Research has documented a negative association between women’s educational attainment and early sexual intercourse, union formation, and pregnancy. However, the implications that school progression relative to age may have for the timing and order of such transitions are poorly understood. In this article, I argue that educational attainment has different implications depending on a student’s progression through school grades relative to her age. Using month of birth and age-at-school-entry policies to estimate the effect of advanced school progression by age, I show that it accelerates the occurrence of family formation and sexual onset among teenage women in Mexico. Focusing on girls aged 15–17 interviewed by a national survey, I find that those who progress through school ahead of their birth cohort have a higher probability of having had sex, been pregnant, and cohabited by the time of interview. I argue that this pattern of behaviors is explained by experiences that lead them to accelerate their transition to adulthood compared with same-age students with fewer completed school grades, such as exposure to relatively older peers in school and completing academic milestones earlier in life. Among girls who got pregnant, those with an advanced school progression by age are more likely to engage in drug use, alcohol consumption, and smoking before conception; more likely to have pregnancy-related health complications; and less likely to attend prenatal care visits. Thus, an advanced school progression by age has substantial implications for the health and well-being of young women, with potential intergenerational consequences.
Past research has found a strong negative relationship between educational attainment and the occurrence of transitions such as sexual onset, childbearing, and marriage or cohabitation formation around the world (Basu 2002; Clark and Mathur 2012; Dincer et al. 2014; Monstad et al. 2008). However, school progression implies different social experiences for students depending on their age relative to their peers. Women who start school at a younger age are exposed to older peers and will graduate from school at an earlier age than other women in their birth cohort (Argys and Rees 2008; Skirbekk et al. 2004). A small body of literature focused on a handful of high-income countries has studied the relationship among school starting age, school progression relative to age, and family formation. Black et al. (2011) found that starting school at a younger age increases the likelihood of teenage pregnancy in Norway. Skirbekk et al. (2004) found that graduating from school at a younger age accelerates the transition to motherhood and marriage in Sweden but seems to affect only the timing and not the eventual occurrence of these events. In contrast, McCrary and Royer (2011) found no effect of starting school at a younger age on the timing of the first birth in the United States. This last study also evaluated but did not find effects on infant health outcomes. However, the social and institutional contexts in these countries are remarkably different from those of the majority of teenage girls in the world, particularly those living in low- and middle-income1 countries (UNESCO 2016; UNESCO Institute for Statistics (UIS) 2015; United Nations 2015).
In this study, I assess the causal effect of progressing through school ahead of one’s birth cohort on the timing of sexual onset and family formation among teenage girls (aged 15–17) in Mexico, a developing middle-income country. I refer to this phenomenon as having an advanced school progression by age (ASPA). In contrast to previous studies, I evaluate the effect of school progression by age under social and institutional conditions that are pervasive globally. In Mexico, 13 % of girls aged 12–14 and 41 % of girls aged 15–17 were out of school in 2012, which is close to the corresponding figures globally (17 % and 38 %, respectively) in the same year (UIS 2015).2 Thus, worldwide, the period between ages 15 and 17 is a critical point in the transition to adulthood, in which girls have a high probability of leaving school and beginning family formation. This is especially true in the absence of enforced compulsory schooling laws. Although Mexico has laws that mandate completion of lower secondary (Grades 7–9) and upper secondary school (Grades 10–12), these rules are not enforced (Instituto Nacional para la Evaluación de la Educación (INEE) 2017; Rendall and Parker 2014). Similarly, as of 2012, one-half of the world’s countries do not mandate compulsory schooling beyond ninth grade (UNESCO 2016; UIS 2015). And in those countries with laws to keep students enrolled in school through adolescence, these rules have often been weakly enforced (Lloyd 2005; UNESCO 2016).
I show that in this context, school progression relative to age can be consequential not only for the timing of first union formation and first pregnancy but also for the age at sexual onset. More importantly, I show that an ASPA has potential intergenerational implications beyond accelerating family formation and sexual onset because it increases the probability of maternal risky behaviors and detrimental health outcomes. In addition to assessing these effects in the educational context of a middle-income country, this article contributes to the literature by evaluating the implications of school progression for outcomes that have not received attention in previous work: maternal health, prenatal care, and the probability of cohabitation. Finally, this study expands knowledge by showing that an ASPA affects not only the timing of school leaving and family-related transitions but also the order in which family formation events occur relative to leaving school.
By virtue of age-at-school-entry rules in Mexico, month of birth strongly predicts ASPA. To tackle concerns about unobserved heterogeneity, I take advantage of this natural experiment and use month of birth as an instrument for whether respondents have completed one more school grade than expected for their age by the time of interview. A conceptual innovation of this analysis is acknowledging that in a middle-income country such as Mexico, sexual onset, family formation, and schooling are closely related and are simultaneously determined decisions that do not necessarily follow a rigid order. To account for this, I use instrumental variable bivariate probit models to estimate the effect of an ASPA on sexual onset and family formation. These models allow me to estimate the correlation between the unobserved joint determinants of school progression and of the demographic transitions of interest. Then, I use reduced-form multinomial logit models to estimate the impact of an ASPA on the sequencing of family formation and sexual onset relative to school dropout, which has not been assessed by existing studies. To the best of my knowledge, this is also the first study to apply a causal inference approach to assess the effect of school progression by age in a context other than a high-income country, thus expanding our understanding on the complex relationship between schooling and the transition to adulthood among disadvantaged populations in the world.
