Secondary Schools and Teenage Childbearing: Evidence from the School Expansion in Brazilian Municipalities

Abstract

This article investigates the effect that increasing secondary education opportunities have on teenage fertility in Brazil. Using a novel dataset to exploit variation from a 57 percent increase in secondary schools across 4,884 Brazilian municipalities between 1997 and 2009, the analysis shows an important role of secondary school availability on underage fertility. An increase of one school per 100 females reduces a cohort's teenage birthrate by between 0.250 and 0.563 births per 100, or a reduction of one birth for roughly every 50 to 100 students who enroll in secondary education. The results highlight the important role of access to education leading to spillovers in addition to improving educational attainment.

1. Introduction

A 2012 report by the World Bank on teenage pregnancy stresses the correlation between teenage childbearing and socioeconomic variables, including poverty, inequality, public health expenditure, and female labor force participation. The report shows that, despite substantial reductions in teenage pregnancy rates in virtually all countries, observers continue to see that rates differ vastly between high- and low-income countries, with Brazil – as a middle-income country – being placed somewhere in the middle of the distribution (World Bank 2012).

Improved access to education, not explored in detail in the World Bank report, potentially provides an important channel through which teenage pregnancy and the above socioeconomic variables are correlated. Observational evidence shows a strong negative relationship between school availability and teenage childbearing in Brazil. This paper plots this relationship over time for Brazil in fig. 1. Between 1997 and 2009,¹ 9,402 secondary schools were introduced in Brazilian municipalities (a 57 percent increase), raising the average school density in municipalities from 1.06 per 100 teenagers to 1.54 per 100 teenagers. Over the same period, the birth rate for teenage girls² decreased by 21 percent from 8.1 births per 100 to 6.4 births per 100. This suggests an additional secondary school per 100 teenage females is associated with a decrease in births of 3.33 births per 100 teenage females. Evidence based on cross-sectional data shows a very similar picture. Figure 2 presents the state-level relationship between school density and the rate of teenage childbearing. Based on this clear negative relationship the study calculates that an additional secondary school per 100 teenage females is associated with a larger decrease in births of 17.9 births per 100 teenage females.³

This paper investigates whether the negative association between secondary school availability and teenage childbearing is based on a causal relationship. This is done by looking at the effect of a large secondary school expansion across 4,884 Brazilian municipalities on teenage childbearing. Conditioning on municipal fixed effects and other determinants of school introductions, this expansion provides a plausibly exogenous source of variation in school availability. The study's main results show that, on average, one additional secondary school per 100 females in the age cohort decreases teenage childbearing by between 0.250 and 0.563 births per 100 females. While being smaller than the effects that are measured using casual observation in figs. 1 and 2, the main estimates are statistically significant and economically relevant.

This study's estimates rely on the assumption that the secondary school expansion is orthogonal to levels and trends in municipal teenage childbearing. The paper provides an analysis of the expansion and finds that, controlling for variables reflecting school supply and demand, where and when a school is introduced is independent of variation in teenage childbearing. The study also finds no evidence of pre-trends in childbearing prior to school introductions. The study's main estimates are robust to a number of different specifications and robustness checks, including estimates identified only from variation in the timing of school introductions across municipalities of similar size.

Brazil is particularly well suited for studying this research question. The school expansion that is examined constitutes one of the largest expansions of secondary schools on record. The study uses information from 13 waves of the annual Brazilian school census, containing detailed information on the universe of Brazilian schools, to create a new dataset reflecting the availability of secondary schools in every Brazilian municipality between 1997 and 2009. This information is combined with vital statistics data from Brazil capturing the universe of live births including information on the age of mothers at date of conception over the same period, creating a rich and unique dataset. This seems to be the first paper to document and utilize data on the rapid growth of secondary schools across Brazil over the two decades starting in the 1990s.

