Migration, health knowledge and teenage fertility: evidence from Mexico

Abstract

Migration may affect fertility and child health care of those remaining in the country of origin. Mexican data show that having at least one household member who migrated to the United States decreases the occurrence of pregnancy among teenagers by 0.339 probability points. This finding can be partially explained by the fact that teenagers in migrant households have a higher knowledge of contraceptive methods and likely practice active birth control. I use potential migration, measured as historic migration rates interacted with the proportion of adult males in the household, as an instrument to account for the endogeneity of migrant status.

1 Introduction

Far from being a new phenomenon, Mexican migration to the United States has seen increasing flows of migrants over time. In both absolute and relative terms, their number has been larger than any other immigrant influx in the past century, corresponding nowadays to approximately seven million people and therefore 31 % of the foreign born population (U.S. Census Bureau 2006).¹ According to the 2000 Mexican Census, from 1995 to 2000, 4.1 % of all Mexican households saw at least one family member migrating to live in the United States, while an additional 1.8 % of households had family members migrating back and forth between the two countries or returning to Mexico (INEGI 2000). Overall, almost 10 % of the population born in Mexico now resides in the United States (UNDP 2007).² This paper aims at investigating the impact of this international migration on occurrence of pregnancy and child health care of women, especially teenagers, remaining in Mexico. My results show that having at least one household member who migrated to the United States decreases the occurrence of pregnancy among teenagers by 0.339 probability points. The impact observed is very large, representing a decline to 0.09 from the sample mean. This finding can be partially explained by the fact that teenagers in migrant households have a higher knowledge of contraceptive methods and likely practice active birth control. Being in a migrant household increases the knowledge of at least one contraceptive method among teenagers by 0.307 probability points on average. Less significant results are obtained for child care outcomes: being in a migrant household only raises the likelihood of the child to be delivered by a doctor in the sample of all women, especially in rural areas. This probability is estimated to increase by 0.569 probability points on average.

I focus on teenage women because delaying onset of fertility is still a policy priority in the country. Although fertility rates have decreased considerably in the last decades—the number of children per women of childbearing age decreased from 6.8 in 1970 to 2.2 in 2006³—adolescent fertility rates have declined much more slowly. In 2005, the pregnancy rate for women between 12 and 19 years old was still 79 out of 1000, in public institutions 21 % of children were born to women under the age of 20 (CONAPO 2007), and the proportion of teenage mothers was around 17 % (INEGI 2005).⁴ Adolescent pregnancy and childbearing are associated with a range of adverse health outcomes, including high risk of pregnancy-related diseases, unsafe abortion practices, and maternal mortality.⁵ Adolescent mothers are also less likely than women aged 19–23 to use either antenatal or delivery care and have their infants immunized (Reynolds et al. 2006).⁶ Moreover, especially among the poor, adolescent childbearing is linked with lower future monthly earnings for mothers, and contributes to the persistence of poverty from one generation to the next, thus affecting long-term human capital accumulation and growth. Adolescent child bearers are themselves often born to adolescent mothers: in Mexico two-thirds of adolescent mothers have mothers who gave birth in their teens (Buvinic 1998).

The major concern in this study is the endogeneity of migrant status: family planning decisions can be related to the characteristics of migrant households themselves. The direction of the potential selection of individual into migration is a priori unclear. It is difficult to disentangle the extent to which the impacts on occurrence of pregnancy and child health care reflect the unobserved characteristics of migrants, their households or their communities as opposed to the migration experience itself. I tackle the problem by using an instrumental variable method. Potential migration—measured as historic migration levels interacted with the proportion of adult males in the household—is used as instrument for current migration. Historically, migrants have mainly come from the central-western region of the country. The states of Jalisco, Michoacán, and Guanajuato, which together accounted for about 30 % of all Mexico-U.S. migrants throughout the twentieth century, are still among the principal sending states. In order to obtain variations at the household level, 1924 migration levels are interacted with the proportion of adult males in the household: in the Mexican context, the likelihood that at least one household member migrates increases with the proportion of males in the household.