My findings show that progressing through school at a younger age than other members of the same birth cohort increases the likelihood of having sex, becoming pregnant, and entering a cohabitation earlier in the life course compared with girls who started school later. I find that relative to staying in school and delaying family formation, girls with an ASPA are more likely to enter their first coresidential union while still enrolled in school and are more likely to leave school before getting pregnant. These findings suggest that plausible causal mechanisms are the exposure to relatively older peers in school and completing academic milestones earlier in life. Among girls who experienced their first pregnancy, an ASPA increases the probability of smoking, using drugs, and drinking alcohol before conception. In addition, it reduces the number of reported prenatal care visits and increases the probabilities of experiencing serious complications, such as bleeding excessively during delivery or miscarriage, and during the 40 days after the end of the pregnancy. Thus, beyond accelerating family formation and sexual onset, I find that an ASPA has substantial and potentially intergenerational implications for women’s well-being.
Theoretical and Empirical Background
Does the Timing of Sexual Onset and Family Formation Matter?
In the United States, recent studies have questioned the nature and durability of the effects of teenage pregnancy on mothers and children (Hotz et al. 2005; Lee 2010; Ribar 1994). Although adolescent mothers may be better able to overcome the difficulties of early childbearing in the United States, the prevalence and consequences of teenage pregnancy in low- and middle-income countries are substantially different. Evidence from these contexts suggests that teenage childbearing is associated with permanent reductions in work hours (Arceo-Gomez and Campos-Vazquez 2014) and educational attainment (Arceo-Gomez and Campos-Vazquez 2014; Eloundou-Enyegue 2004; Timaeus and Moultrie 2015); reduces the probability of receiving prenatal care (Macleod and Tracey 2010; Vieira et al. 2012); and increases the risk of developing anemia, postpartum hemorrhage, puerperal endometritis, and hypertensive problems during pregnancy, among other health complications (Conde-Agudelo et al. 2005; Mahavarkar et al. 2008). These health risks are shared by the offspring of teenage mothers. Evidence from countries of diverse income levels shows that mortality, pre-term delivery, and low birth weight rates are higher for babies born to teenage mothers (Branson et al. 2015; Chen et al. 2007, Fall et al. 2015; Nour 2006). Children of younger mothers are also less likely to complete secondary schooling in low- and middle-income countries (Fall et al. 2015).
Early sexual onset can disrupt an adolescent’s development and motivate future delinquent behavior (Armour and Haynie 2007). Both early sexual onset and union formation are associated with a greater risk of sexually transmitted diseases (Kaestle et al. 2005; Nour 2006, 2009), such as HIV (Wand and Ramjee 2012). Early union formation is also associated with higher risks of depression (Nour 2006, 2009) and higher malnutrition rates among the offspring (Raj et al. 2010). Transitioning to the first union before age 20 is negatively associated with union duration for both marriages (Lehrer 2008) and cohabitations (Kalmijn et al. 2007). In turn, union instability is associated with detrimental socioeconomic effects for both women and their children (McLanahan and Kelly 2006; McLanahan and Percheski 2008; McLanahan et al. 2013).
An ASPA may impact the timing of a woman’s sexual onset, first union formation, and first pregnancy through at least two counteracting sets of causal mechanisms: (1) human capital accumulation (or the acquisition of skills and knowledge); and (2) social age, constructed through peer effects and events that signal adulthood readiness (Basu 2002; Clark and Mathur 2012; Lam et al. 2013; Skirbekk et al. 2004; Zuilowski and Jukes 2012). In the absence of enforced compulsory schooling laws, as in the case of Mexico, the decision of when to leave school is closely linked to women’s transition to adulthood. Leaving school and experiencing events such as first sexual intercourse, first pregnancy, and first union formation are likely to be affected by very similar processes.
Human Capital Formation: Skills and Knowledge
Completing one more school grade during adolescence serves a number of instrumental goals, such as acquiring skills and knowledge, increasing potential earnings, and gaining autonomy and assertiveness to decide the timing of sexual onset and childbearing (Basu 2002). In economic theory, the relation between education and fertility is often explained as an increase in the opportunity cost of having a child (Axinn and Barber 2001). It follows that more years of education would provide incentives to delay sexual onset and union formation because both of these factors increase the risk of pregnancy. Schooling could also delay transition to childbearing by giving girls the information to use contraception effectively. Several studies have used compulsory schooling laws as instrumental variables to estimate the effect of completed school grades on fertility tempo and quantum, finding negative relationships (Black et al. 2008; Dincer et al. 2014; Monstad et al. 2008). The analytical strategies in these studies were designed to evaluate the average effect of one more year of schooling on the timing and number of births, regardless of age relative to peers.
