ABSTRACT
We examine the impact of raising the minimum age of marriage to 18 years old in Mexico. Using a difference-in-differences model that takes advantage of the staggered adoption of this reform across states, we find a large reduction in the number of registered child marriages. However, we find no effect on school attendance or early fertility rates. We provide evidence that this is driven by a substitution of formal marriage for informal unions. This suggests that when informal unions are a viable option for young couples, age-of-marriage reforms are not enough to prevent early unions and their negative consequences.
I. Introduction
Approximately 650 million girls and women alive today were married before the age of 18, where marriage includes both formal marriages and informal unions in which partners cohabitate as if married (UNICEF 2018). Eradicating this practice is key in the fight to reduce global poverty, as child marriage leads to lower educational attainment, earlier age at first childbirth, higher fertility, higher infant mortality, and worse health and educational outcomes for the children born into child marriages (Field and Ambrus 2008; Sekhri and Debnath 2014; Chari et al. 2017; García-Hombrados 2022). Although most countries’ laws set the minimum age of marriage at 18, they typically provide exceptions upon parental consent, pregnancy, authorization from the courts, or due to religious or customary laws, making child marriage legal in practice (Arthur et al. 2018).1
A common proposal for ending child marriage is to eliminate all exceptions to the minimum age of marriage.2 As long as there is adequate enforcement, these reforms should reduce or even eradicate formal child marriages. But the enforcement of minimum-age-of-marriage laws may be weak if these laws are incompatible with prevailing social norms and practices (Acemoglu and Jackson 2017).3 Perhaps more importantly, raising the minimum age of marriage may not be effective in societies where informal unions are a viable option for young couples, as a drop in the number of formal marriages may be offset by an equal rise in the number of informal unions, leaving child marriage rates unchanged. Banning child marriage could even have negative welfare effects if people in informal unions do not have the same legal benefits or social recognition as those who are formally married. Alternatively, if laws banning child marriage have an expressive function (Benabou and Tirole 2011), they may change social norms, reducing the incidence of both formal and informal child marriages.4
We provide empirical evidence on the impact of raising the minimum age of marriage using a natural experiment in Mexico. Researching the impact of changes in minimum-age-of-marriage laws is challenging because these policies are commonly set at the national level, providing no counterfactual for credibly identifying their causal impact. Mexico provides a compelling case study for various reasons. Between 2008 and 2018, most states in Mexico increased their minimum age of marriage. These reforms occurred at different points in time, allowing us to exploit variation across states and over time using a two-way fixed effects difference-in-differences model. Another advantage of the Mexican context is the availability of granular data on marriages, births, and school attendance. Finally, Mexico ranks seventh in terms of the number of women who were child brides, so studying the impact of changes in age-of-marriage laws in this context is important in and of itself.5
We first examine the extent to which the reform was enforced. Using microdata from marriage certificates, we find that banning child marriage leads to a large reduction in the rate of formal child marriage, particularly for 16- and 17-year-olds, the age group with the highest rate of child marriage before the onset of the prohibition. We find a 49 percent reduction in the formal child marriage rates for girls of this age.6 Importantly, we show that these estimates are not biased by different pre-trends in states that enacted the ban on child marriage, by couples marrying in states in which child marriage was still legal, or by misreporting of age in marriage certificates.
After establishing that the reform led to a large and statistically significant reduction in formal child marriage rates, we study whether banning child marriage reduces school dropout and early motherhood, two important and detrimental consequences of early unions (Field and Ambrus 2008; Chari et al. 2017). We present a simple conceptual framework that illustrates that in societies where informal unions carry a reputation penalty, banning child marriage will reduce fertility and increase school attendance. We use an event-study specification to examine this question empirically, since the impact of the reform could change over time, biasing our difference-in-difference estimates (Goodman-Bacon 2021; de Chaisemartin and D’Haultfœuille 2020). Using data on school attendance from the Mexican labor force survey and birth registry microdata, we find that the reform had no effect on school attendance and early motherhood. We explore the mechanism behind this result using data on the civil status of young mothers at the moment of delivery. We find that for births where the mother is younger than 18, banning child marriage led to a drop in the share of married mothers and an equivalent rise in the share of mothers in an informal union.7,8 This shows that there are minimal social sanctions for informal unions in Mexico. This change in civil status could potentially have negative effects for young mothers and their children if girls in informal unions do not have the same legal rights or spousal support as those that are married. We examine whether the reform had an impact on prenatal investments and newborn health outcomes, and find no effects. These findings suggest that in places where cohabitation is socially acceptable, minimum-age-of-marriage laws are ineffective at avoiding the detrimental consequences of early unions.
This work is related to the research studying the determinants of early marriage and the impact of policies aimed at reducing this practice. Corno, Hildebrandt, and Voena (2020) study how aggregate economic conditions affect child marriage rates in Sub-Saharan Africa and India, where marriage payments are a source of consumption smoothing. They find that negative shocks lead to more child marriages in societies with bride price, and less in places with norms of dowry. Jensen (2012) finds that providing job recruitment services to young women in rural India raises their probability of working and their investments in schooling or training and reduces their likelihood of marrying over the study period. Baird, McIntosh, and Özler (2011) study the impact of a program granting cash transfers in Malawi and find that after two years, unconditional cash transfers reduce child marriages and delay fertility, whereas cash transfers conditional on attending secondary school have no impact. Buchmann et al. (2018) study a program in Bangladesh that provides girls with empowerment training, in-kind transfers conditional on delaying marriage until the age of 18, or both. While financial incentives led to a sizable reduction in child marriages, school dropout, and teenage childbearing, the empowerment treatment had no effect on child marriage, and there was no complementarity between the two treatments. Finally, Bandiera et al. (2020) find that providing vocational and empowerment training to adolescent girls in Uganda increases the probability of working and reduces the likelihood of marriage or cohabitation and teenage pregnancy.
There are a few studies in this literature focusing on age-of-marriage laws. García-Hombrados (2022) studies the effect of raising the legal age of marriage in Ethiopia. He shows that the reform was effective at reducing child marriages and that the resulting delay in the age of cohabitation decreased infant mortality. In the closest work to ours, Bharadwaj (2015) studies the impact of a 1957 reform in Mississippi that increased the minimum age of marriage from 12 to 15 for women and from 14 to 17 for men, introduced parental consent requirements for individuals below the age of 18, and implemented a compulsory three-day waiting period and blood tests. Using a difference-in-differences strategy, the author compares counties in Mississippi to those in neighboring states and finds that three years after the law change, there was a large decrease in the overall marriage rate, a drop in overall birth rates, and a rise in school enrollment rates. While cohabitation was extremely rare in the United States in the late 1950s (Lundberg, Pollak, and Stearns 2016), informal unions and unregistered marriages are relatively common nowadays.9 The question of how increases in the minimum age of marriage affect child marriage rates should be revisited in a context where informal unions are common, as they could undermine the effectiveness of legal prohibitions. The main contribution of our study is that it shows that banning child marriage is not effective in places where formal marriage is not the only option for young couples. Our work is also related to Collin and Talbot (2019). This study uses a large sample of developing countries to test whether there is a discontinuity in the age-of-marriage distribution at the legal minimum age, where marriage includes formal and informal unions. The authors find a statistically significant discontinuity in around half of the cases and attribute this to a weak enforcement of age-of-marriage laws.10 As the data used in Collin and Talbot (2019) do not allow them to distinguish between formal and informal unions, it is unclear if age-of-marriage laws do not bind due to lack of enforcement or because minors enter informal unions when they are barred from getting legally married. In our setting, we can pin down the mechanism behind this null effect. In particular, we are the first to find that in a context where informal unions are socially acceptable, raising the minimum age of marriage leads young couples to substitute marriage for informal unions. Ours is also the first study examining the issue of child marriage in Latin America, a developing region that is currently home to almost 10 percent of the world’s child brides (UNICEF 2018). Reducing the incidence of child marriage in this region is especially important because, despite the rapid worldwide decrease in child marriages, child marriage rates have remained constant over the last 25 years in Latin America.