The principal channels through which migration is expected to affect occurrence of pregnancy and child health care are wealth, exposure to the host country’s norms and knowledge transmission to the sending country, and cost of migration. First, remittances sent back from migrants likely change household income and allow families to spend resources on different categories of expenditure, such as education and child health care. Higher levels of educational attainment and health investment are expected in households experiencing migration and receiving remittances, with consequences on human capital accumulation and family planning decisions (Cox et al. 2004). Access to education by adolescent girls may delay the onset of fertility (Breierova and Duflo 2004) and increase the opportunity cost of women’s time, thus leading to changes in preferences regarding the quality and quantity of children (Becker 1960) and endowing girls with a better ability to process information, which, in turn, potentially increases knowledge of contraception options (Rosenzweig and Schultz 1987; Duncan et al. 1991).⁷

Second, through exposure to the host country’s practices, migrants acquire health knowledge and become more familiar with the use of contraceptive methods. In turn, they provide examples of behaviors that may be considered and copied by other family members at home. As much of the literature suggests (Fargues 2006; Beine et al. 2013; Lindstrom and Munoz-Franco 2005), migrants’ behavioral norms tend to converge to those of their host country. Through the process of living and working in the receiving country, migrants acquire information and become aware of alternative models of gender roles and family relationships that they may accept and adopt. This assimilation process—a change in individual preferences—is gradual. The impact of the host country’s values and norms on migrants’ behavior increases with the length of migration: the longer a migrant has been in the receiving country, the more her outcomes are similar to those of a native. The adjustment in fertility behavior, however, also occurs in response to the economic opportunities and constraints present in the country of destination, and can therefore take place in the short term. Both processes happen simultaneously so it is challenging to disentangle the effects of assimilation—change in preferences—from those of adaptation—adjustment in fertility behavior due to economic opportunities and constraints (Lindstrom and Saucedo 2002). What is observed is that migrants share similar economic and institutional environments with natives and acquire information and behavioral norms of the country of residence. Fertility rate of the Mexican-born population in the United States is converging towards the non-Hispanic U.S. fertility rate (IPUMS 2010). Information and behaviors that flow from receiving to sending country communities can represent an additional pathway through which migration affects health outcomes and fertility behaviors. Kovsted et al. (2002), for instance, suggested that although maternal education is important in determining child health and mortality, its effect diminishes or disappears when health knowledge is introduced as an explanatory variable. Health knowledge crowds out the effects of mother’s education and is thus not necessarily associated with the income or education of parents. The information and behaviors obtained through the experience in the receiving country therefore have the potential to result in less risky sexual behaviors and—in the long run—in better health outcomes also in the sending country.⁸ The strong family ties of Mexican society and the recurrence of Mexican migration to the United States make the role of migration networks as a transfer of behavioral norms from the host to the migrants’ home country highly plausible.

Third, the migration of a household member decreases the cost of migration for other family members and women, especially teenagers, can decide to delay pregnancy in anticipation of future migration. Having at least one household member who migrated to the United States increases the likelihood that another member who remained in the home country will try to migrate. Thus, women who do not want to get pregnant in Mexico, as this would increase their cost of migration, will acquire more knowledge of contraceptive methods and likely practice active birth control.

My data do not allow to provide a complete decomposition of the channels through which migration matters for the occurrence of pregnancy and child health care. I show some evidence of the increased knowledge of contraceptive methods among teenagers in households that experienced migration to U.S. and I argue that these girls may be better endowed to process information (thanks to higher access to education-wealth effect), decide to delay pregnancy and actively acquire more information (due to the reduced cost of migration) and are more exposed to behavioral norms of their household members’ host country (knowledge transmission to the sending country). Knowledge of contraceptive methods is therefore used as a measure of more general health knowledge.

