp ec if ca te go ry w er e as ke d fo r ea ch y ea r si n ce th e ye a r o f b ir th u n ti l 2 0 0 2 . 80 T ab le B 2: L it er at u re R ev ie w - H ig h in co m e C ou n tr ie s C ou n tr y Id en ti fi ca ti on S tr at eg y O u tc om es R es u lt s (1 ) (2 ) (3 ) (5 ) H ei la n d , K or en m an a, S m it h (2 01 9) U S H H F E Y rs of ed u ca ti on ≈ ze ro in ou tc om es Z it o (2 01 8) U S P S M A tt it u d es & n or m s ↑ ri sk av er si on . N o se lf -w or th or re la ti on sh ip eff ec ts D u n ca n , L ee , R os al es -R u ed a, K al il (2 01 8) U S O L S , H H F E Y rs of ed u ca ti on & b eh av io r p ro b le m s 1 y r d el ay in b ir th ↑ 0. 02 to 0. 04 S D in sc h o ol ac h ie ve m en t & ↓ p ro b le m s D ia z & F ie l (2 01 6) U S S m o ot h in g- d iff . & IP W E d u ca ti on al at ta in m en t an d ea rn in gs ↓ co ll eg e co m p le ti on , ea rl y ea rn in gs Y ak u sh ev a (2 01 1) U S P S M Y rs of ed u ca ti on ≈ 0 fo r h ig h -r is k te en s & lo w eff ec ts fo r te en s at lo w ri sk of T P A sh cr af t, F er n án d ez -V al & L an g (2 00 6, 20 13 ) U S IV (m is ca rr ia ge s) Y rs E d u ca ti on , G E D S co re , em p lo y m en t & m ar ri ag e G E D ↓ b y ab ou t 5 p p & ↓ 0. 15 y rs ed u c. E m p lo y m en t: ↓ 5 p p . M ar ri ag e ↓ 3 p p . K an e, M or ga n , H ar ri s, G u il ke y (2 01 3) U S O L S , P S M & M L Y rs E d u ca ti on ↓ 0. 7 an d 1. 9 y rs . of ed u ca ti on L ee (2 01 0) U S P S M E d u ca ti on , la b or fo rc e ↓ ea rl y so ci o ec on om ic ou tc om es F le tc h er & W ol fe (2 00 9) U S O L S & IV (m is ca rr ia ge s) G ra d u at io n , ea rn in gs ↓ 5- 10 p p h ig h sc h o ol gr ad u at io n , ↓ $1 ,0 00 to $2 ,4 00 an n u al in co m e F ra n ce sc on i (2 00 8) U K O L S , H H F E Y rs ed u ca ti on , b m i ↓ y rs ed u ca ti on , em p lo y m en t. ↓ C h il d h ea lt h in si n gl e p ar en t H ot z, M cE lr oy & S an d er s (2 00 5) U S IV (m is ca rr ia ge s) Y rs of ed u ca ti on , ea rn in gs N o ed u ca ti on eff ec ts , ↑ ea rn in gs at ol d er ag es K ap la n , G o o d m an , W al ke r (2 00 4) U K O L S , P S M & IV (m is ca rr ia ge s) E d u ca ti on at ta in m en t, em p lo y m en t ↓ la rg e ed u c. at ta in m en t, n o la b ou r eff ec ts L ev in e & P ai n te r (2 00 3) U S P S M , H H F E Y rs E d u ca ti on ↓ y rs ed u ca ti on & b ig ge r fo r te en ag er s at ri sk B it le r (2 00 1) U S IV (A b or ti on la w s) T im in ig of ab or ti on s ≈ ze ro in ou tc om es K le p in ge r, L u n d b er g, P lo tn ie k (1 99 5, 19 99 ) U S IV (t ee n ag e fe rt il it y ) & H H F E (1 99 9) Y rs of ed u ca ti on & w ag es ↓ -2 .1 4 y rs of ed u ca ti on , ↓ 2 y rs w or k ex p er ie n ce R ib ar (1 99 4) U S IV (a ge at m en ar ch e) H ig h sc h o ol co m p le ti on ↓ la b or fo rc e p ar ti ci p at io n , h ou rs of w or k G er on im u s & K or en m an (1 99 2) U S H H F E H ig h sc h o ol co m p le ti on ↓ sm al l eff ec ts in sc h o ol co m p le ti on 81