This study contributes to a literature examining the relationship between education and fertility in young women. Lowering the explicit cost associated with education, by providing free school uniforms (Duflo, Dupas, and Kremer 2015) or the removal of school fees (Chicoine 2020), leads to a greater number of girls in primary education and significantly reduces childbearing for young women in African countries. Increasing the time spent in education also affects teenage childbearing. In Chile, a greater number of full-day schools, as opposed to half-day schools, decreases the probability of adolescent motherhood (Berthelon and Kruger 2011). Papers using variation from changes to mandatory schooling laws in high-income countries find that increases in the mandatory schooling age lead to a large and significant decrease in motherhood for young women in Norway, the United States and the UK (Black, Devereux, and Salvanes 2008; Monstad, Propper, and Salvanes 2008; Geruso and Royer 2018). However, a UK policy allowing young women to leave school six months earlier does not lead to an increase in teenage motherhood (James and Vujic 2019).

The findings in the literature indicate that there are multiple structural channels at work in the education-fertility relationship. School attendance restricts the time outside school available to young men and women (an incarceration effect), but education may also a change preferences or opportunity costs of young men and women (a human capital effect) (Black, Devereux, and Salvanes 2008).

One may also expect expanding access to schools to lead to changes in fertility through indirect channels. Secondary school education may increase a young adult's value in the marriage market (Fort, Schneeweis, and Winter-Ebmer 2016), or promote greater assortative matching, both of which may result in earlier childbearing. Previous studies have had limited success disentangling the relative importance of these different channels.

This paper is mainly interested in testing for a causal relationship between the school expansion and teenage fertility, while remaining agnostic about the channels through which improving a community's school access might affect teenage childbearing. In addition to the channels mentioned above, the introduction of a school may influence childbearing outcomes in difficult to identify, but nonetheless important ways. For example, improving access to secondary schools may disproportionately benefit human capital opportunities for girls relative to boys, and this may play a role in changing social norms towards early motherhood (Duflo 2012). The paper's reduced-form estimates reflect the net effect of all of these different channels.

The paper makes two key contributions to the existing literature. First, its approach is different from several previous studies that exploit changes to mandatory school attendance age in countries with relatively good access to secondary schools, such as the United States and Norway (Black, Devereux, and Salvanes 2008; Monstad, Propper, and Salvanes 2008). Increasing the school leaving age in these settings elicits a behavioral response from youths for whom the benefit of school attendance (subjective or objective) is low (Brunello, Fort, and Weber 2009; Fort, Schneeweis, and Winter-Ebmer 2016).⁴ In contrast, the school expansion leads to a behavioral change from young men and women who attend school when access is good – but not when access is poor – without explicit compulsion. In affluent countries with good school access, this margin of the population is not expected to respond to a change in mandatory schooling laws. For this reason, estimates based on mandatory schooling laws may be uninformative for inferring the effect of changes in access to schooling. Despite the stark difference in the approach chosen in this paper, its estimates are similar to the findings of studies using compulsory schooling. For example, compelling youths to attend school until age 16 reduces the probability of a teenage birth by 4.7 percent in the United States and 3.5 percent in Norway (Black, Devereux, and Salvanes 2008). The present study finds that one more school per hundred females reduces the rate of teenage births for a cohort by 1.4 percent.⁵

The second contribution comes from the focus on a middle-income, as opposed to a high-income, country. Teenage childbearing in Brazil is significantly higher than in many high-income countries, but lower than in many of the poorest countries.⁶ In addition, the focus on Brazil makes it possible to study a recent and rapid expansion of secondary schools over a relatively short time period, which is not possible in the context of high-income countries, where such expansions have taken place decades earlier and often in a more gradual fashion.⁷ The study is therefore able to provide the first quantification of the role of expanding education access in the substantial reduction of teenage fertility rates observed across many middle-income countries over the last two decades. The expansion of secondary schooling in Brazil provides a blueprint for understanding the effect of the expansion of the educational system on fertility. The estimates provided here are therefore of relevance for a large group of low- and middle-income countries currently experiencing or on the brink of similar expansions of their educational systems.

The remainder of this paper is organized as follows: section 2 provides the background on the provision of secondary education in Brazil. Section 3 discusses the data to be used in the main analysis. Section 4 introduces the empirical strategy and shows that the introduction of schools in Brazilian municipalities is uncorrelated with latent factors that might impact fertility decisions. Section 5 presents and discusses the results. Section 6 provides additional robustness checks. Section 7 concludes the article.