Next, Section II provides background on child marriage in Mexico, and Section III discusses the potential impact of banning child marriage. Section IV describes the data and provides summary statistics, and Section V discusses our estimation strategy. Section VI presents our results, and Section VII discusses their implications. Section VIII provides evidence from several validity and robustness checks, and Section IX concludes.
II. Child Marriage in Mexico
Child marriage is commonly defined as a union in which at least one of the parties is below the age of 18 (UNICEF 2018) and includes formal marriage and cohabitation as if married. Throughout the paper, we refer to formal marriages as formal or registered marriages and to cohabitation as if married as informal unions. At the start of our study period in 2008, all Mexican states allowed persons younger than 18 to get formally married, albeit with some restrictions. Minors needed to have a certain age and the consent of their parents or guardians. The minimum age of marriage with parental consent varied across states. For example, the threshold was set at 14 in the states of Chihuahua and Durango and 16 in Chiapas and Baja California Sur. In addition to the consent of the parents or guardians, a few states required that minors have the authorization of a judge or the mayor of their municipality. Those who were younger than the age of marriage with parental consent or did not have this consent could only get married with the permission of a judge or the municipal mayor. A few states also allowed girls who were younger than the minimum age to get married if they were pregnant.
Between 2008 and 2014, some states eliminated all exceptions for marriage below the age of 16, but still allowed those 16–17 years old to get married, as shown in Figure 1. In December of 2014, the Federal Congress sanctioned a law defining the rights of children and adolescents. This law set the minimum age for marriage for both women and men at 18, without exception, and urged all federal entities to reform their legislation to incorporate this change. Since marriage laws are a competency of the states, it is the prerogative of state congresses whether to adopt the reform or not.11 As summarized in Online Appendix Table A.1, the adoption of the reform was gradual. By the end of 2015, only eight states had changed their marriage laws in accordance to federal legislation. By December of 2018, the end of our study period, the reform was adopted by 30 of the 32 Mexican states.12
Progressive Adoption of the Child Marriage Ban
Notes: This figure shows the evolution in the number of states that banned child marriages (solid line) and banned child marriages only for minors below the age of 16 (dashed line). We obtained the date of the reform for each state from the states’ civil and family codes.
Child marriage includes informal unions as well as formal marriages. Informal child marriages may be punished by law if one of the partners is below the age of consent. Mexican states have a “hard” and a “soft” age of consent. The hard age of consent is typically 12 or 14, and intercourse with a person that is younger than this age is considered rape. The soft age of consent is 18 in most states, and intercourse with individuals who are above this age is not punished by law. There is a legal gray area between the two ages of consent in which the crime of estupro may apply. An adult commits estupro by having intercourse through seduction or deceit with an adolescent between these two ages of consent. Although the law is vague about what constitutes seduction or deceit, the typical example is promising the minor that a marriage will occur and then reneging on this promise. While estupro is punishable with jail time, this crime can only be prosecuted if the minor or their legal guardians present charges, and this is quite uncommon. In 2016, for example, only around 1,000 cases were brought forward for estupro in all of Mexico (UNICEF 2019).
From a legal standpoint, marriage and informal unions are similar but not equivalent institutions. Informal unions only generate obligations and rights while the union lasts. Similarly to married couples, partners under informal unions are entitled to life insurance, inheritance, social security coverage, and maternal benefits.13 Marriage, on the other hand, generates obligations and rights even after the divorce, for example, allowing one of the spouses to claim a pension from the other spouse, something that does not happen in the case of an informal union. Informal unions are protected by law when there is no legal impediment for marriage, and the couple has been cohabitating for at least two years or has a child. This means that the rise in the minimum age of marriage not only prevents minors from getting married, it also impedes them from being in a legally recognized informal union.
To understand the scope of the change in legislation and its potential to reduce child marriage rates, it is important to analyze the prevalence of this practice before the reform was introduced. Figure 2 shows the evolution in the incidence of child marriage for the cohorts that turned 18 before the reform. We obtained these data from the Encuesta Nacional de la Dinámica Demográfica (ENADID), a nationally representative demographic survey conducted in 2014 that has detailed information on relationship, schooling, and fertility history. Our sample includes almost 84,000 women who were 20–54 years old at the time of the survey. Around 23 percent of respondents were formally married or in an informal union before turning 18, and this percentage is relatively constant across cohorts. Although overall child marriage rates in Mexico have not varied over the last decades, there has been a significant change in the type of union. While formal marriages accounted for approximately 75 percent of all child marriages in the older cohorts, as seen in Figure 2, less than one-third of the child marriages of women born in the early 1990s were formal unions. Right before the ban on child marriage, approximately 6 percent of Mexican women got formally married before turning 18.
Evolution in the Share of Women Who Were Child Brides
Notes: This graph depicts the share of women who were formally married and in an informal union before the age of 18, by their birth year. These two categories are mutually exclusive. Women in an informal union are those who had an informal union before the age of18 but did not get formally married by this age. The source of these data is the Encuesta Nacional de la Dinámica Demográfica (ENADID), a demographic survey conducted in 2014. We took the sample of 83,554 women who were 20–54 at the time of the survey and computed the share of women in each category using sampling weights.
Despite the declining trend in formal child marriages, this practice was far from being eradicated before its ban. Using microdata from marriage certificates, Figure 3 shows the number of marriages in 2013 by the age of the bride, for every 1,000 girls and women of each age. Formal marriage rates were highest for women in their 20s, although child marriage was relatively frequent as well. There were 40,298 child marriages in total, accounting for almost 7 percent of total marriages. Most child marriages had a bride aged 14–17, with the largest share being 16-and 17-year-olds, who got formally married at a rate similar to that of women in their early 30s. For every 1,000 girls ages 16–17, 16.08 got formally married in 2013. In contrast, there were only 2.30 marriages with a bride ages 14–15 for every 1,000 girls of this age. In 2008, before some states enacted a specific ban for this age group, the marriage rate for girls ages 14–15 was slightly larger, with 5.65 marriages for every 1,000 girls of this age group. There were very few marriages with a groom below the age of 18, as shown in Online Appendix Figure A.3, because child marriages had a groom that was 4.62 years older than the bride, on average.
Number of Registered Marriages per 1,000 Girls and Women of Each Age in 2013
Notes: This figure depicts the number of legally registered marriages by the age of the bride, per 1,000 women and girls of each age in 2013. The number of marriages was obtained from marriage certificate microdata and the population from CONAPO.
A. Comparing Child Brides to Nonchild Brides
Having established that child marriage was relatively common before the reform, we now compare the baseline characteristics of women who were child brides to those who were not. Since there is no longitudinal survey following women from childhood to marriage, we use the Encuesta Demográfica Retrospectiva (EDER), a nationally representative survey conducted in 2017 to collect retrospective information on a wide array of demographic and socioeconomic characteristics. To minimize measurement error from long-term recall, we limit our sample to the 4,438 women who were 24–34 at the time of the survey (that is, 20–30 in 2013). We then divide them into three groups according to their civil status in the period before age 18. We compare women who were formally married, in an informal union without getting married, and single before the age of 18. Online Appendix Table A.2 shows that, on average, women who were formally married or in an informal union before the age of 18 come from families of a lower socioeconomic status than those who were single. They are more likely to belong to an indigenous group, have parents with low educational attainment, and to live in a house of low structural quality with few durable assets when they were 14 years old.14 At the age of 14, the women who eventually became child brides were less likely to be attending the appropriate grade level at school and had a higher probability of being dropouts.