So far the literature has not largely investigated the interaction between migration and health outcomes or, even more specifically, sexual behavior. When it has, it mainly focuses on the potential effects in the country of destination, such as access to welfare benefits (Bertrand et al. 2000), health services (Menjivar 2002; Deri 2005; Devillanova 2008), and fertility (Chou 2011), but little is said about the impact of migration on these outcomes in the sending region. There are some recent contributions in this direction. Beine et al. (2013) examined the relationship between international migration and source country fertility and provided evidence of a strong transfer of fertility norms from migrants to their country of origin. Analogously, in their study on rural Guatemala, Lindstrom and Munoz-Franco (2005) found that contraceptive use increases and fertility falls with variables such as having family members in urban or international destinations and living in a community where urban migration is common. De (2013) showed that women in Mexico belonging to migrant families have a higher propensity to use modern contraceptives. Kanaiaupuni and Donato (1999), Frank and Hummer (2002), Hildebrandt and McKenzie (2005) used the same Mexican data to investigate the effects of migration on infant mortality and birth weight, finding positive effects for children living in households with at least one migrant. In Pakistan, Mansuri (2006) found that migration has a positive impact on early child growth outcomes, especially for young girls. My paper is related to this strand of the literature and attempts to provide a first, in-depth investigation of the impact of migration on occurrence of pregnancy in the country of origin, especially for adolescent women for whom delaying onset of fertility is still a policy priority in many countries.

The paper is structured as follows. Section  2 provides information on the data used in the study. The empirical strategy is presented in Sect.  3. Section  4 illustrates and interprets the results. Conclusions are drawn in the last section.

2 Data

This study uses data from the Mexican National Survey of Demographic Dynamics (ENADID—Encuesta Nacional de Dinámica Demográfica) of 1997 and 2006.⁹ The households for which information is collected are not the same across years. Therefore, I use a pooled cross-sectional dataset. Detailed individual and household characteristics, including demographics, education and dwelling infrastructures are available in the pooled sample for 468,503 individuals in 107,654 households across all 32 Mexican states. The survey also offers information on permanent and temporary migration to the United States. For migrants, data are collected on age, sex, relationship with the head of the household, year and duration of migration, state of destination, reasons to migrate and remittances.¹⁰ Moreover, all women aged 15–54 in each household are asked detailed questions about their fertility history, knowledge and use of contraceptive methods, and child health care. For the purpose of the analysis, attention is placed on women for whom information on migration is available. The final dataset therefore comprises 39,133 women aged 15–54 in 28,433 households. Among these women, almost 21 % are less than 20 years old.

International migration The ENADID survey asks each household member whether she has ever been to the United States in the 5 years prior to the day of the interview.¹¹ This implies that I do not get information on the migration experience before 5 years prior to that day. ENADID does not ask the year of the first trip to U.S., only the last trip. Due to the way the question is asked, those who migrated and returned to Mexico before 5 years prior to the survey are not migrants in my sample. I am not particularly concerned that this definition may bias my results because Mexican migration to the United States tends to be recurrent.¹² If one migrates once, it is likely that she will migrate again. If one migrates before the five-year span, she will likely migrate also in this interval. Therefore, she is part of the migrant sample. In my pooled sample, almost 54 % of the individuals who migrated moved back and forth.

I am interested in the impact of previous migration experience on subsequent occurrence of pregnancy and child health care. My outcomes of interest need not be affected by the same shocks leading to the migration decision. As suggested by Hildebrandt and McKenzie (2005), I divide the 5-year period into two periods: the first 2 years and the second 3 years. The choice of using January 1994 and 2003 as thresholds depends on the way the question on the probability of getting pregnant has been posed in the survey. The information on children and on the occurrence of pregnancy was collected from 1994 in the 1997 dataset and from 2003 in the 2006 dataset. Moreover, reducing the period of analysis to less than 3 years would create a too short interval of time to expect any impact of the previous migration experience. A household is defined as a migrant household if at least one adult member made her last trip to the United States between January 1st 1992 and January 1st 1994 or between January 1st 2001 and January 1st 2003, depending on the dataset.¹³ The occurrence of pregnancy and child health care outcomes are therefore recorded since January 1st 1994 or January 1st 2003. Migrant households account for 10 % of the pooled sample. Migrants mainly move in search of a job (about 82 %) and to California and Texas (36 and 15 %, respectively). They are mainly young males aged on average 27 who cross the border illegally: almost 84 % of them are undocumented.¹⁴ The typical migrant is either the (eldest) son of the household head (48 %) or the head of the household (30 %). Only 20 % of migrant households have at least one woman migrating.