Social Age: Peer Effects and Adulthood Readiness
Completed education relative to a woman’s biological age can also impact the timing of sexual onset, union formation, and childbearing through social and contextual channels. The social age that individuals acquire through school experience can be shaped by peer effects as well as by adulthood-readiness markers acquired when progressing through school grades (Skirbekk et al. 2004). Spending nearly one-half of their time at school and interacting with people other than their family could be nearly as important as the transmission of formal knowledge, skills for employment, and sex education (Basu 2002). As an institution that demarcates stages across the life course, school also places individuals in a particular position and social context, and conveys specific expectations and social norms to each stage (Pallas 1993, 2003). In this sense, school progression as a source of meaning is closely linked to the transition to adulthood. For women from disadvantaged backgrounds and for many women living in low- and middle-income countries, completing certain school levels during adolescence (such as lower secondary education) may signal an imminent readiness to start the next stage in the life course (Blossfeld and Huinink 1991; Zuilowski and Jukes 2012).
Evidence also suggests that teenagers’ sexual behavior is strongly influenced by peer’s sexual behavior and attitudes toward contraception (Loewenstein and Furstenberg 1991). As a consequence, mechanisms such as the increase in opportunity costs, which requires forward-looking calculations, may not be so relevant to explaining the relation between completed school grades and transitions to family formation for teenagers. In contrast, holding biological age constant, the proximate social context at school could be a strong predictor of the timing of family formation and sexual onset. For instance, Lam et al. (2013) showed that teenagers who are exposed to relatively older peers accelerate the timing of sexual onset in urban South Africa because these peers are more likely to be sexually active themselves.
Some studies have specifically used age-at-school-entry rules as exogenous instruments for school progression by age. Skirbekk et al. (2004) exploited differences in women’s age at graduation from lower secondary school induced by age-at-school-entry and compulsory schooling rules in Sweden, finding that an older age at graduation delayed the occurrence of first and second births as well as first marriage relative to biological age. The authors attributed these findings to differences in women’s social age, driven by the average student age within a school cohort (Skirbekk et al. 2004). Similarly, Black et al. (2011) found that in Norway, girls who start school at an earlier age because of age-at-school-entry rules are more likely to get pregnant during their teenage years. Along the same lines, Argys and Rees (2008) used kindergarten age-at-school-entry rules to instrument for the age of a student relative to her peers in the United States. They found that teenage girls, but not boys, who are young for their grade and exposed to older peers are significantly more likely to use drugs than their same-age counterparts who are relatively older for their grade.3 Using a similar strategy, other studies have found that a later age at school entry reduces the probability of juvenile criminal involvement in the United States (Cook and Kang 2016; Depew and Eren 2016).4 All these studies support the notion that an ASPA can accelerate the adoption of adult-like, and even deviant, behaviors among girls who are young for grade, either through exposure to relatively older peers or through the acquisition of adulthood-readiness markers earlier in the life course. As an exception, McCrary and Royer (2011) found no effect of starting school at a younger age on the timing of the first birth and negligible effects on infant health outcomes in the United States.
Overall, the implications of school progression relative to age for the timing of family formation have been scantly studied (McCrary and Royer 2011). To the best of my knowledge, the consequences for early cohabitation and maternal health have not been previously assessed. Although completing lower secondary and upper secondary school in Mexico became mandatory in 1993 and 2013 (Diario Oficial de la Federación 1993, 2012), respectively, these rules have not been enforced (INEE 2017; Rendall and Parker 2014). As mentioned earlier, the lack of effectively enforced rules guaranteeing that youth will remain in school until the end of adolescence is a very prevalent characteristic at the global level (UNESCO 2016; UIS 2015). In an attempt to account for the social and institutional realities of teenage women in Mexico and in a majority of countries, this study conceptualizes school progression, family formation, and sexual onset as potentially simultaneous decisions, where such transitions may have different possible sequencing orders relative to school dropout.
Schooling and Family Formation Among Teenagers in Mexico
Mexico is a country with a high degree of inequality and few opportunities of social mobility for disadvantaged youth (Binder and Woodruff 2002) as well as remarkably low wage returns to secondary relative to primary education (López-Acevedo 2006). The Mexican education system shares multiple characteristics with those of other low- and middle-income countries. In 2012, the cumulative dropout rate to the last grade of lower secondary school was high in Mexico (11 %) and equivalent to that of Pakistan, Cameroon, Nepal, and Uganda (UIS 2015). In the same year, 45 % of women aged 15–17 in the poorest quintile and 22 % of their counterparts in the richest quintile were no longer attending school in Mexico (UNICEF 2016). These rates are similar to the percentages of out-of-school female adolescents in the same age group in South and West Asia (53 %) and Latin America and the Caribbean (23 %), correspondingly (UIS 2015). In addition, the percentage of repeaters in elementary school in Mexico in 2012 (2.9 %) was comparable with that of middle-income countries (2.8 %) but was almost twice that of North America and Western Europe (1.6 %) (UIS 2015). Because of grade repetition and interruptions in school progression, approximately 18 % of students aged 12–14 and approximately 30 % of students aged 15–17 in Mexico were older for their school grade in 2015 (INEE 2016). Thus, age distribution throughout secondary school was skewed toward older ages, which likely magnifies the effects of exposure to older peers.