These baseline differences between women who were child brides and those who were not are amplified with time. Online Appendix Figure A.4 shows the maximum educational attainment for women of different cohorts, splitting the sample by whether they were child brides or not. We obtained these data from the ENADID survey and use the same sample as the one used to analyze the evolution of child marriage.15 In the cohorts born in the early 1990s, only 15 percent of the women who had been child brides (either through a formal marriage or informal union) have a secondary school degree or more, compared to 65 percent of the women who were not child brides.16 There are stark differences in the rates of early motherhood in both groups as well. As shown in Online Appendix Figure A.5, 88 percent of the women who were married or in an informal union before the age of 18 gave birth before turning 20, whereas only 16 percent of the women who were not child brides had a child before this age.
III. Conceptual Framework
Child marriages in Mexico can be separated into two main categories: marriages that occur after the couple engages in premarital sex and the girl gets pregnant and marriages that occur for reasons unrelated to pregnancy (for example, love, opportunity for a better life). We explore the reasons behind child marriage in Mexico using data from the 2014 ENADID survey. We limit our sample to the 11,317 women who got formally married before the age of 18 and were 20–54 years old at the time of the survey. We find that 38 percent of these women conceived their first child before they got married, and this share is relatively constant across cohorts.17 These figures show that in most cases, formal marriage is not the result of a pregnancy.
When the laws banning child marriage are introduced, formal marriage is no longer an option. In Online Appendix B, we develop a simple theoretical framework to explain the effects of banning child marriage on fertility and school attendance. In this model, the impact of banning child marriage depends on the reputation cost of being in an informal union. If there are no social sanctions for informal unions, banning child marriage should produce a complete substitution from marriage to informal unions, leaving fertility and school attendance unaffected. If, on the other hand, informal unions carry a reputation penalty, banning child marriage leads to a reduction in fertility and school dropout, and this reduction is increasing in the fraction of child marriages that occur for reasons unrelated to pregnancy.
As the impact of banning child marriage likely depends on the social norms around informal unions, we use data from the World Value Survey to compare social norms in Mexico to other countries. There are 77 countries in total, and 62 if we exclude the United States, Canada, Australia, New Zealand, and Western Europe. Figure 4 shows the share of respondents in each country who answered that they would not like to live next to an unmarried couple. Only 13 percent of Mexican respondents preferred not to have an unmarried couple as their neighbor, placing Mexico in the 42nd percentile overall and in the 29th percentile in the subsample of developing countries. As compared to other developing countries in Asia or Africa, informal unions are socially accepted in Mexico. Mexico is not an outlier, however, as its norms around informal unions are comparable to other Latin American countries with similarly high rates of child marriage, such as Brazil, Uruguay, and Peru. One caveat is that this question does not specifically ask about informal unions by minors, and opinions about informal unions of underage persons may differ. In a nationally representative survey conducted in Mexico in 2012 (Encuesta Nacional sobre Política y Prácticas Ciudadanas), individuals were asked whether they would be accepting of certain actions by their hypothetical teenage son or daughter (15–18-years old). Two-thirds of the respondents age 40 or older agree with their teenage children moving in with their partners. All in all, the reputation cost for being in an informal union is small in Mexico.
Cross-Country Comparison of the Norms around Informal Unions
Notes: The data for this figure come from the fifth and sixth waves of the World Value Survey. For countries that participated in both waves, we use the most recent data. This graph displays the share of respondents ages 40 and older in each country who answered that they would not like to live next to an unmarried couple.
IV. Data and Descriptive Statistics
To calculate the number of formal marriages with a bride younger than the age of 18, we relied on marriage certificate microdata from 2008–2018 provided by INEGI, the Mexican statistical institute.18 This database contains all the legally registered marriages conducted in Mexico and specifies the date and state in which the marriage took place, the age of the bride and groom, their state of residence, level of education, and occupation. During this period there were approximately six million marriages, of which 6 percent had a bride younger than 18 years old.19 Since most registered child marriages in Mexico have a bride who is younger than 18 years old but a groom who is 18 or older, we focus on girls.20 Using these data, we constructed a monthly panel for every state with the number of formal marriages with a bride ages 14–17. Even though there are marriages in which the bride is 12 or 13, they account for less than 0.6 percent of the child marriages in our sample. To abstract from potential spillovers to states in which child marriage was still legal, we use the brides’ state of residence. In Section VIII.A we show that our results are not sensitive to this choice.
We obtained data on live births in 2008–2018 from the Ministry of Health birth microdata. These data are derived from SINAC, a system created at the end of 2007 to obtain timely and detailed birth data. Hospital staff input information on the mother and newborn into the system on the day of the birth and provide the parents with a document generated by SINAC that is required for obtaining their child’s birth certificate later.21 This data set reports the date and state where the birth occurred, the length of gestation, the child’s birth order, the number and timing of prenatal care visits, the newborn’s birthweight, and the mother’s birth date, state of residence, and civil status. There were almost 23 million live births in 2008–2018, of which 9 percent had a mother below the age of 18. Using these data, we constructed a monthly panel for every state with the number of live births, using the mothers’ state of residence. Since the database has information on the date of birth and the length of gestation, we use the date of conception, as the reform could only affect fertility up until this moment. We restrict our sample to women who were 14–17 years old at the moment of conception.22 We drop 0.32 percent of observations for which the mother’s age is missing and keep only one observation for multiple births. Since we have information on births that occurred in 2008–2018, but we conduct our analysis at the moment of conception, our sample is composed of all births conceived between January 2008 and March of 2018 by women who were 14–17 years old at the moment of conception.
To calculate the formal child marriage rate and the fertility rate of young mothers, we also need information on the population of each state, by age and gender. We obtained biannual population data for 2008–2018 from the Consejo Nacional de Población (CONAPO).23 Our analysis focuses on girls ages 14–17 and splits them up into those who are 14–15 and 16–17, as the timing of the marriage ban differs for these two age groups in some states. Following standard definitions, we define the monthly child marriage rate of each age group in each state as the number of formal marriages with a bride of this age, for every 1,000 girls of this age group living in this state. Similarly, we define the monthly rate of early motherhood as the number of live births conceived in that month by a mother of this age, for every 1,000 girls of this age living in that state.
To analyze school attendance, we relied on the 2008–2018 waves of the Encuesta Nacional de Ocupación y Empleo (ENOE), a quarterly labor force survey with self-reported information on school attendance.24 We limit our sample to girls who were 14–17 years old at the moment of the survey. On average, each survey round interviews around 15,000 girls in this age group. We also obtain time-varying control variables from several sources. We created a panel with the political party of the governor in all states for 2008–2018 using data from miscellaneous sources, to account for the fact that the enactment of the child marriage prohibition might depend on the party in power. We also put together a monthly panel with several economic indicators at the state level. The unemployment rate was obtained from INEGI, the poverty rate and average income of employed individuals from CONEVAL, and the female labor force participation of women ages 20 and older from ENOE, the Mexican labor force survey.25 We also obtained the total population from CONAPO.
Descriptive Statistics
Our main independent variable throughout the analysis is a dummy for whether formal child marriage was prohibited in a given state, month, and year.26 In some states, this variable differs for girls ages 14–15 and 16–17. We went through the civil and family codes of each state to find out the date when marriage was banned for each of these ages. We only consider that child marriage is banned if the legislation allows no exceptions. For the federal entities that banned child marriage, we obtained the date when the articles that establish the minimum age for marriage were modified. Online Appendix Table A.1 summarizes this information for each state. Table 1 presents summary statistics of the variables used in our regressions.