Individual and household characteristics Table 1 presents summary statistics for individual and household characteristics of the pooled sample, for both teenagers and all women.

Table 1 Summary statistics—individual characteristics

Overall, teenagers in households with at least one member who migrated to the United States have less educated parents and live in larger families mainly in rural areas compared to teenagers in non-migrant households. No statistically significant differences between the two groups are recorded for level of education: on average, teenagers in both migrant and non-migrant households attend some years of junior high school.¹⁵ The same conclusions can be drawn for the sample of all women. These descriptive statistics are in line with much of the literature that suggests that Mexican migrants usually come from the lower middle ranges of the socioeconomic scale (Massey et al. 1994; Borjas and Katz 2007).¹⁶

Occurrence of pregnancy ENADID asks women aged 15– 54 whether they ever got pregnant, whether they gave birth or have had an abortion or miscarriage, and the number of children. The year of birth (or death, in case of death after birth) and the year of abortion are recorded. Hence, when I refer to the occurrence of pregnancy, I am exactly capturing the event of getting pregnant and not necessarily whether the woman gave birth: cases of miscarriage and especially abortion are included. Each Mexican state has a different law on abortion and this could affect the number of children actually born.¹⁷ The sample includes only women and teenagers whose partner did not migrate since January 1st 1994 or 2003. As reported in Table 2, almost 41 % of teenagers and 74 % of all women in the sample got pregnant in the last 3 years prior to the survey.¹⁸ The difference in means between migrant and non-migrant households for both samples of teenagers and all women is statistically significant: women in migrant households—especially if less than 20 years old—are less likely to get pregnant than women in non-migrant households.

Table 2 Summary statistics—outcomes of interest

Child health care Child health care is measured by likelihood of the child to be delivered by a doctor, number of visits to the doctor during the child’s first year of life, likelihood of the child to receive at least some vaccinations,¹⁹ and number of doses across all vaccinations. In Mexico, obstetric and pediatric health care provided by the public sector has high levels of effective coverage and is fairly accessible.²⁰ Thus, small variations are expected in these outcomes of interest.

Women for which this information is available gave birth to a surviving child or to a child who died at or after birth since January 1st 1994 or 2003. For each outcome, the actual number of observations depends on the amount of missing values. The reduction of the sample with respect to the previous analysis does not correspond necessarily to cases of abortion or miscarriage, or death at or after birth. Abortion, miscarriage and child mortality do not differ between the two groups: migrant households do not record either higher child mortality, abortion, or miscarriage in my sample. I can compare surviving children between the two groups.

Unlike for the occurrence of pregnancy, no remarkable differences between non-migrant and migrant households are found for number of visits to the doctor and the likelihood of receiving some vaccinations (Table 2). On average, teenagers’ children visit a doctor 3.26 times in the first year and almost 97 % of their children receive at least some doses of each vaccine, although teenagers’ children in migrant households receive a significantly lower number of doses across all vaccinations.²¹ However, teenagers and women in migrant households are more likely to be assisted by a doctor when delivering than their counterparts in non-migrant households. The difference in means is statistically significant. Overall, more than 80 % of women have their children delivered by a doctor, suggesting that still around 20 % of deliveries may have been conducted in less safe conditions.

Knowledge of contraceptive methods ENADID asks women aged 15–54 detailed information on the knowledge of eleven different methods of contraception and their use.²² A priori, information on the use of contraception is more relevant to explain the different occurrence of pregnancy between migrant and non-migrant households. Among sexually active teenagers, almost 70 % reports having used or using contraceptive methods.²³ Nonetheless, self-reported use of contraceptive methods is often unreliable: people tend to state that they use a different method than the one they actually use (Becker and Costenbader 2001). Therefore, I prefer to investigate the impact of migration on knowledge of at least one contraceptive method—it is necessary to have knowledge about at least one contraceptive method to prevent pregnancy—and infer from that the possible impact on its use: knowledge of contraceptive methods and their use are indeed positively correlated. Estimates of the impact of migration on the self-reported use of contraceptive methods are, however, provided in Table 7 in the Appendix. An initial comparison of migrant and non-migrant households reveals that migrant households know more about contraception and this difference is statistically significant. Overall, 18 % of teenagers reports having no knowledge about any method to prevent pregnancy (Table 2).²⁴