2. Background Information

Secondary education in Brazil typically lasts for three years and is preceded by nine years of primary school.⁸ Of the 11,007 public secondary schools in 1997 for the municipalities in the present study, 97 percent were under state control. State secretariats of education are responsible for the regulation and general management of secondary schools, including the recruitment of teachers and curriculum content (JBIC 2005). There is no minimum age for initial enrolment to secondary school, but it is targeted at age 15, and students must have completed primary school first. There is no maximum age limit and, because of common late enrolment and grade retention in primary schools, age-grade mismatch at secondary school is frequent (Foureaux Koppensteiner 2014). In 2010, about 30 percent of students in the first grade of secondary school were above the target age (IBGE 2012).

Over the last two decades, many low- and middle-income countries have undergone an expansion of their secondary education system, driven by improvements in primary school completion rates and an increased demand for a more highly skilled workforce (World Bank 2005). In Brazil, secondary schooling was an overlooked part of the education system until the beginning of the 1990s (Guimarães de Castro and Tiezzi 2004). Prior to the 1990s, secondary education was highly geared to the elites in preparation for entrance to higher education and was considered of little relevance for the education of the broader population. Following the end of the military dictatorship, the introduction of the Constitution of 1988 and the General Education Law (LDB 1996) of 1996 made access to secondary education a key aim on the political agenda, mandating it to be available (although not mandatory) for all those completing primary education. Following the rapid expansion of primary education, which led to universal enrolment towards the end of the 1990s, secondary education started to expand rapidly as well (Marchelli 2010).

The expansion followed the increasing demand from larger numbers of final year primary students (Di Gropello 2006; Moore et al. 2008; De Felizio 2009) – section 4 tests this formally. While the expansion in primary education was largely the responsibility of Brazilian municipalities, in part funded by federal resources through the program FUNDEF (Fundo de Manutenção e Desenvolvimento do Ensino Fundamental), the 26 Brazilian states and the federal district were in charge of the expansion of secondary education using financial support from the federal program PROMED (Programa de Melhoria e Expanção do Ensino Médio). The explicit aim of PROMED was to address excess demand for secondary education from primary students (Monteiro de Barros Araújo and Mendes Najjar 2017).

The expansion resulted in a 57 percent increase in the number of secondary schools, from 16,562 in 1997 to 25,964 in 2009. This was driven primarily by a 68 percent increase in the number of publicly funded schools, from 11,007 to 18,526. The school expansion had a non-trivial impact on school access across Brazilian municipalities. In particular, there was a remarkable increase in the availability of schools to the poorer northern states of Brazil; fig. 3 maps the change in secondary school availability across municipalities.⁹ In 1997, 316 municipalities, representing 6.5 percent of all Brazilian municipalities, had no secondary school. By 2009, this number had dropped to 12. There was also a notable 34 percent increase in the provision of private secondary education from 5,555 to 7,438. The number of students in secondary schools increased from under 6.4 million in 1997 to 8.3 million in 2009 (INEP 2003; INEP 2011).¹⁰

3. Data

The primary data used in this study come from two sources: the Brazilian school census (Censo Escolar) and Brazilian vital statistics data (SINASC) from the Ministry of Health. In addition, the study uses auxiliary data, including population estimates for Brazilian municipalities from the Brazilian Census Bureau, and municipal expenditure data from a variety of sources. Descriptive statistics for the key variables in the analysis are reported in table S1.1 in the supplementary online appendix. As municipal boundaries changed over the period of interest, the analysis is based on 4,884 minimal comparable areas (see S2 in the supplementary online appendix for details). For simplicity, the study continues to refer to the unit of observation as the municipality.