V. Estimation Strategy
We use a two-way fixed-effects difference-in-differences model, which exploits variation in the enactment of laws banning child marriage across states and over time. To estimate the impact of the reform on registered child marriages, we use the following specification:
(1)
where Yst is the number of marriages in month–year t with a bride ages 14–15 or 16–17 living in state s, for every 1,000 girls of this age. We also estimate separate regressions for each age between 14 and 17. Our main explanatory variable, Child marriage bannedst, is a dummy variable equal to one if girls of the corresponding age group were not allowed to get married in state s in month–year t, and zero if they were. Xst is a set of state-specific controls measured in period t, namely the unemployment rate, poverty rate, labor force participation of females ages 20 and older, average income of employed people, population (in ln), and dummy variables for whether the state governor belongs to PRI, PAN, or PRD. Prior to the ban on child marriage, some states started requiring minors to obtain the authorization of a judge or the mayor of their municipality to get married. We control for whether this requirement was in place. We include state fixed effects (γs) to control for the time-invariant characteristics of states that affect marriage decisions and may also be correlated with the occurrence and timing of the child marriage prohibition. The month–year fixed effects (γt) control for the trend and seasonality in child marriages common to all states. Finally, Ust are the unobserved factors affecting child marriage rates in state s and period t, such as religious preferences and social attitudes. We allow for arbitrary within-state correlation of the errors by clustering our standard errors at the state level (Bertrand, Duflo, and Mullainathan 2004). Since we only have 32 clusters, we report wild-bootstrap p-values following Cameron, Gelbach, and Miller (2008).
As depicted in Online Appendix Figure A.6, there was substantial heterogeneity in the formal child marriage rate across states before the national push towards banning this practice. Importantly, our state fixed effects γs control for these and any other time-invariant differences across states. Our assumption for identifying the causal effect of banning child marriage is that conditional on state fixed effects, time fixed effects, and controls, the timing of the child marriage ban is uncorrelated with the error term. This assumption would be violated if the first states to ban child marriage were those in which child marriage rates were declining at a lower or higher rate. This could occur, for instance, if the prohibition of child marriage was driven by changes in religious preferences or social attitudes. We report the results of several tests validating our identification strategy in Sections VI and VIII.
In a two-way fixed-effects difference-in-differences estimation as in Equation 1, the estimate is a weighted average of the average treatment effects obtained from all possible two-by-two difference-in-difference estimators in the data, where the weights are proportional to the group sizes and the treatment variance within each pair (Goodman-Bacon 2021; de Chaisemartin and D’Haultfœuille 2020). Some two-by-two difference-in-difference estimators could have negative weights, potentially leading to large biases if treatment effects are heterogeneous (Abraham and Sun 2021; de Chaisemartin and D’Haultfœuille 2020). A concern in our setting is the possibility that the effect of banning child marriage varies over states or over time. This could occur, for example, if there is a lag between the enactment of the law and its implementation, or if the impact of the reform on school attendance and early motherhood takes some time to materialize. Importantly, only a small share of the two-by-two difference-in-difference estimates in our setting receive negative weights, mitigating these concerns.27 Since there is no reason why the impact of banning child marriage on fertility or school attendance should remain constant over time, most of our estimations are conducted using the following two-way fixed-effects event-study specification, which also allows us to test for differential pre-trends:
(2)
where Child marriage banneds(t+j) is a dummy variable taking the value of one j months relative to the month in which the reform was enacted in state s, in states that banned child marriage, and zero in all other months and states. As in Equation 1, we run separate regressions for girls ages 14–15 and 16–17, as some states banned child marriage for younger girls first. Following common practice, we exclude Child marriage banneds(t-1), thus normalizing relative to the month before the reform was put in place. Since some states implemented the reform towards the end of our sample window, as can be seen in Online Appendix Figure 1, the estimates of the longer lags are only estimated using early-adopters, and could be contaminated by sample composition changes. We discuss this issue when interpreting the results from these estimations in Section VI. The dependent variable in our fertility regressions is the number of first live births conceived in month–year t by a mother ages 14–15 or 16–17 living in state s, for every 1,000 girls of this age. In our main specification, we focus on first births instead of all births because, if any, the reform should have an impact on the extensive margin. Approximately 81.4 percent of the births in our sample with a mother younger than 18 at the time of conception are first births. We also report results using all births. Taking advantage of the fact that the birth registration data include the length of gestation, we conduct our analysis at the moment of conception, as the reform could only affect fertility up to this point.
We also study the impact of the reform on school attendance using a similar event-study specification. Unlike our analysis for registered marriages and fertility, we have a repeated (quarterly) cross-section of individual observations. We thus run the following regression:
(3)
where Yist is a dummy variable equal to one if person i living in state s is attending school in quarter–year t. We perform this analysis for girls who are 14–15 and 16–17 years old at the time of the survey. Our main regressors, Child marriage banneds(t+j), are dummy variables equal to one j quarters relative to the quarter in which child marriage was banned in state s for the relevant age group. Our vector of controls Xist includes the standard controls, as well as dummies for age and town size. We use the sampling weights from the ENOE survey in these estimations.
VI. Results
A. Formal Child Marriages
Table 2 shows the results of our difference-in-differences estimations on the impact of raising the minimum age of marriage on formal child marriage rates. In our preferred specification, displayed in Column 4, we control for month–year fixed effects, state fixed effects, and time-varying state characteristics. In the case of marriages with a bride ages 14–15, the estimates are negative, although they are not statistically significant. When focusing on girls ages 16–17, we find that outlawing child marriage results in 0.695 fewer formal marriages per month for every 1,000 girls of this age, a 49 percent reduction over the mean. The estimate is significant at the 1 percent level based on standard errors clustered by state (in parentheses) and wild-bootstrap p-values (in brackets). The legislation change does not cause a larger reduction in formal child marriages because child marriage rates were dropping in all states before the reform.28 Back-of-the-envelope calculations reveal that the law change averted approximately 50,000 formal child marriages. Column 5 reports the results of regressions weighting by the female population of the relevant age group, and the estimates are almost unchanged.
Effect of Banning Child Marriage on Formal Child Marriage Rates
To understand the timing of the effects we use the event-study specification in Equation 2. We plot the estimated coefficients and their 95 percent confidence intervals based on wild-bootstrap clustered standard errors for each month in the year before child marriage was banned, and 24 months after. We only report estimates for the first 24 lags because at the end of our sample period there were few states for which more than 24 months had passed since the reform.29 Even with this restriction, the coefficient for the longest lag is estimated using only 26 of the 31 states that banned child marriage for 14- and 15-year-olds and only 22 of the 30 states that banned it for all minors. The estimates of the longest lags must therefore be viewed with caution due to differences in sample composition. Figure 5 presents the event-study estimates for 16- and 17-year-olds. The drop in child marriage rates is realized right when the ban occurs and persists at similar levels for the following two years, indicating that there were few delays in the implementation of the reform. The impact in Month 0 (that is, the month in which the law was changed) is lower than in following months because the law changes were often conducted towards the end of the month. Importantly, the states that banned child marriage do not exhibit differential pre-trends in formal child marriage rates up to 12 months before the prohibition, as the lead coefficients are small and statistically indistinguishable from zero. We also report the results of these estimates in Table 3. For ease of interpretation, we grouped the estimates into four-month periods.
Event-Study Estimates of the Effect of Banning Child Marriage on Formal Marriage Rates of 16- and 17-Year-Old Girls
Notes: The sample includes all Mexican states in 2008–2018, and the unit of observation is a month–state. The figure plots the coefficients and 95 percent confidence intervals based on wild-bootstrap clustered standard errors of a regression in which the dependent variable is the monthly number of marriages with a bride ages 16–17 per 1,000 girls of this age who reside in state s in month–year t. The regressors of interest are dummy variables for each month relative to the period in which child marriage was banned for this age group in state s, with the month before the reform being the omitted category. We only report 12 lead coefficients for ease of interpretation and 24 lag coefficients because there are few states for which more than 24 months have passed since the reform was enacted. The regression also includes state fixed effects, month–year fixed effects, states’ unemployment rate, poverty rate, labor force participation of females ages 20 and older, average income of employed individuals, population (in ln), dummy variables for whether the state governor belongs to PRI, PAN, or PRD, and a dummy for whether girls of this age required the authorization of a judge/mayor to get married. Standard errors are clustered at the state level.