3 Empirical strategy

3.1 Econometric specification

Summary statistics show the characteristics of migrant households that are generally associated with high fertility rates and low education levels. Women, and in particular teenagers, in migrant households have less educated parents and live in larger families mainly in rural areas. Nonetheless, an initial comparison of migrant and non-migrant households reveals some differences in the outcomes of interest which suggest a lower occurrence of pregnancy and slightly higher child care for women in migrant households. These phenomena may be the result of the migration experience itself (and not of the selection mechanism into migration). In order to assess the impact of international migration on the outcomes of interest, I estimate the following regression function:

$$\begin{aligned} F_{ij}= & {} \beta _{0}+\beta _{1}{\textit{migrants}}_{j}+\beta _{2}X_{ij}+\beta _{3}\sum _{s=1}^{32}\sum _{t=1}^{2}{\textit{state}}_{js}{\textit{year}}_{jt}S_{st} \nonumber \\&+\sum _{s=1}^{32}\delta _{s}{\textit{state}}_{js}+\sum _{t=1}^{2}\phi _{t}{\textit{year}}_{jt}+\epsilon _{ij} \end{aligned}$$

(1)

where $F_{ij}$ stands for the outcomes of interest—occurrence of pregnancy, child health care and knowledge of at least one contraceptive method—for a woman i in household j. $migrants_{j}$ is an indicator of whether the household j has a member who migrated to the United States. The set of exogenous individual and household characteristics $X_{ij}$ includes age and age squared, proportion of adult males in the household, household size, whether the woman is in a relationship, whether the household is in a rural area, and woman’s level of education. I prefer to use the last education level achieved instead of years of schooling completed because the education level is often obtained prior to the migration experience and less subject to its influence. Especially for teenagers, being in school could be influenced by the family migration experience, so years of schooling completed and dropping out of school during high school (from 15 to 17 years old) or university (from 18 years old) could be potentially endogenous. Not including education at all will, however, lead to misleading results: the literature suggests that access to education by adolescent girls may delay the onset of fertility (Breierova and Duflo 2004; Durand et al. 2001). I do not control for monthly income because of its potential endogeneity: it also likely contains remittances due to the migration experience. Data on remittances are indeed inaccurate and the available information does not allow distinguishing between the amounts due to labor income and transfers within the household. Two additional controls for the mother’s education level and the presence at home of a grandmother—or another woman older than 50 years old—are included when the sample is restricted to teenagers. The rationale is that teenagers who live with a higher educated mother, or another adult women beyond the mother, may get better supervision and be endowed with a better ability to process information. $S_{st}$ is a vector of observable state-level characteristics that capture the socioeconomic development of the Mexican state of residence. It includes two measures of health care provision taken from CONAPO (1998, (2002), namely number of hospitals per 10,000 inhabitants (by state) and number of doctors per 1000 inhabitants (by state) in the two years of the survey. $state_{js}$ is a dummy for the 32 states of residence and $year_{jt}$ is a dummy for the year of the survey (1 stands for 1997 and 2 for 2006). Robust standard errors are clustered at the state-by-survey year level.²⁵

I consider separately the sample of all women and teenagers. In addition, I include the interaction between being in a migrant household and living in rural areas to assess whether there are differences between rural and urban areas. The regression function is estimated as follows:

$$\begin{aligned} F_{ij}= & {} \beta _{0} + \beta _{1} {\textit{migrants}}_{j} + \beta _{2} X_{ij} + \beta _{3} \sum _{s=1}^{32} \sum _{t=1}^{2} {\textit{state}}_{js} {\textit{year}}_{jt} S_{st} \nonumber \\&+ \beta _{4} {\textit{migrants}}_{j} \times {\textit{rural}}_{j} + \sum _{s=1}^{32} \delta _{s} {\textit{state}}_{js} + \sum _{t=1}^{2} \phi _{t} {\textit{year}}_{jt} + \epsilon _{ij} \end{aligned}$$

(2)

where $X_{ij}$ also includes the location of the household $({\textit{rural}}_{j})$.