The teenage birth rate for women of ages 15–19 in Mexico was 65 per 1,000 females in 2010–2015, compared with rates of 30 in the United States, 6 in Norway, and 5 in Sweden over the same period (United Nations 2015). In Mexico, roughly 78 % of women who experienced their first pregnancy as teenagers were not enrolled in school when they got pregnant, and more than 50 % were already married or cohabiting (Stern and Menkes 2008). Similar to other national contexts, such as South Africa, teenage pregnancy is often not a cause of school desertion for girls but instead is part of an early transition to adulthood that was already underway (Macleod and Tracey 2010). In such disadvantaged social contexts, girls may identify the completion of certain school grades during adolescence as the right moment to start a family, which may eventually lead them to abandon school.
Data and Methods
Measuring the causal effect of ASPA is challenging because it is correlated with confounders that also contribute to delay the transitions of interest, such as social class background, personal efficacy, and attitudes toward school (Bonell et al. 2005; Skirbekk et al. 2004). I focus on the effect of ASPA as measured by having completed one more school grade than expected for a girl’s age by the time of interview.5 To isolate the effect of having completed more school grades than other students of the same age, I use a plausibly random instrumental variable for ASPA: month of birth.6 Between 1996 and 2006, the Mexican Education Ministry required that children had turned age 6 by September 1 in order to be enrolled in elementary school that year (Diario Oficial de la Federación 1996, 2006). Although children born in August would have turned age 6 just in time to start school in September, children born in October are usually held back almost an entire year before being able to enroll. Assuming that grade repetition is unrelated to school starting age, and holding age constant, teenage women who were born in August will be more likely to have completed one more school grade than expected for their age at the time of interview compared with those born in October. For instance, students are expected to have completed at least eight school grades by the time they turn age 15 (INEE 2016), but those students born in August would have completed nine grades because they started school one year earlier.7 As will be discussed later, the likelihood of repeating a grade is unrelated to month of birth in Mexico. ASPA represents a complex array of experiences, which comprises greater accumulation of human capital by age, exposure to relatively older peers, and the completion of academic milestones at an earlier age than other members of the birth cohort. However, all these experiences stem from progressing through school grades ahead of other girls of the same age, which is the concrete predictor that month of birth is an instrument for.
I use the 2009 and 2014 National Surveys of Demographic Dynamics (Encuesta Nacional de la Dinámica Demográfica, ENADID), made available by the National Institute of Statistics and Geography (Instituto Nacional de Estadística y Geografía (INEGI) 2016a). These nationally representative cross-sectional surveys provide retrospective sexual activity, union, and pregnancy histories. My main analytical sample has 3,530 women who were born between 1991 and 1998 and turned age 6 between 1997 and 2004, when the aforementioned policy was in place. I restrict the sample to respondents aged 15–17 because the median woman in Mexico leaves school and initiates the transition to adulthood between these ages (Echarri-Cánovas and Pérez-Amador 2007), but I provide descriptive analyses for women aged 18–20 for comparison. The questions of interest were not asked to any woman younger than age 15. I focus on women born in either August or October, so that the greatest distance between any two dates of birth is three months. Focusing on women born in the months immediately after and before the cutoff makes the assumption of random assignment of the instrument more feasible.8 Using data from birth certificates made available by the Mexican Civil Registry, Table 1 shows evidence in support of this assumption: the birth order and sociodemographic characteristics of babies born in August and October between 1991 and 1998 are equivalent, on average. See the online appendix for methodological details about Table 1, and further discussion of the instrumental variable assumptions in this study. Overall, 60 % of respondents were compliers: their completed school grades responded to the instrument (month of birth) as expected. The percentage of compliers did not change monotonically across cohorts, although compliance was the highest among girls born in 1998 (67 %) and the lowest for girls born in 1991 (55 %) (see Table A1 in the online appendix).
During the period in which cohorts 1991–1998 were of school-appropriate age, basic education in Mexico comprised elementary school (primaria), from Grades 1 to 6, and lower secondary school (secundaria), from Grades 7 to 9.9 Age-at-school-entry rules, school calendars, grading reporting rules, and curricula for both private and public schools are centralized and dictated by the Ministry of Education, which also produces and distributes textbooks for Grades 1 to 9 at no cost for parents or schools (SEP 2000). Students enrolled in public schools pay no tuition, and access to primary schooling is nearly universal: as of 2005, nearly 100 % of children between ages 6 and 11 were enrolled in elementary school (INEE 2006).
Bivariate Probit for Endogenous Outcomes
Bivariate probit estimation has been widely applied in the literature related to binary decisions or outcomes that are jointly and simultaneously determined, such as fertility and employment (Ekert-Jaffe and Stier 2009), fertility and schooling (Ribar 1994), or homeownership and childbearing (Öst 2012). As it applies to this study, bivariate probit estimation with an instrumental variable is aimed to correct the endogeneity bias that would result from a naïve regression that used ASPA to explain the timing of sexual onset and family formation. Such analysis would ignore the fact that women who are more likely to have an early sexual onset may also be less likely to complete school grades by the normative age.