Schooling Data

The study uses 13 waves of the Brazilian school census, collected annually for the Ministry of Education by the Anisio Teixeira Institute of Research on Education (INEP), which provides administrative data on the universe of schools in Brazil (Glewwe and Kassouf 2012). The school census includes detailed information on the universe of public and private schools in Brazil, such as enrolment by grade, age, and sex, information on the number of classes, the physical characteristics of the schools, as well as information on teachers.¹¹ The school census is available from 1995, but because of inadequate quality of the data, the first two census years were discarded. The census data are used to create a dataset on the number of secondary schools, the number of classrooms, and the number of students between 1997 and 2009 – when the vast majority of the school expansion had occurred – collapsing the data by municipality and year. By the study's definition, a new school is introduced in a municipality if a new unique school identification number appears in the school panel.¹² Information on municipality codes makes it possible to locate every school in Brazil to the corresponding municipality. The school census provides information on primary school enrolment, the availability of nursery classrooms, and the number of preschool classrooms, which are used as controls.

Childbearing Data

Data on birth outcomes come from the microdata of Brazilian vital statistics, which cover approximately 45 million births occurring between 1997 and 2009. Vital statistics data are based on birth certificates issued by health institutions or midwifes attending homebirths and are collected through the states’ health secretariats. The vital statistics microdata are publicly available through the System of Information on Life Births (SINASC) of the Brazilian public health system (DATASUS). These data provide information on the age and municipality of residence of the mother, as well as gestational length of the pregnancies, and the mother-reported race of the child.

For each year, these data are collapsed to create a summary measure of births by municipality and mother's age at conception.¹³ Age at date of conception is calculated using information on gestational length recorded in the birth certificates to provide a municipal panel of births by mother's age at conception.¹⁴ Brazilian vital statistics data show close-to-universal coverage of all occurring births; information from the 2010 population census shows that more than 99 percent of all births occurring between 2000 and 2010 were registered and entered into the vital statistics data that were used in this study. The advantage of using vital statistics data to learn about fertility in the population comes from the universal coverage of the data for the entirety of Brazilian municipalities over the period of interest. Information about the residence of the mother during pregnancy is particularly important, as information on the place of birth may be misleading if there is a discrepancy between place of residence of the mother and the place of occurrence of birth, which is more likely for relatively small municipalities that do not have clinics with birth facilities.

Population Estimates

The Brazilian Census Bureau (IBGE) provides official population estimates for each municipality based on the 1990 and 2000 census and the 1996 and 2006 population counts. These data provide population estimates by sex and age group that are used in all the regressions to account for cohort sizes.

Municipality Controls

The study uses a rich set of municipality-level time-varying controls on the characteristics of the municipalities from a variety of sources. These include municipality GDP, and the fraction of municipality-level expenditure on education, health, welfare, transportation, and housing, provided by IBGE. In addition, the study uses information on the number of Bolsa Família¹⁵ recipients and the total amount of Bolsa Família payments in the municipality. These data are available annually for the 1997–2009 period.¹⁶ Details on the source of these data appear in the supplementary online appendix (table S1.2).

4. Empirical Strategy

The study is interested in the effect that an increase in municipal school availability has on the birthrate. The measure of school availability is referred to as secondary school density, denoted by |${S_{it}}$|⁠. This is calculated as the number of schools divided by the cohort size of females. Secondary school density, as with birth outcomes, is measured as the number schools per 100 females in the cohort.

Identification Strategy

Variation, to identify the intention-to-treat effect (ITT) of an increase in school density, comes from differences across municipalities in the within-municipality change in the number of secondary schools. To illustrate, the analysis starts with a stylized two-by-two difference-in-differences analysis (table 1). The study compares the change in the birthrate between 2000 and 2009 for municipalities that received at least one new school over this period (treated) and municipalities that did not receive a school over this period (control). Relative to the control municipalities, the treated municipalities saw an increase of 0.79 secondary schools per 100 students, roughly an 80 percent increase. While both control and treated municipalities experienced a decrease in birthrates between the two periods, the average decrease in treated municipalities is more accentuated with a difference of 0.65 births per 100 students enrolled. Attributing the difference in birthrate decrease to the difference in secondary schools increase, a unit change in school density decreases the teenage birthrate by 0.82 units (‒0.65/0.79). As will be shown in table 2, this basic calculation is similar to what is found in the full regression analysis