3.2 The endogeneity problem

The parameter of major interest in Eq. (1) is $\beta _{1}$. Nonetheless, its identification is complicated. Migrant households are not randomly drawn from the population: family planning decisions and health care can be related to the characteristics of migrant households themselves. More educated and wealthier people tend to be more sensitive to family planning. They can decide to migrate to improve their children’s opportunities. As a result, migrant households would have lower fertility and better health outcomes than non-migrant households. Vice versa, if wealthier people decide not to migrate, then migrant households tend to be the poorest and have higher fertility and worse health than non-migrant households. There is a problem of potential selection of individual into migration whose direction is a priori unclear. More formally, in Eq. (1) the error term $\epsilon _{ij}$ is composed of two parts:

$$\begin{aligned} \epsilon _{ij}=\eta _{ij}+u_{ij} \end{aligned}$$

where $\eta _{ij}$ is an unobservable individual term correlated with migration and $u _{ij}$ is an unobservable individual term that is not correlated. $migrants_{j}$ in (1) is likely to depend on some factors captured by $\eta _{ij}$. A simple comparison of migrant and non-migrant households would therefore incorrectly estimate the gains or losses from migration. In order to separate its effect from the impacts of the selection mechanism, I use an instrumental variable strategy. I need an instrument that is correlated with the migration decision but uncorrelated with the individual unobservable attributes.

The instrument for current migration is potential migration. Foerster (1925) provides information about state-level migration networks from Mexico to the United States for the year 1924. There is substantial evidence that migration networks, formed as a result of U.S. demand conditions and the pattern of development of the north–south railroad system in the early 1900s, play an important role in determining current migration from Mexico (among others, Massey et al. 1994; Winters et al. 2001; Munshi 2003).²⁶ Those states which were the main sending states at the beginning of the century (Fig. 1) are still among the principal sending states nowadays.²⁷

Fig. 1

Historical migration rates and Railroad system in early 1900s. Source Hildebrandt and McKenzie (2005), Woodruff (2007)

Historic migration patterns can thus be used as an instrument for current levels of migration (Hildebrandt and McKenzie 2005). The assumption is that in a community with high historical levels of migration, a household will have a higher likelihood of having one migrant member than an otherwise identical household living in a community with low initial migration rates. Nonetheless, the use of historical migration networks at the state level may give rise to possible concerns: 1924 migration patterns may be correlated with state-level unobservables that can affect the state average outcomes. Variations at the household level would be preferable. Historical migration networks are thus interacted with the proportion of adult males (more than 12 years old) in the household, including those who have migrated and are still in the United States.²⁸ In the Mexican context, the likelihood that members of a household will migrate increases with the number of males in the household: typical migrants are the sons of the household head or the male head himself. ENADID data suggest that households with a single adult male are less likely to have a member who migrates (9 %) than households with three or more adult males (14 %), conditional on the same household size and the same number of adults in the household. For instance, take two households with five adult members (more than 12 years old) and the same household size: household A has two men and three women, household B has three men and two women. Household A is less likely to undertake migration than household B. Variations at the household level are given by the proportion of males among adult members.

$migrants_{j}$ in a reduced form framework is therefore modeled as follows:

$$\begin{aligned} migrants_{j}= & {} \alpha _{0} + \alpha _{1} potential migrants_{j} + \alpha _{2} X_{ij} + \alpha _{3} \sum _{s=1}^{32} \sum _{t=1}^{2} state_{js} year_{jt} S_{st}\nonumber \\&+ \sum _{s=1}^{32} \sigma _{s} state_{js} + \sum _{t=1}^{2} \gamma _{t} year_{jt} + \eta _{ij} + v_{j} \end{aligned}$$