The outcome in Eq. (1) is the latent propensity of having an ASPA. This binary outcome equals 1 if the respondent had completed more school grades than expected for her age according to the Mexican school system, and 0 otherwise. According to the Mexican education system, students are expected to have completed at least up to eighth grade by the time they are age 15, at least up to ninth grade by the time they are age 16, and at least up to tenth grade by the time they are age 17 (INEE 2016). Because the 2009 survey was conducted in May (before the school year ended), and the 2014 survey was conducted in August (after the closure of the school year, which ends in early July), the expected number of school grades was adjusted accordingly. A girl who is 15 in January of a given survey year would be expected to have completed eight years of education by May but nine by August, after the end of the school year. Thus, a 15-year-old girl interviewed in May would obtain a 1 in the outcome variable if she had completed more than eight grades, and 0 otherwise. In contrast, a 15-year-old girl interviewed in August would obtain a 1 if she had completed more than nine grades, and 0 otherwise (see Table A2 in the online appendix for coding details).
Equation (1) includes controls for respondent’s age,10 state dummy variables, a 2014 survey year dummy variable, and two municipal-level characteristics: percentage population with less than elementary school in 2000 (CONAPO 2012) and the ratio of higher secondary schools (preparatoria and técnica) to the population aged 15–19, measured in the year when respondents were 11 years old and faced the decision of whether to start secondary school (INEGI 2016b). This set of controls (X) is intended to capture differences in the aggregate local propensity to stay enrolled in school during adolescence. State dummy variables capture potential regional differences in educational attainment and adherence to school enrollment rules. The survey year dummy variable is included to capture any period differences that may be related to both schooling and the outcomes of interest, as well as any differences in instrument compliance levels between girls born in 1991–1993 (all interviewed in 2009) and girls born in 1996–1998 (all interviewed in 2014).
Equation (2) uses ASPA as a predictor of family formation transitions and is jointly estimated with Eq. (1) using an iterative maximum likelihood procedure. The bivariate probit is able to identify the effect of an ASPA in Eq. (2) because Eq. (1) includes the exogenous instrument, M (Angrist and Pischke 2009). In addition, Eq. (2) controls for the same set of covariates X included in Eq. (1). I capture family formation transitions using a series of binary variables that are analyzed separately and that record whether by the time of interview the respondent had ever had sex, had ever been pregnant, had ever been married, or had ever cohabited.
Assessing the Order of Events: Reduced-Form Multinomial Logistic Models
The bivariate probit models evaluate the effect of ASPA on family formation in the presence of shared unobserved determinants of both variables. However, without enforced compulsory schooling laws, deciding when to leave school becomes a part of the transition to adulthood for young women. In the bivariate probit models, the order of school desertion and family formation transitions is unclear because completed education is measured only at the time of interview, and family formation and sexual onset are captured as having ever occurred. Causal mechanisms may be better understood by assessing the effect of month of birth on the order in which events happen relative to leaving school, if this has occurred. Only the latest wave of ENADID asked what age respondents were when they left school. Thus, I restrict this part of the analysis to 3,121 respondents in this latest wave, this time encompassing ages 15–20 to conserve statistical power.
I estimate reduced-form multinomial logistic regressions to evaluate the effect of month of birth on the probability of observing each of these statuses at the time of interview: (1) leaving school and family formation (sexual onset) happened at the same age, (2) leaving school occurred first, (3) family formation (sexual onset) occurred first, and (4) none of these events has occurred (reference). These models include the same set of individual and municipal-level controls used in the bivariate probit models. They assess the effect of the instrument—month of birth—as a proxy for a higher likelihood of having an ASPA. Estimates are likely to be underestimated and conservative due to the presence of women whose completed school grades by age did not respond to the instrument (noncompliers).
Health Implications: Reduced-Form Linear Probability Models
The ENADID asked a series of questions about risky behaviors and maternal health outcomes surrounding the last pregnancy. To ensure that all observations correspond to the first pregnancy, I restrict the sample to women who had only had one pregnancy that had ended by the time of interview.11 Using reduced-form linear probability models, I assess the effects of month of birth on whether the respondent drank alcohol, smoked, or used drugs before getting pregnant; the number of prenatal care visits; whether the pregnancy resulted in a live birth;12 whether she had vaginal bleeding during the pregnancy; whether she experienced excessive bleeding during delivery or miscarriage; and whether she had excessive bleeding during the 40 days after delivery or miscarriage. These models include the same controls used in the bivariate probit analysis. I include women aged 15–20 in this exercise to maximize the chances of observing the first pregnancy by the time of interview, which results in a sample of 648 women. Questions on risky behaviors and on excessive bleeding after delivery or miscarriage were asked only in 2014 and thus have a smaller sample (534).
Main Analysis: Bivariate Probit Models
Table 2 shows means for the outcomes and covariates included in the analysis, comparing characteristics of women aged 15–17 and born in August and October. As a comparison, I provide the same set of descriptive statistics for women aged 18–20. For the transition to adulthood indicators—having ever had sex, having ever been pregnant, having ever been in a union, and currently attending school—the mean differences between women born in August and October are large and significant for those in the younger group, but they become smaller and mostly nonsignificant for the older group. In Mexico, the risk of leaving school and experiencing sexual onset and family formation increases rapidly after age 15, which highlights the importance of analyzing these dynamics among the younger age group.
Table 3 shows three naïve linear probability models assessing the effect of ASPA on family formation and sexual onset for women aged 15–17, using the same set of controls as the rest of the models in this study. These models exclude month of birth as an instrument, so they do not account for the endogenous relationship between family formation transitions and school progression. I observe a strong negative association between an ASPA and the probabilities of having ever had sex, having ever been pregnant, having ever been married, and having ever cohabited (p < .001).