Event-Study Estimates of the Effect of Banning Child Marriage on Formal Child Marriage Rates
Figure 6 plots the estimates of this event study for 14- and 15-year-olds. The point estimates are quite similar to those in the difference-in-differences estimation reported in Table 2, but they are more precise, particularly in the first year after the reform. Our events-study estimates show that banning child marriage for this age group led to an average reduction of approximately 0.109 marriages per month for every 1,000 girls of this age group. Since child marriage was uncommon at these ages, these point estimates are six times smaller than those of girls ages 16–17. For exposition purposes, the remainder of the paper will focus on the impacts of the reform for girls ages 16–17, since the reduction in child marriages is mostly driven by this age group. We include results on the impact of banning child marriage for 14- and 15-year-olds in the Online Appendix.
Event-Study Estimates of the Effect of Banning Child Marriage on Formal Marriage Rates of 14- and 15-Year-Old Girls
Notes: The sample includes all Mexican states in 2008–2018, and the unit of observation is a month-state. The figure plots the coefficients and 95 percent confidence intervals based on wild-bootstrap clustered standard errors of a regression in which the dependent variable is the monthly number of marriages with a bride ages 14–15 per 1,000 girls of this age who reside in state s in month–year t. The regressors of interest are dummy variables for each month relative to the period in which child marriage was banned for this age group in state s, with the month before the reform being the omitted category. We only report 12 lead coefficients for ease of interpretation and 24 lag coefficients because there are few states for which more than 24 months have passed since the reform was enacted. The regression also includes state fixed effects, month–year fixed effects, states’ unemployment rate, poverty rate, labor force participation of females ages 20 and older, average income of employed individuals, population (in ln), dummy variables for whether the state governor belongs to PRI, PAN, or PRD, and a dummy for whether girls of this age required the authorization of a judge/mayor to get married. Standard errors are clustered at the state level.
After establishing that banning child marriage leads to a reduction in the rate of registered marriages, particularly for 16- and 17-year-old girls, we examine whether the affected cohorts get formally married once they turn 18 or delay marriage even longer. If the 16–17-year-old girls who would have gotten married in the absence of the prohibition get married once they turn 18, we should see an increase in the marriage rate of 18-year-olds shortly after the ban (that is, once the affected cohorts start turning 18). Given the dynamic nature of these potential effects, the most appropriate specification is an event study. As can be seen in Online Appendix Figure A.7, the marriage rate of 18-year-olds does not increase after the reform, indicating that most couples postpone formal marriage even longer or dissolve before they reach the minimum age of marriage.
B. School Attendance and Early Motherhood
Figure 7 presents the event-study estimates of the impact of banning child marriage on school attendance for 16- and 17-year-old girls. Our framework predicts that if there are social sanctions for being in an informal union, banning child marriage reduces dropout rates. However, we find small and statistically insignificant impacts on school attendance. These null results persist seven quarters after the reform.30 In particular, we can reject an average increase in the likelihood of attending school larger than 2.4 percentage points (a 3.4 percent increase over the mean). This confirms that the reputation cost of informal unions is negligible in Mexico. We observe similar impacts if we separately analyze girls in each age group, as shown in Table 4. We only report the estimates for the first seven quarters after the reform because there are few states with longer lags. The coefficient for the longest lag is estimated using 24 of the 30 states that enacted the reform. Importantly, there are no differential trends. Given that 6 percent of the 17-year-old girls in our sample are not attending school because they finished high school, we also conduct this analysis using a dummy for whether the girl attends school or completed high school as the dependent variable. Our conclusions are unchanged, as shown in Online Appendix Figure A.8. We also estimate the impact of the reform using administrative data on school enrollment by age, gender, and state, obtained from the Instituto Nacional para la Evaluación de Educación. We combine these data with the midyear population estimates from CONAPO to compute the share of girls enrolled in school at the start of the academic year (in August). Since this data set ends in 2017 and only has annual information, there is not enough variation for an event-study estimation.31 We present our difference-in-difference estimates of the effect of banning child marriage on school enrollment in Online Appendix Table A.4. Once again, we find that the reform has no impact on schooling decisions.
Event-Study Estimates of the Effect of Banning Child Marriage on School Attendance of 16- and 17-Year-Old Girls
Notes: The sample is composed of all 16–17-year-old girls interviewed in ENOE in 2008–2018. This figure plots the coefficients and 95 percent confidence intervals based on wild-bootstrap clustered standard errors of a regression in which the dependent variable is a dummy for whether the girl attended school at the moment of the survey. The regressors of interest are dummy variables for each quarter relative to the period in which child marriage was banned for this age group in the girl’s state of residence, with the quarter before the reform being the omitted category. We only report five lead coefficients for ease of interpretation and seven lag coefficients because there are few states for which more than seven quarters have passed since the reform was enacted. The regression also includes state fixed effects, quarter–year fixed effects, states’ unemployment rate, poverty rate, labor force participation of females ages 20 and older, average income of employed individuals, population (in ln), dummy variables for whether the state governor belongs to PRI, PAN, or PRD, a dummy for whether girls of this age required the authorization of a judge/mayor to get married, age dummies, and town-size dummies. These estimates are weighted using the sampling weights provided in ENOE. Standard errors are clustered at the state level.
Event-Study Estimates of the Effect of Banning Child Marriage on School Attendance of 16- and 17-Year-Old Girls
As predicted in our conceptual framework, the impact of banning child marriage on fertility depends on the social penalty faced by girls in an informal union. If social norms discourage girls from entering an informal union once child marriage is illegal, the prohibition of child marriage could lead to a reduction in birth rates. However, girls affected by the ban could conceive out of wedlock or in an informal union, thus reducing or even eliminating the effect of the reform on early fertility. Consistent with the latter, Figure 8 shows that the reform did not reduce the fertility rate of girls 16–17-years old, the age group for which the reform had an impact on formal marriage rates.32 In particular, we can reject an average drop larger than 0.2 births per 1,000 girls, a 4 percent reduction over the mean. These findings confirm once again that in Mexico, informal unions are socially accepted. We also report the results of these estimates in Table 5, with estimates grouped into four-month periods for ease of interpretation. We report the estimates for 15 lags because by March of 2018 (the end of our sample period), there were few states for which more than 15 months had passed since the reform. In particular, the coefficient for the longest lag is estimated using 22 of the 28 states that enacted the reform before March 2018. Importantly, there are no differential pre-trends in fertility rates. If we focus on all births instead of just first births (Online Appendix Table A.5), we do not find an effect either.
Event-Study Estimates of the Effect of Banning Child Marriage on Early Motherhood of 16- and 17-Year-Old Girls
Notes: The sample includes all Mexican states from January 2008 to March 2018, and the unit of observation is a month–state. This figure plots the coefficients and 95 percent confidence intervals based on wild-bootstrap clustered standard errors of a regression in which the dependent variable is the monthly number of (live) first births conceived in that month by girls ages 16–17, per 1,000 girls of this age who reside in that state. The regressors of interest are dummy variables for each month relative to the period in which child marriage was banned for this age group in state s, with the month before the reform being the omitted category. We only report 12 lead coefficients for ease of interpretation and 15 lag coefficients because there are few states for which more than 15 months have passed since the reform was enacted. The regression also includes state fixed effects, month–year fixed effects, states’ unemployment rate, poverty rate, labor force participation of females ages 20 and older, average income of employed individuals, population (in ln), dummy variables for whether the state governor belongs to PRI, PAN, or PRD, and a dummy for whether girls of this age required the authorization of a judge/mayor to get married. Standard errors are clustered at the state level.