Clearly, this conclusion relies on the assumption that school introductions do not vary systematically with changes in unobservables across municipalities that influence birth outcomes.¹⁷ As an initial test, panel B of table 1 reports the results of a falsification exercise, using birth outcomes for an older cohort group. The cohort aged 29 in 2009 would have been 20 years old in 2000; that is, older than the secondary school target age group. It should be noted that this is an imperfect exercise; it cannot rule out the possibility that an older cohort was affected by the school introduction, as there is no age limit to enroll in secondary school, and there is frequent late enrolment into secondary schools. However, the study expects to see an effect that is considerably less than that of the younger target-age cohort. As with the younger cohort, the older cohort experiences a reduction in birthrates over the two periods. However, there is no birthrate difference between treatment and control municipalities over time.

Exogeneity of Brazil's Secondary School Expansion

The interpretation of the conditional correlation between birth rate and school density as an intention-to-treat effect requires the identifying assumption that school introductions are conditionally random with respect to births, consistent with the controlled experiment above. Section 2 argues that supply-demand factors largely determine the school expansion across municipalities and over time. To examine this empirically, the study estimates models of the between-municipality differences in the school expansion. The first step is investigating the probability a municipality is part of the expansion, for which the outcome is a binary indicator for receipt of a new school between the years 1998 and 2009. Then the timing of the introduction of schools across municipalities is investigated. For municipalities that receive a school, the outcome is an ordinal variable equal to the number of years from 1997 that a school expansion is realized.¹⁸ These two outcomes are regressed on 1997 municipal characteristics and a set of state dummy variables (table 2).

Consistent with the identifying assumption, the correlation between the teenage birthrate and both the introduction of a secondary school or the timing of the expansion is neither significant nor large in magnitude. Rather, it is found that supply-demand factors are important in explaining the school expansion. Municipalities with high secondary school density in 1997, which is interpreted as a supply factor, are less likely to receive a school in the preceding periods, and if they do receive a school, the expansion takes place later than in municipalities with low initial school density. The municipality's enrolment rate in the final year of primary school, which is interpreted as a demand factor, is positively correlated with how early the municipality receives a new secondary school, but does not determine the receipt of a school. This is consistent with a policy objective to equalize secondary school access across Brazilian municipalities, while giving priority to those municipalities that have a higher immediate demand (Soares 1998).

The outcome, |${B_{it}}$|⁠, is the cohort birth rate in municipality i in year t. The event for municipality i is the year in which a change in the number of secondary schools is first observed, |${e_i} \in [ {1997,\ 2009} ]$|⁠, and the indicator function |$1[ {t - \ {e_i} = \ d} ]$| takes a value of 1 when the difference between year|$\ t\ $|and |${e_i}$| is d and 0 otherwise. The parameters |${\lambda _d}$| reflect the average (demeaned) birthrate d periods away from the event, |$d\ = \ 0$|⁠, normalizing |${\lambda _{ - 1}} = \ 0$|⁠. |${\vartheta _i}$| reflects the mean birth rate for municipality i and within-municipality deviations from the mean are captured by |${\mu _{it}}$|⁠. Estimated values of |${\lambda _d}$|⁠, for |$d \in [ { - 4,\ 4} ]$| are plotted in fig. 4, for classroom density (4a) and the birthrate (4b) as outcomes (bars around point estimates reflect 95 percent confidence interval). From the resulting figures it is concluded that there are no systematic trends in municipal rates of teenage childbearing prior to the introduction of a secondary school.²⁰

Estimation Strategy

The outcome is the cohort birth rate for municipality i in year t, as defined in equation (1). The explanatory variable of interest, |${S_{it - 4}}$|⁠, is secondary school density in municipality i in year |$t - 4$| (when the cohort reflected in |${B_{it}}$| was 15).²¹|$P{E_{i,\ t - 5}}$| is the enrolment rate for students in the final-year primary classes at year |$t - 5$|⁠.²²|${{\boldsymbol{X}}_{{\boldsymbol{it}}}}$| is a vector of municipal controls including the number of preschools, total male-to-female sex ratio, total and per-recipient Bolsa Família transfers, and – both in logs and four-year log-differences–municipal GDP and public spending (welfare, education,²³ health, transportation, and housing). Unobservable heterogeneity is captured by a year trend, |$yea{r_t},{\rm{\ a\ state\ }} \times \ {\rm{year\ specific\ shock}},{\rm{\ }}\delta _t^s$|⁠, a municipal fixed effect, |${\eta _i}$|⁠, and a time-varying municipal component, |${\epsilon _{it}}$|⁠.