Results change drastically when I use month of birth as an instrument for ASPA in a bivariate probit, which takes into account the endogenous relationship between this predictor and family formation transitions. Table 4 presents the two simultaneous equations of the bivariate probit analyses for women aged 15–17 born in August or October. The first equations, labelled “ASPA,” show the strong positive relationship between being born in August and the probability of being a school grade ahead than other same-age youth. The F test for the instrument, month of birth, in Eq. (1) is 82 (not shown), well above the recommended threshold of 10 for reliable instrument strength. Although the magnitude of effects represented by probit coefficients is not directly obvious, the coefficients for ASPA in the second equations show positive and significant impacts on the probabilities of having ever had sex, having ever been pregnant, having ever been in a union, and having ever cohabited (p < .05). The effect on having ever been married is nonsignificant, which suggests that accelerated union formation among young-for-grade women is driven by early cohabitations. Translated into local average treatment effects (LATE), having an ASPA increases the probability of having ever had sex by 17 percentage points among Mexican teenage girls. It increases the probability of having ever been in a union by 7 percentage points, of having ever cohabited by 10 percentage points, and of having ever been pregnant by 9 percentage points (numbers not shown in table). At the bottom of Table 4, ρ indicates that the correlation between the unobserved determinants of ASPA and each family formation transition (except for marriage) is near –.9 and is highly significant (p < .001). This means that unobserved factors that increase educational attainment by age also delay the occurrence of sexual onset, first pregnancy, and first union formation, in the form of cohabitation. Some of these factors could be parental monitoring, personal motivation, or school quality. A model that does not account for this high negative correlation would necessarily render biased results (Chiburis et al. 2011).
As a robustness check, I estimate linear two-stage least-squares (2SLS) regressions using the same instrumental variable, outcomes, and controls included in the main bivariate probit models. These estimates of LATE are overall positive and significant for all transitions except for marriage, although their magnitudes are about twice as large as those of bivariate probit models (see Table A3 in the online appendix). Although 2SLS models offer further evidence of positive treatment effects, bivariate probit estimates are preferable because they are less prone to bias when the probabilities of observing the outcomes are closer to 0 than to .5 and when the sample size is smaller than 5,000 (Chiburis et al. 2011), which are the characteristics of my sample.
Assessing the Order of Events: Reduced-Form Multinomial Logistic Models
Panel 1 in Table 5 presents a model in which the events of interest are leaving school and sexual onset. Results show that being born in August has a positive and significant effect on the probabilities of both experiencing sexual onset before school dropout and leaving school before sexual onset, relative to experiencing neither sexual onset nor school dropout by the time of interview (p < .05). This suggests that the acceleration of sexual onset could be due to both exposure to older peers (when first sex happens before leaving school) and completed school grades representing an adulthood-readiness marker (when first sex happens after leaving school).
Panel 2 in Table 5 shows odds ratios from an analogous model in which the events of interest are leaving school and first union formation. Results show that a higher likelihood of an ASPA, as indicated by being born in August, has a positive and significant effect on the occurrence of one particular scenario: entering the first coresidential union before leaving school, relative to none of these events occurring by the time of interview (p < .05). Panel 3 in Table 5 shows another analogous model, in which the events of interest are leaving school and first pregnancy. Results show that being born in August has a positive and significant effect on the likelihood of leaving school before getting pregnant for the first time, relative to none of these events occurring by the time of interview (p < .05).
These findings suggest that young-for-grade girls are more likely to meet their first coresidential partner and form their first union—most likely, cohabitation—while still enrolled in school, possibly influenced by older peers. These results also suggest that progressing through school ahead of other same-age students accelerates the acquisition of adulthood markers, which makes teenage girls more likely to interrupt their schooling before getting pregnant for the first time. Thus, any policy measure intended to increase educational attainment and reduce teenage birth rates among young-for-grade girls will not be well served by focusing only on avoiding teenage pregnancy: adolescents with an ASPA tend to start their transition to family formation well before the first pregnancy. Such policies would also need to target their higher propensity for early union formation and school-leaving.
Health Implications: Reduced-Form Linear Models
Table 6 shows reduced-form linear regressions assessing the effect of being born in August or October on maternal risky behaviors and health surrounding the first pregnancy. According to Table 6, women who were born in August—and who thus were more likely to have started school about one year earlier than those born in October—were also 9 percentage points more likely to report drinking alcohol, 8 percentage points more likely to report smoking, and 3 percentage points more likely to admit to using drugs before their first pregnancy (p < .01). These risky behaviors are often considered to be adult-like activities, so their higher prevalence among girls with an ASPA is consistent with the mechanisms discussed in the Social Age: Peer Effects and Adulthood Readiness section and with the findings in Argys and Rees (2008).
The second panel in Table 6 shows analogous models assessing the effect of being born in August instead of October on a series of maternal health indicators related to the first pregnancy. Being born in August, and thus having a greater likelihood of being young for grade, increases the probability of experiencing excessive bleeding during the event that ended the pregnancy (either delivery or miscarriage) and during the 40 days after the end of the pregnancy by 6 percentage points each (p < .05). These results suggest that getting pregnant at a younger age increases the risk of complications. Finally, even though month of birth does not have a significant effect on the likelihood of the first pregnancy ending in a live birth, being born in August results in 0.8 fewer prenatal care visits (p < .01).