Event-Study Estimates of the Effect of Banning Child Marriage on Early Motherhood of 16- and 17-Year-Old Girls
C. Informal Unions
A possible reason for why the reform has no impact on school attendance and early motherhood is that the decrease in formal marriages is offset by an increase in informal unions. This is hard to examine empirically, as there is no register of informal unions, and young girls have incentives to underreport being married or in an informal union when they are surveyed.33 We take advantage of the fact that the birth registration microdata has information on the self-reported civil status of the mother at the moment of delivery and that women who are already having a child at a young age have fewer incentives to lie about their civil status. Importantly, 62 percent of women who were child brides prior to the reform had their first child before the age of 18 (Online Appendix Figure A.5), so the sample of young mothers makes up for a big portion of the women who were affected by the reform. Since the increase in the minimum age of marriage has no impact on early motherhood rates, we can explore whether the reform led to an increase in informal unions using the sample of girls who were mothers before age 18.
Using an event-study estimation, we examine the impact of banning child marriage on the share of 16–17-year-old mothers by civil status.34 Since civil status is measured at the moment of birth, we perform this analysis using the month of delivery instead of the month of conception. Given that our analysis is conducted at the moment of birth rather than the moment of conception, we report the estimates for 24 lags (instead of 15 lags as in the fertility estimations). The results are presented in Figure 9 and Table 6. Consistent with our results on formal child marriage rates, we find that banning child marriage gradually reduces the share of 16- and 17-year-old mothers who are formally married.35 This reduction in the share of married mothers is completely counteracted by an increase in the share of mothers in an informal union.36 Importantly, there are no differential pretrends in the civil status of mothers, substantiating the causal interpretation of our estimates. Taken together, these results indicate that informal unions carry no social sanction in Mexico, leading girls to substitute marriage for informal unions after the prohibition. In societies where informal unions are a valid option for young couples, banning child marriage will not reduce school dropout and early fertility.
Event-Study Estimates of the Effect of Banning Child Marriage on Civil Status of 16–17-Year-Old Mothers
Notes: The sample includes all Mexican states in 2008–2018, and the unit of observation is a month–state. These figures plot the coefficients and 95 percent confidence intervals based on wild-bootstrap clustered standard errors of a regression in which the dependent variable is the share of first births delivered by girls ages 16–17 that are in a marriage and informal union at the moment of delivery, respectively. The regressors of interest are dummy variables for each month relative to the period in which child marriage was banned for this age group in state s, with the month before the reform being the omitted category. We only report 12 lead coefficients for ease of interpretation and 24 lag coefficients because there are few states for which more than 24 months have passed since the reform was enacted. The regression also includes state fixed effects, month–year fixed effects, states’ unemployment rate, poverty rate, labor force participation of females ages 20 and older, average income of employed individuals, population (in ln), dummy variables for whether the state governor belongs to PRI, PAN, or PRD, and a dummy for whether girls of this age required the authorization of a judge/mayor to get married. Standard errors are clustered at the state level.
Event-Study Estimates of the Effect of Banning Child Marriage on Civil Status of 16- and 17-Year-Old Mothers
VII. Implications
The change in civil status that we observe for young mothers could potentially have negative effects for them and their children. One way to explore this hypothesis is to examine whether the increase in the minimum age of marriage has an impact on prenatal investments and newborn health outcomes. There a several reasons why pregnant women in informal unions may adopt less healthy behaviors than married women, translating intro worse health outcomes for their children at birth. As explained in Section II, the rise in the minimum age of marriage prevents couples with underage members from being legally recognized. This lack of legal recognition and the lower exit costs from the relationship may discourage fathers from being involved during the pregnancy. This could reduce their provision of emotional support and how much they promote the engagement in healthy behaviors by the mother. A lower involvement by fathers could result in women investing less in prenatal care. This lack of legal recognition could also affect the girls’ perception about the wantedness of the pregnancy (Kane 2016). These perceptions could influence birth outcomes either directly or indirectly via other protective behaviors (Weller, Eberstein, and Bailey 1987; Kroelinger and Oths 2000; Shah et al. 2011; Shah, Gee, and Theall 2014).
Online Appendix Table A.7 shows that prior to the reform, young mothers who were formally married invested more in prenatal care, on average, than mothers of the same age with a different civil status. For instance, 75 percent of the girls who had their first child at ages 16–17 and were married received their first prenatal visit during the first trimester of pregnancy, as opposed to only 70 percent and 62 percent of the mothers that were in informal unions and single. Married mothers attended 7.2 prenatal visits on average during their pregnancy, compared with 6.5 and 6.2 visits by mothers who were in informal unions or single. It should be noted, however, that there is less of a gap in infant health for children from married and cohabitant mothers. As these differences (or lack thereof) could be driven by selection factors, a more rigorous analysis is needed.
Using an event-study specification, we estimate the impact of banning child marriage on the investments in prenatal care of 16- and 17-year-old mothers and the health outcomes of their offspring at birth. We consider that a girl can be affected by the reform up until the moment of conception, since the potential protection effect of marriage occurs during the pregnancy (Kane 2016). We present the results in Table 7, where the estimates are grouped into four-month periods for ease of interpretation. We find that the reform did not modify the prenatal investment of mothers who had their first child at ages 16–17.37 In particular, we do not find a clear effect in the probability of receiving prenatal care or attending their first prenatal visit during the first trimester of pregnancy. There is no impact either on the number of prenatal visits, or on the likelihood of having a cesarean delivery. Consistent with the null impacts on prenatal investments, we find no effect either on newborn health. We can reject an increase larger than 0.3 percentage points in the probability of having a first birth with low birth weight and an increase larger than 0.2 percentage points in the probability of having a premature baby. Importantly, there are no differential pre-trends. We do not find an effect either if we allow the reform to have an impact up until the moment of birth, instead of conception (Online Appendix Table A.9).
Event-Study Estimates of the Effect of Banning Child Marriage on Prenatal Investment ofl6—17-Year-Old Mothers and Newborn Health
Although we do not observe any short-term impacts on newborn health, this shift in civil status may have medium- and long-term effects. As we show in Section VI.A, the girls who were affected by the reform do not get legally married once they turn 18. This could be a result of the dissolution of these unions, as informal unions are easier to terminate than marriages. The dissolution of early unions could potentially have positive effects on women’s welfare, particularly when the quality of the match is not high. This hypothesis is supported by the findings of a previous literature that shows that when marriages are easier to terminate, through unilateral divorce, there is an increase in the likelihood that a relationship with domestic violence ends (Stevenson and Wolfers 2006). Alternatively, couples may stay together but postpone their marriage even longer or not get married at all. Substituting marriage for informal unions could be beneficial if it increases the bargaining power of women and decreases spousal violence within couples that stay together (Brassiolo 2016) or detrimental if domestic violence is used to prevent women from exiting the relationship (García-Ramos 2021). A shift from marriage to informal unions may be undesirable if unions that are easier to dissolve lead couples to invest less in marriage-specific capital (Stevenson 2007).38 Furthermore, people in informal unions do not have the same rights as those who are formally married after the relationship dissolves, as discussed in Section II. As the recent enactment of the laws raising the minimum age of marriage does not allow us to examine medium- and long-run effects, we leave this for future research.
VIII. Validity and Robustness Checks
A. Accuracy of Marriage Certificate Data
Blank, Charles, and Sallee (2009) show that using marriage certificate data to study the impact of age-of-marriage laws can lead to biased estimates for two reasons. First, underage individuals could potentially travel to states where child marriage is permitted and get married there. Since our marriage certificate data has information on the state of occurrence of the marriage and the state of residence of the bride and groom, we can examine whether this issue is likely to bias our estimates. Only 2 percent of the child marriages in our period of analysis took place in a state that does not coincide with the bride’s or groom’s state of residence, leaving little room for bias from spillovers. As shown in Columns 1 and 2 of Online Appendix Table A.10, the magnitude and statistical significance of our estimates is almost unaffected if we conduct our analysis using the state in which the marriage took place instead of the bride’s state of residence.39’40 The second concern raised by the findings of Blank, Charles, and Sallee (2009) is that underage people can still marry in their state of residence by lying about their age. The possibility that young people lie about their age to get around the child marriage prohibition is less of a concern in our setting, as state laws require the submission of birth certificates to get married.41 If underage brides were lying about their age as a response to the ban in child marriages, we should see a rise in the marriage rate for 18-year-olds immediately after the reform. As we show in Online Appendix Figure A.7, this is not the case.