The resulting estimate, |$\hat{\alpha }$|⁠, reflects the intention-to-treat effect of a one unit increase in secondary school density on teenage birthrates. Municipal differences of the within-variation in the secondary school expansion between the years 1997 and 2005 identify this coefficient.²⁴ For an unbiased estimate of |$\alpha $| the studs assumes strict exogeneity between the error term and school introductions. This assumption is supported by the analysis in section 4. This assumption is subjected to further tests and specifications in sections 5 and 6. The remarkable stability of the results across specifications indicates that this assumption is plausible.

5. Results

Estimates for equation (3) are presented in table 3. The coefficient of interest, |$\alpha $|⁠, is reported in the first row [Secondary school density (t-4)]; estimates for selected control variables in |${{\boldsymbol{X}}_{{\boldsymbol{it}}}}$| and |$P{E_{i,\ t - 5}}$| are reported in the remaining rows. All regressions include a year trend and municipality fixed effects, municipality-clustered standard errors are reported in parenthesis.²⁵ For each specification, both the within-variation R-squared and between-variation R-squared are reported.

Column (1) shows results of the benchmark specification including all municipalities in the sample. Based on this estimate, for a unit increase in school density a decrease of 0.269 births per 100 is expected to appear. If the mechanism operated entirely through school enrolment, this would provide an intuitive sense of the magnitude to interpret this effect. An increase in school density leads to an average enrolment increase of 28.87 pupils per 100 cohort females.²⁶ Given this, for every 107 students enrolled a reduction of one birth (28.87/‒0.269) is expected to appear. Column (2) includes municipal spending information and state-specific year trends. The addition of control variables increases the within-variation explained by approximately 50 percent (a 6.5 percentage point increase). However, very little change is observed in the estimated value of |$\alpha $| relative to the specification in column (1).

In columns (3) to (5), the identification assumption is further weakened by restricting the sample to municipalities that are more similar. In column (3), the sample excludes 34 municipalities with populations greater than 500,000.²⁷ Estimates do not meaningfully change from the previous columns. In column (4), the sample is further restricted to the 2,550 municipalities with populations between 10,000 and 100,000. In column (5) the study further excludes from the sample all municipalities that did not receive a new school between 1997 and 2005. As these are the years from which the study's identifying variation comes, the final specification relies on the relatively weak identifying assumption that the timing of school introduction is strictly exogenous, rather than whether a school is introduced or not. The magnitude of the estimates increases in the final two specifications, indicating an effect of between ‒0.563 and ‒0.499. Column (5), which requires the weakest identifying assumptions, implies that there is one fewer teenage childbirth for every 48 students enrolled due to the increase in secondary school provision (23.8/‒0.499).

The stability of the estimates across these various specifications provides additional support for this study's identification strategy. If there were time-varying characteristics that both influence birth outcomes and are correlated with school introductions, then the estimates would be expected to change as municipalities are restricted to those that are more “similar” in terms of school introductions. Indeed, it is found that the estimates are remarkably consistent across the different samples, lending additional credibility to the identification strategy.