It could be argued that social age might not be the only channel through which school progression by age could accelerate sexual onset and family formation transitions. For instance, existing studies suggest that U.S. children who start school at an earlier age have lower test scores than those who started later, are more likely to repeat grades, and are less likely to enroll in college (Bedard and Dhuey 2006). In supplementary analyses, I used data from the Mexican Family Life Survey (Rubalcava and Teruel 2013) to assess whether month of birth affects the probability of having ever repeated a grade and found a nonsignificant relationship (see Table A4 and section A.2.1 in the online appendix for methodological details). Using the 2014 ENADID, I estimated a linear regression evaluating the effect of being born in August instead of October on the total number of school grades completed at age 22, when Mexican students should have finished college. The coefficient for month of birth is nonsignificant for this older group of women, which suggests that there are no cumulative long-term differences in educational attainment due to school starting age (results not shown but available upon request). Finally, I used 2012 PISA scores (OECD 2009, 2013) for 15-year-old female students in Mexico to assess whether having an ASPA, and thus one more completed school grade by age 15, has an impact on test scores and found positive and significant effects (see Table A5 and section A.2.2 in the online appendix for methodological details). This evidence suggests the effects of an ASPA are not due to a human capital disadvantage relative to same-age students who are one school grade below. The robustness checks in this section support the notion that an ASPA accelerates family formation and sexual onset through the social age assigned by school, and not through disadvantages in academic performance, higher repetition rates, or lower long-term educational attainment.
Using the case of Mexico, I show that teenage girls who progressed through school grades at a younger age relative to other women in their birth cohort are more likely to accelerate their sexual onset, first pregnancy, and first union formation in the form of cohabitation. I apply a causal inference research design in which family formation events and completed school grades are conceptualized as mutually dependent, simultaneous decisions and offer evidence of the highly significant and extensive negative correlation between the unobserved determinants of these outcomes. This shows that any research design that does not account for such endogeneity will necessarily render biased results. Supplementary analyses suggest that my findings are not driven by disadvantages in academic performance, grade repetition rates, or lower long-term educational attainment.
This study makes several contributions to the few prior studies on school progression relative to age and family formation. First, it evaluates this question in Mexico, a country with an education system that shares multiple characteristics with those of other low- and middle-income countries, such as lack of enforced compulsory schooling laws and high dropout rates in secondary school. Using the Mexican case provides insight into the complex implications of schooling for young women in a social context similar to that of teenagers in other developing countries, where approximately 80 % of the global population resides (Population Reference Bureau 2013). Second, among the evaluated outcomes, this study assesses the implications of school progression for maternal health, prenatal care, and cohabitation, which have not received attention in previous studies.
Third, this study presents evidence that an advanced school progression by age has substantial and potentially intergenerational implications for women’s well-being. I find that among girls who have experienced their first pregnancy, an advanced school progression by age increases the probability of excessive bleeding during delivery or miscarriage, and during the 40 days after the end of the pregnancy. This finding is worrisome because severe bleeding is the leading cause of maternal mortality worldwide, and postpartum hemorrhage is responsible for 25 % to 30 % of pregnancy-related deaths in developing countries (Sosa et al. 2009). In addition, my analyses show that an advanced school progression by age reduces the number of prenatal care visits during the first pregnancy. Because at least part of the adverse pregnancy-related outcomes for adolescent mothers can be explained by lower access to prenatal care (Vieira et al. 2012), this relationship suggests that biologically younger girls might have less access to the medical services needed to have a healthy pregnancy, even if they perceive themselves as “older.” Thus, these results suggest that accelerating transitions such as the first pregnancy has serious detrimental consequences for August-born teenage girls, including making them more prone to life-threatening complications, even if this event occurs no more than 10 months earlier than among their October-born counterparts.
I also find that among girls who experienced a first pregnancy, an advanced school progression relative to age increases the probability of drinking alcohol, smoking, and using drugs before getting pregnant. This last finding is consistent with what Argys and Rees (2008) found for female young-for-grade girls, although that study did not analyze teenagers who had ever been pregnant. Combined with a greater likelihood of experiencing a teenage pregnancy, these risky behaviors pose a threat to the well-being of both women and their unborn offspring. Early initiation of drinking alcohol, smoking, and using drugs is associated with a higher risk of later nicotine dependence (Kendler et al. 2013), problem drinking in adulthood (Warner 2003), and a higher likelihood of developing drug addiction (Lynskey et al. 2003), respectively. The early initiation of these problem behaviors is also associated with a higher probability of having antisocial personality and major depressive disorders later in life (McGue and Iacono 2005). Assuming that pregnant adolescents are likely to continue smoking and using drugs and alcohol at least until they find out they are pregnant, these behaviors have other important health implications for mothers and babies, such as greater risk of congenital cardiac defects (Mateja et al. 2012) and maternal depression (Lassi et al. 2014). In addition, substance use is positively associated with risky sexual behaviors (Ritchwood et al. 2015), which may at least partially contribute to the higher probability of teenage pregnancy among young-for-grade women.