B. Unobservable Confounders
As discussed in Section V, our assumption for identifying the causal effect of banning child marriage is that conditional on state fixed effects, month–year fixed effects, and controls, there are no time-varying state-specific factors correlated with the ban in child marriage and our outcome variables. One potential source of bias is that child marriage prohibitions could be driven by a state-specific decline in the value placed on marriage. If that were the case, we should also observe a drop in the marriage rates of older women not affected by the ban. However, as can be seen in Online Appendix Table A.11 and Online Appendix Figures A.14 and A.15, banning child marriage has no impact on the marriage rates and the share of married mothers for women of other ages. In the case of fertility, our null impacts could be driven by unobservable determinants of fertility that correlate with the timing of the child marriage ban. For example, states might have decreased the funding for contraception programs, leading to higher childbearing rates that counteract a reduction in fertility resulting from the reform. However, as shown in Online Appendix Figures A.16–A.17, the ban in child marriage is not correlated with changes in the fertility rates of women in other age groups, further corroborating our findings.
C. Exclusion of Oaxaca and Zacatecas
In June of 2016, the state of Zacatecas changed the punishment for intercourse with a person ages 16–17 from estupro to rape. The state of Oaxaca followed suit in November of 2018. It is unclear whether marriage and/or informal unions were legal for this age group after these changes. As a robustness check, we exclude these two states from our analysis. As shown in Online Appendix Table A.12, our main estimates are robust to this exclusion.
IX. Conclusions
We examine the impact of increasing the minimum age of marriage on child marriage rates, school attendance, and early fertility in Mexico. Using microdata derived from marriage certificates, we find that banning child marriage leads to a large and statistically significant reduction in the number of registered child marriages. However, the reform does not reduce the dropout or fertility rates of the affected cohorts. Using data on the civil status of mothers reported in birth registration data, we find that the reduction in the share of married mothers caused by the child marriage ban is counteracted by an equivalent rise in the share of young mothers in an informal union. These findings indicate that in Mexico, the social sanctions faced by girls in informal unions are negligible. This change in civil status could potentially have negative effects if girls in informal unions do not have the same legal benefits or spousal support as those who are married. However, we do not find any effects on prenatal investments or newborn health. Even though we do not find negative short-run consequences, the medium- to long-term impacts are unclear. As the reform was recently enacted, we will explore this in future research. These results suggest that in places where cohabitation is a socially acceptable alternative to formal marriage, raising the minimum age of marriage is not enough to reduce child marriage rates or prevent its detrimental consequences.
Our findings are especially relevant for other countries in Latin America and the Caribbean, where norms around informal unions are similar, and child marriage and early fertility rates are comparable to those in Mexico, as shown in Online Appendix Figure A.18. One option for policymakers in these contexts is to modify age-of-consent laws to make informal child marriages illegal as well. Alternatively, recent findings from Malawi (Baird, McIntosh, and Özler 2011), India (Jensen 2012), Bangladesh (Buchmann et al. 2018), and Uganda (Bandiera et al. 2020) show that policies providing young girls with economic opportunities or direct incentives to delay marriage may be effective at reducing child marriage rates. Future research should corroborate whether these findings also generalize to other regions, where the drivers of child marriage may differ.
Footnotes
They thank Eric Arias, Cristian Crespo, Emma Duchini, Ana Figueiredo, Andrés Gago, Aixa García Ramos, Andrea Ichino, Ricardo Maertens, Ernesto Schargrodsky, Rodrigo Soares, Michelle Sovinsky, Eleftheria Triviza, Joe Vecci, and seminar and conference participants at Universidad Torcuato Di Tella, Universidad Nacional de la Plata, the Center for Evaluation and Development, Corporación Andina de Fomento, JILAEE, Seminario de Micro Aplicada Perú, the LACEA Annual Conference, the Symposium of the Spanish Economic Association, and the Eighth Development Workshop in Chile, as well as two anonymous referees for their comments. This research is funded by the CAF Development Bank research grant on human capital formation and youth access to high-quality employment in Latin America. Bellés-Obrero acknowledges support from the German Research Foundation (DFG) through CRC TR 224 (Project A02), and project ECO2017-82350-R. There are no further sources of funding or conflicts of interest to disclose. Replication files are provided at http://dx.doi.org/10.17632/ng6bw2hhhc.1
Supplementary materials are freely available online at: http://uwpress.wisc.edu/journals/journals/jhr-supplementary.html
↵1. A survey of marriage laws in 193 countries conducted by the World Policy Center in 2013 revealed that when all exceptions were considered, 87 percent of these countries had a minimum age of marriage lower than 18.
↵2. Several countries have recently set the minimum age of marriage at 18, without exception. Some examples include Chad, Costa Rica, Ecuador, Guatemala, Honduras, Malawi, Nepal, Panama, Turkmenistan, and Zimbabwe (Girls Not Brides 2017).
↵3. Enforcement may even be unfeasible in contexts where lack of widespread birth registration prevents age verification at marriage (Jensen and Thornton 2003). This is not a concern in Mexico because birth registration is almost universal. A comparison of the number of registered births in 1999 from vital statistics data and the number of newborn babies in the 2000 census shows that more than 96 percent of these births were registered (Péres Paredes and Meneses Mendoza 2008).
↵4. As laws signal what is right and wrong, they can affect preferences over the regulated behavior and lead to higher compliance by changing social norms. The impact of laws beyond deterrence has been empirically corroborated in the case of compulsory voting laws with symbolic fines for abstention in Switzerland and Austria (Funk 2007; Hoffman, León, and Lombardi 2017) and seat belt laws with low enforcement in the United States (Cohen and Einav 2003).
↵5. For further details, see https://www.girlsnotbrides.org/where-does-it-happen (accessed June 6, 2022).
↵6. The reform only led to a 49 percent reduction in registered child marriages because formal child marriage rates were decreasing in the entire country in our period of analysis for reasons unrelated to the reform. Our estimates capture the impact of the law change, above and beyond the countrywide drop in formal child marriage rates.
7. In a working paper, Lyn and Rainer (2020) study the same question as this paper using different data, and reach similar results on early fertility and school attendance. However, the data used in that study has some limitations that we explain in detail in Footnotes 20 and 23. Unlike our study, Lyn and Rainer (2020) find a decrease in the share of girls in informal unions using data from the Mexican labor force survey. These data are not adequate to study this question because girls underreport being married or in an informal union when surveyed. This is less of a concern in our study, since we use birth registry data. When we estimate the impact on informal unions using these alternative data, we find no effect, since girls do not report this behavior in the first place. We discuss this in detail in Footnote 32.
8. Importantly, the sample of young mothers makes up for a big portion of the girls who were affected by the reform, as discussed in detail in Section VI.C.
↵9. In the 74 developing countries that participated in the DHS and MICS surveys in 2008–2017, 36 percent of the 15–19-year-old girls who lived with their partners were not formally married. The share of girls who cohabitate without being legally married is probably higher, as DHS and MICS statistics do not distinguish between registered and unregistered marriages, and marriage registration is low in many developing countries (Center for Reproductive Rights 2013; Centre for Human Rights 2018).
↵10. The authors do not consider exceptions to the minimum age of marriage based on religion, customs, pregnancy, or judicial authorization, so it is unclear whether enforcement is also low when all exceptions are considered.
↵11. The states of Baja California Sur and Veracruz had already modified their legislation to ban child marriage earlier in 2014.