Heterogeneous effects on teenage births by municipality characteristics are also investigated. The study focuses on two sets of municipality characteristics: measures that reflect the state of the local economy (the ratio of wages for workers with at least secondary education, versus below-secondary education and the municipal unemployment rate) and measures of population density and dispersion (population per |$k{m^2}$| of land and proportion of rural students enrolled). Estimates, corresponding to the study's preferred specification of column (4) in table 3, are stratified by the median values for each of these variables (see table S3.3 in the supplementary online appendix). Substantial differences in estimates across the different municipal characteristics are not found. The effect of school density on birth rates is almost identical for municipalities in which the observed earnings for secondary education is above the median versus those below the median. There is a small difference for municipalities in which the local unemployment rate is high versus those in which it is low. Less dense municipalities (as measured by population per |$k{m^2}$| and rural student enrolment) also tend to have a larger effect.

6. Testing for Potential Threats to the Identification Strategy

The preliminary analysis in section 4 is consistent with the assumption of exogeneity of the secondary school expansion across municipalities. The stability of estimates across different subsamples of municipalities reinforces this finding. This section discusses additional threats to the identification strategy and how these are tested. The possibility is further explored that the expansion of secondary schools is correlated with other municipal-level programs not controlled for by the municipal spending variables that are included in equation (3).

Unobserved Municipality Characteristics

One concern for the causal interpretation of the estimates in table 3 is that there may be unobserved municipal characteristics that are correlated with teenage births and influence a municipality's receipt of a secondary school, leading to a spurious correlation between birth rates and school density. This possibility is investigated using two strategies. First, a binary indicator for the introduction of a secondary school in year t is regressed on the lag of birth rates, |${B_{it - 1}}$| (panel A, table 4). Second, a binary indicator for the introduction of a secondary school is regressed on the percent change in the number of births between year t and |$t - 4$|⁠, |${{B_{it}} - {B_{it - 4}}} /{B_{it - 4}}$| (reported in standard deviations, panel B, table 4). Regressions also include the lagged value of primary school enrolment, |$P{E_{i,\ t - 1}}$| and the vector of controls |${X_{it}}$|⁠, as well as state-time trends and municipal fixed effects. For display purposes, coefficients and standard errors are multiplied by 100.

These results from these regressions are consistent with school introductions being conditionally independent of birth rates. Correlations between municipal birthrates and school introductions are small and statistically insignificant.

Births to Older Mothers

As there is no explicit age restriction to enrolling into secondary education in Brazil, there is no age cut-off at which it is possible to say treatment is received or not. However, as the majority of secondary school students enroll between the ages of 15 and 17, the study expects the secondary school expansion to have less effect on the fertility of older women compared to younger women. Finding a qualitatively similar effect on births at ages 30 and older may indicate that unobservable factors are influencing municipal childbearing more generally.

To test this, the municipal births for women age 15 to 40 (⁠|$b_{it}^a$|⁠, |$a\ = \ 15,\ \ldots ,\ 40$|⁠) are regressed on the contemporaneous school introductions. Regressions include age-specific male and female sex-ratios, but otherwise contain the same control variables as in equation (3). The conditional correlations between contemporaneous school density and the percent change in age-specific births are reported in fig. 5. Estimates for all groups over 23 years of age are statistically insignificant and small in magnitude. The study interprets this as further evidence that there are no unobserved municipal changes that affect both childbearing and the introduction of secondary schools.

Selective Migration

Migration might lead researchers to estimate a negative value for |$\alpha $| if the introduction of a secondary school induces an outflow of sexually active teens, creating a spurious negative relationship between school introductions and teenage childbearing. Appendix S3 in the supplementary online appendix provides a detailed analysis of migration patterns in Brazil, using information from the 2010 Brazilian population census available publically through IPUMS International. These data make it possible to analyze migration patterns for a subset of the Brazilian municipalities (831 municipalities). The key findings here are summarized here.

Migration data from the population census demonstrates the following: a) relative to other migrant and nonmigrant women, women who migrate between ages 10 and 18 are significantly more likely to be teenage parents; b) migrants tend to leave municipalities with low school expansion in favor of municipalities with greater school expansion; c) the difference in school expansion between the destination and origin municipality is higher for teenage mothers than for non-teenage mothers. These stylized facts all work against finding a negative relationship between school density and teenage birthrates. As a result, migration may lead to an underestimate of the true magnitude. Therefore, the results reported in table 3 should be considered a lower bound on the parameter |$\alpha $|⁠.