(3)

where $potentialmigrants_{j}$ is a continuous variable equal to the 1924 state migration rate interacted with the proportion of adult males in the household. As an instrument for the interaction term $migrants_{j}\times rural_{j}$ in (2), I use $potentialmigrants_{j}$ interacted with the dummy for the location of the household $(rural_{j})$. The instrument affects the household’s propensity to send members abroad but is unlikely to be correlated with any unobservable household or women attributes that affect the occurrence of pregnancy and child health care (and knowledge of contraceptive methods). The assumption here is that, conditional on household demographic characteristics and state fixed effects, the proportion of adult males does not affect the outcomes of interest differently by state, but it does affect the propensity that someone in the family would migrate. One could argue that children’s gender composition influences women’s fertility and therefore the proportion of adult males in a household can affect the occurrence of pregnancy. Angrist and Evans (1998), among others, show that parents of same-sex siblings are significantly and substantially more likely to go on to have an additional child. My data do not suggest it is the case here. As reported in Table 8 in Appendix, further child-bearing is not more likely for women with the oldest two children of the same sex. No statistically significant differences are found between women with two girls compared to women with two boys either. There might be a preference for firstborn to be male, but the preference for male children fades out when passing from two to three children and on average women have 3.25 children.²⁹ The potential concern of endogeneity through children’s gender composition is of second order in this context. Moreover, this unlikely affects teenagers who, on average, have one child: only 3 % of my teenage sample has more than two kids. The coefficients of the instrument of the reduced form (3) with and without controls are reported in columns 1–4 of Table 3.

Table 3 First-stage

The first-stage results for the instrumental variable estimation show that F-statistics on the incidence of migration are above 10, suggesting that after controlling for the remaining exogenous regressors, the historical migration rates interacted with the proportion of adult males in the household can be a strong determinant of whether a household currently has a migrant member.³⁰ The coefficients of the instrument are positive, as expected, and highly statistically significant.

The estimates reported in this study are obtained through OLS and standard two-stage least squares estimation (2SLS). The IV estimation method for linear regressions has been preferred to a IV probit estimator also in the case of binary outcomes because it requires fewer distributional assumptions and consistently estimates average treatment effects in the case of binary endogenous variables (Newey 1987; Angrist 1991).³¹

4 Results

I proceed as follows. First, I investigate the impact of international migration on occurrence of pregnancy and child health care of women remaining in the country of origin. Knowledge of measures to prevent unwanted pregnancy can partly explain changes in fertility behaviors and accounts for broader health knowledge that can affect child health care outcomes. Therefore, by using the same specification, I secondly measure the impact of international migration on knowledge of at least one contraceptive method by women remaining in the sending country. This knowledge can be the result of increased access to education thanks to changes in household income due to remittances (wealth effect), the migration experience per se—the transfer of behavioral norms from receiving to sending country communities, and the reduction of the cost to migrate (teenagers actively acquire more information). My data do not allow to properly disentangle these channels. However, I can document that remittances sent back from migrants increase households’ wealth in this context. Results are reported in Sect. A.3 in Appendix.

Occurrence of pregnancy and child health care Table 4 presents the estimation results for the occurrence of pregnancy. The difference between OLS (columns 1 and 2) and 2SLS (columns 3 and 4) estimates of migrants indicates a positive selection into migration. By simply comparing migrant households and non-migrant households I would have obtained results driven by the characteristics of migrants’ households themselves: migrant households’ characteristics are generally associated with higher fertility rates than non-migrant households. The coefficient estimates for the individual characteristics confirm what is expected. The age of the woman matters only for the entire sample, with an expected inverted U-shape. The higher the education level of a woman, the lower her chance of getting pregnant, especially for teenagers. The higher the education level of her mother, the lower the teenager’s chance of getting pregnant. Being in a larger family reduces the occurrence of pregnancy for all women, while the presence at home of a grandmother—or a woman older than 50 years old besides the mother—does not seem to affect the outcome. Being in a relationship evidently increases the occurrence of pregnancy. The location of the household in a rural area does not significantly affect the outcome of interest. The proportion of adult males in the household is statistically significant here, plausibly capturing a size effect: conditional on the same household size, the higher the number of adult males, the lower the number of adult women who can get pregnant in the same household.