↵12. Online Appendix Figures A.1 and A.2 show the geographical dispersion in the adoption of this reform.
↵13. Informal unions (legally known as concubinato) are defined by the civil and family codes of the different Mexican states. For instance, in article 313 bis of the civil code of the state of Aguascalientes or article 297 of the civil code of the state of Puebla.
↵14. Importantly, 87 percent of the formal child marriages and informal unions in the sample started at age 15 or higher, so these characteristics are mostly measured before the child marriage.
↵15. Although some of this information is also available in EDER, we use ENADID due to its larger sample size. While EDER has information on 4,438 women ages 20–30 in 2013, there are almost 27,000 women of these ages in ENADID. We do not use the ENADID survey to analyze baseline differences between these two groups, as it does not have retrospective information on life conditions during childhood.
↵16. Compulsory schooling in Mexico is composed of six years of primary school and three years of middle school, known as secundaria. Secondary school (bachillerato or preparatoria) is not mandatory and lasts for three years. The women born in the early 1990s were 20–24 at the time of the interview. Although some of them might still finish secondary school, it is worth noting that only 6 percent of those who had not finished high school were still studying at the time of the survey.
↵17. We back out the month and year in which respondents became pregnant with their first child by subtracting nine months from the month–year in which their first child was born. We then compare this date to the month–year in which they got married.Sweden Netherlands Haiti Argentina Norway
↵18. The replication files for this paper can be found in Lombardi and Bellés-Obrero (2020).
↵19. The age of the bride is missing in 0.30 percent of observations, which we drop from our sample.
↵20. Almost 94 percent of the registered child marriages in 2008–2018 had a bride below the age of 18, whereas only 19 percent had an underage groom.
↵21. For births taking place outside of a medical unit, the mother is required to attend a health institution shortly after the birth to obtain this documentation. These account for a small share of births, since the vast majority of births are overseen by a doctor. While coverage is not universal, SINAC registered approximately 90 percent of births in its first years and has a 95 percent coverage since 2013. Lyn and Rainer (2020) use a different data set compiled by INEGI using microdata from birth certificates. However, these data are not suitable to calculate birth rates in recent years because a sizable share of parents take some time to obtain their child’s birth certificate. As only 80 percent of children obtain their birth certificate before their first birthday, the recommendation is to wait four years to have sufficiently high coverage. Further details on the coverage of both databases can be found in Hernández, Tapia, and Alarcon (2015).
↵22. Another advantage of the SINAC database is that unlike the data from birth certificates, it reports gestation length, allowing us to precisely pin down the month of conception. The length of gestation is missing in 0.4 percent of births, for which we assume a gestation of 40 weeks. The SINAC data set also has detailed information on prenatal investments and health outcomes that are absent from the data set derived from birth certificates.
↵23. CONAPO compiles population counts for each state at the start and middle of each year, by gender and age. These statistics are derived from the decennial censuses and population counts taking place between censuses. Statistics for 2016 onwards are projected. We calculated the monthly population for each age group and gender using linear interpolation.
↵24. While our analysis uses all waves of the Mexican labor force survey (ENOE) in 2008–2018, Lyn and Rainer (2020) only use certain survey waves that also measure child labor (the Módulo de Trabajo Infantil). The ENOE is carried out four times a year, whereas this special module is conducted once every two years. Using these data implies losing a large share of observations and much of the identifying variation without any advantage.
↵25. The unemployment rate, poverty rate and average income of employed people were also calculated by INEGI and CONEVAL using data from ENOE. Since ENOE is a quarterly survey, we assume the same value within the months of each quarter.
↵26. In the case of our school enrollment regressions, this regressor varies at the quarterly level.
↵27. For example, in our estimates on the impact of banning child marriage for 16–17-year olds, less than 8 percent of the two-by-two difference-in-differences estimators receive a negative weight. We computed the proportion of negative weights using the twowayfeweights command in Stata.
↵28. Age-of-marriage laws are mostly enforced in Mexico. Online Appendix Table A.3 shows that in the period after the reform, the monthly number of formal child marriages was very small in most states, and most of the nonenforcement took place in the first three months after the law change. The majority of the child marriages occurring after the reform are probably couples that took the matter to the courts and were allowed to get married with the ruling of the judge (García Sánchez 2019).
↵29. We bin longer lags together and estimate them using a single dummy variable that is not reported in the plot.
↵30. We also analyze whether banning child marriage has an impact on the time devoted to caretaking activities and housework using data from ENOE. On average, 16 percent of 16–17-year-old girls devote time to nonpaid caretaking, whereas 93 percent devote some time to housework. The average weekly time spent on caretaking and housework is 2.8 and 11.9 hours, respectively. We run our event study estimates for these outcomes, and find no effect (results upon request).
↵31. While 17 of the 26 states that banned child marriage by August 2017 passed the reform a year or more before this date, only five of these states banned child marriage two or more years before August 2017.
↵32. We also find no impact of banning child marriage on school attendance or fertility rates of 14–15-year olds, as shown in Online Appendix Figures A.9 and A.10.
↵33. For instance, in the sample of 17-year-olds interviewed in the ENOE labor force survey in 2008–2011, almost 88 percent report being single. In a retrospective survey of respondents from the same cohort (the ENADID survey), only 79 percent report that they were single at the age of 17. Given this underreporting, it is unlikely that we can uncover the impacts on civil status using these data. In contrast to Lyn and Rainer (2020), we find no impact on civil status using data from the ENOE survey, as shown in Online Appendix Figure A.11.
↵34. Only 0.3 percent of these mothers are divorced, separated, or widowed. We include divorced mothers inside the definition of married women, while single mothers also include those that were separated or widowed.
↵35. We find similar results if we examine all births instead of first births (Online Appendix Table A.6). Unlike the case of older girls, we find no impact of banning child marriage on the civil status of 14- and 15-year-old mothers at the time of delivery, as shown in Online Appendix Figure A.12. This null effect is likely due to the smaller reduction in child marriage rates for this age group and the fact that the fertility rate at this age is substantially smaller than that of their 16–17-year-old counterparts.
↵36. We also find a small decrease in the share of single mothers one year after the reform. We believe this could be driven by a normalization of informal unions.
↵37. We do not find any impact on prenatal investments and newborn health of 14- and 15-year-olds either, as seen in Online Appendix Table A.8.
↵38. Stevenson (2007) shows that couples that could potentially have access to unilateral divorce are less likely to support a spouse through school, more likely to have both spouses in the labor force, and less likely to have a child.
↵39. As we explain in Section II, a consequence of the ban on child marriage is that informal unions are no longer recognized by law. This implies that there are even more advantages to formal marriage in this context that could be obtained by marrying in a different state. We examine whether the reform led girls to marry in adjacent states using an event-study estimation in which the dependent variable is the monthly number of out-of-state formal child marriages with a bride ages 16–17, for every 1,000 girls of this age living in the state. We exclude 2017 and 2018 since few states shared a border with a state where child marriage was still legal at that time (Online Appendix Figure A.1). As canbe seen in Online Appendix Figure A.13, the reformhad no effect on the rate of out-of-state child marriages, and the point estimates are very small in comparison to the average child marriage rate of 1.133 marriages a month. The fact that couples do not get married in another state indicates that perhaps girls and their partners are uninformed about the advantages of formal marriage, or that the cost of getting married in another state is too high.
↵40. We also redo this analysis focusing on the groom’s state of residence, as the bride and groom reside in different states in 5 percent of the child marriages in our sample. As can be seen in Columns 3 and 4 of Online Appendix Table A.10, our coefficients are almost identical to our baseline specification, which uses the state of residence of the bride.
↵41. The evidence from Blank, Charles, and Sallee (2009) is from the United States in the 1950s, when documentary evidence of proof of age was not generally required to get married.
- Received December 2019.
- Accepted October 2020.