Finally, my analysis shows that being young for grade affects not only the timing of school leaving, sexual onset, and family formation but also the order in which family formation and sexual onset occur relative to school dropout, which has not received attention in the existing literature. I find that relative to staying in school and delaying family formation, girls with an advanced school progression by age are more likely to enter their first union—usually cohabitation—while still enrolled in school but are more likely to leave school before they get pregnant. This is consistent with two potential causal mechanisms that may contribute to the acceleration of family formation and sexual onset: (1) exposure to older peers while attending school, and (2) the acquisition of an adulthood social marker after reaching educational milestones. It follows that progressing through school at an earlier age changes girls’ life trajectories in ways that can be very consequential for their well-being. For instance, a girl who starts cohabiting while still enrolled in school may experience a more unstable first union than a girl who waits until she finishes her schooling, given that the former will have access to a smaller and lower-quality pool of potential partners, likely limited to her schoolmates.
Given that the absence of enforced compulsory schooling laws allows early starters to abandon school at younger ages in Mexico, proper enforcement of these laws could counteract social age effects if compulsory education is extended—and enforced—up to late adolescence. Research has shown that the mandatory human capital accumulation and the “incarceration effect” of having to spend more time in school can reduce teenage pregnancy rates (Black et al. 2008). Policy-makers may also consider designing interventions that target young-for-grade girls, with the goal of delaying their transition to adulthood, particularly in the presence of a high proportion of over-age students. In addition, improving school infrastructure, guaranteeing that schooling remains a feasible alternative for teenagers, and increasing returns to education beyond the ninth grade are key to incentivizing young women to stay enrolled in school through adolescence. Such measures can mitigate the acceleration of family formation and sexual onset induced by an advanced school progression relative to age, and its detrimental implications for the well-being of young women.
I am deeply grateful to Florencia Torche, Paula England, Lawrence Wu, Julia Behrman, José Ortiz, Andrés Villarreal, Michael Rendall, three anonymous reviewers, and the members of the New York University Inequality Workshop for their valuable comments on previous drafts. All remaining errors are my own. I gratefully acknowledge support from the Eunice Kennedy Shriver National Center for Child Health and Human Development Grant P2C-HD041041, Maryland Population Research Center.
Low- and middle-income here are defined according to the United Nation's income-based country categories (United Nations 2018).
In contrast, in Norway, Sweden, and the United States, less than 4 % of girls aged 12–14, and less than 10 % of those aged 15–17 were out of school as of 2012 (UIS 2015).
Past studies have also assessed the effect of school starting age on other outcomes, such as educational attainment and earnings in the United States (Angrist and Keueger 1991; Dobkin and Ferreira 2010) and Sweden (Fredriksson and Öckert 2013), although results have been mixed and highly dependent on the configuration of compulsory schooling laws in each country.
However, given the compulsory schooling laws in North Carolina, Cook and Kang (2016) also found that late starters are more likely to drop out of high school and commit a felony offense by age 19.
About 16 % of women with an ASPA in the data had actually completed two more school grades than expected for their age, possibly because having a daughter born in August makes it easier for parents to negotiate her school enrollment before age 6, despite it being unusual. The other 84 % had completed only one more grade, as expected. The results of all the analyses remain unchanged if the 16 % of women with the most advanced school progression are excluded. But for simplicity, throughout this article, I refer to the entire group of women with an ASPA as having completed one more grade than expected for their age.
This is an instrument similar to that used by Angrist and Keueger (1991), with the exception that this analysis does not use the combination of compulsory schooling laws and month of birth as an instrumental variable.
This is true for girls who are still enrolled in school or for those who have dropped out and spent less than one school year out of school. For girls who dropped out more than a school year before the interview date, this variable is not informative. To test the sensitivity of results to this limitation, I estimated the same bivariate probit models excluding girls who did not complete any grade beyond elementary school (seventh grade or higher) because they are likely to have abandoned school more than a school year before the interview date. Results using this restricted sample are closely similar to those in Table 4.
I exclude from the analysis girls born during September because they are less likely to follow the age-at-school-entry rule. Given that their date of birth would be less than a month away from the cutoff, school officials commonly allow children born in September to enroll before they turn age 6, even if their birth date is not September 1. Focusing the analysis on girls born in August and October reduces uncertainty in enrollment practices for children born in September.
Preschool education was neither mandatory nor widely available in Mexico between 1994 and 2001, which is the year range within which the women in the study (born between 1991 and 1998) would have become 3 years old.
I use a homogenized age-at-interview variable that records full years of age in July of the corresponding interview year. Because interview dates in the 2014 wave ranged from August to September, some respondents were n years old at the time of interview, while some of their counterparts born within less than three months had already turned n + 1. If full years of age at the exact interview date were included in the model as a control for age, they would distort the real distance in biological age for respondents who have a birth date in August. The homogenized age at interview used in the models is perfectly collinear with year of birth, which is not included as a covariate.
This implied excluding 15 % of girls born in October and 17 % of girls born in August among those who had ever been pregnant.
The sample contains too few miscarriage and stillbirth cases to be able to assess the effect of month of birth on the prevalence of each of these outcomes separately.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.