7. Concluding Remarks

This article investigates whether an increase in secondary education access, through an increase in the municipal density of secondary schools, impacts the fertility decisions of teenagers in the middle-income context of Brazil. The study's estimates are consistent with the hypothesis that the school expansion played an important role for teenage fertility in Brazil. A one-unit increase in secondary school density in municipalities leads to an average decrease in a cohort's teenage birth rate of between 0.250 and 0.563 births per 100. These estimates suggest, under some assumptions, a reduction of one birth for roughly every 50 to 100 students who enroll in secondary education due to the expansion.

The data in this study have the advantage of near-universal coverage of births, and significant variation from a large school expansion across 4,884 municipalities. However, a limitation of these data is that it is not possible to link birth outcomes to schools at an individual level. Therefore, results should not be interpreted as capturing the effect of school enrolment on an individual's outcome. Rather, the intention-to-treat effect is measured, reflecting the availability of secondary education opportunities in a municipality.

The findings in this paper also contribute to a literature that looks at the impact that expanding education access has in developing communities. Several studies have previously looked at the quasi-random variation from the primary school expansion in Indonesia. This expansion is found to have a substantial impact on education and adult wages (Duflo 2001, 2004). In addition to labor market outcomes, women have fewer children as adults (Akresh, Halim, and Kleemans 2018). Providing evidence from an RCT, Burde and Linden (2013) find that introducing a primary school in Afghan villages significantly increases academic participation and performance for both boys and girls. However, the gains for girls are disproportionately large, significantly reducing the enrolment and the performance gap between boys and girls. An analysis of Brazil over the same period as this study’s finds a strong association between teenage fertility and income inequality within a municipality (Chiavegatto and Kawachi 2015), and an inverse relationship between regional teenage pregnancy and regional measures of the human development index (Ferreira Vaz et al. 2016). The present study's analysis provides insight into the relative importance of access to secondary education in these relationships and adds to this observational evidence by providing evidence on a pathway through which school access may change gender-based inequalities in a municipality. If secondary access affects both fertility and social mobility, the school expansion documented here may lead to fundamental changes in income inequality in the longer-run.

The between-state variation in teenage pregnancy rates (fig. 2) may be also reflect broader differences in economic and social issues. Kearney and Levine (2012) argue that the substantial geographic variation in U.S. teen pregnancy rates reflects geographic differences in economic opportunity – both perceived and real. Being on a low economic life trajectory leads teenage girls to have children. In the context of the current study, by decreasing the geographic distance necessary to attend secondary school, the Brazilian expansion improved opportunities for secondary school attendance. In this sense, the study's estimated negative relationship between schools and teenage childbearing is consistent with Kearney and Levine's findings and improving access to secondary schooling may improve the (perceived) life trajectory for young women.

Footnotes

Data availability statement

The data and code underlying this article can be accessed at shorturl.at/coGY9.

Notes

Martin Foureaux Koppensteiner is a Senior Lecturer at the School of Economics, University of Surrey, Guildford, UK, and a Research Fellow at IZA, Bonn, Germany. His email address is [email protected]. Jesse Matheson (corresponding author) is a Senior Lecturer at the Department of Economics, University of Sheffield, Sheffield, UK. His email address is [email protected]. The authors received valuable comments on earlier drafts from Donna Feir, Krzysiek Karbownik, Melissa Kearney, Courtney Ward, and Nicolas Ziebarth. Thanks go to seminar and conference participants at the Essen Health Conference, the University of Pernambuco Recife, the Cornell Institute on Health Economics, Health Behaviors and Disparities, the University of Sheffield, the European Society for Population Economics, the Royal Economics Society, and the Canadian Health Economics Study Group for very useful feedback. Thanks also go to Supriya Toor for her excellent research assistance. This is substantially revised version of a paper previously circulated with the title ‘Secondary School Enrolment and Teenage Childbearing: Evidence from Brazilian Municipalities.’ (Foureaux Koppensteiner and Matheson 2019) A supplementary online appendix is available with this article at the World Bank Economic Review website.

References