Table 4 Occurrence of pregnancy

Columns 3 and 4 report the 2SLS estimates for the sample of all women and teenagers separately. Despite the imprecision of point estimates, the results show that having at least one household member who migrated to the United States decreases the occurrence of pregnancy among women. This is especially true for teenagers, whose reduction in occurrence of pregnancy corresponds, on average, to 0.339 probability points. The impact observed is very large, representing a decline to 0.09 from the sample mean. Columns 5 and 6 control for the location of the household. The coefficient for the total effect of being in a migrant household for teenagers is still statistically significant at 5 % (it corresponds to $-$0.326), but the results show that there is not a differential impact due to the fact that the household is located in a rural or urban area. Moreover, teenagers in migrant households in urban areas have a lower likelihood of getting pregnant than their peers in non-migrant households, but results are not statistically significant.

Results for child health care outcomes are reported in Table 5. The only remarkable difference between migrant and non-migrant households can be found for the likelihood of the child to be delivered by a doctor in the sample of all women. Being in a household with at least one migrant is estimated to raise the probability that the child is delivered by a doctor by 0.569 probability points on average. Columns 5 and 6 control for the location of the household. The coefficient for the total effect of being in a migrant household for all women is still statistically significant at 1 % (it corresponds to 0.787) and the results show that women in migrant households in rural areas are more likely to have their child delivered by a doctor than women in migrant households in urban areas. There are no statistically significant differences in child health care for teenagers in migrant and in non-migrant households.

Table 5 Child care

Knowledge of at least one contraceptive method An increase in the knowledge of methods to prevent pregnancy partially explains the reduction in the occurrence of pregnancy. Knowledge is positively correlated with use of contraceptive methods and the use of contraception leads to a reduction in the occurrence of pregnancy. Moreover, knowledge of contraceptive methods can be associated with more general health knowledge and be informative of different child health care measures. Table 6 reports the estimates for the knowledge of at least one contraceptive method. The dependent variable is a dummy equal to one when the woman has knowledge of at least one method to prevent pregnancy.

Table 6 Knowledge of at least one contraceptive method

Being in migrant households increases the knowledge of contraceptive methods among women, especially teenagers (columns 3 and 4). More precisely, being in a migrant household increases the knowledge of at least one contraceptive method among teenagers by 0.307 probability points on average.³² The coefficient estimates for the individual characteristics confirm what is expected. The higher the education level, the higher the knowledge of at least one contraceptive method. The education level of the mother is no longer statistically significant, while the presence of a grandmother in the household increases the likelihood that the teenager will have knowledge about a method to prevent pregnancy. The proportion of adult males in the household is not statistically significant. Women and teenagers in non-migrant households living in rural areas have lower knowledge of contraceptive methods than peers in non-migrants households living in urban areas. Columns 5 and 6 show that both women and teenagers in migrant households have a higher knowledge of contraceptive methods: the coefficient for the total effect of being in a migrant household corresponds to 0.281, statistically significant at the 5 %, for all women and to 0.304, statistically significant at 10 %, for teenagers. Moreover, women in migrant households in rural areas have a higher knowledge of contraceptive methods (0.286 probability points more) than women in migrant households in urban areas.

The results are suggestive of the existence of a health information channel due to the migration experience, which can partially drive the observed reduction in teenagers’ occurrence of pregnancy and the increased likelihood of the child to be delivered by a doctor among migrant households. Teenagers in migrant households have a higher knowledge of contraceptive methods and are more likely to practice active birth control. The impact of migration on the knowledge of at least one contraceptive method is very large, corresponding to a 37.4 % increase. This result is in line with that found by De (2013): in Mexico, women from migrant families are 75 % more likely to use birth control pills and 36 % more likely to use condoms than their non-migrant counterparts. In the sample of all women, the statistically significant impacts are obtained among migrant households living in rural areas: women have a higher knowledge of contraceptive methods and their children are more likely delivered in safer conditions. Such an increased in the knowledge of birth control methods can be explained by better ability to process information due to higher access to education (wealth effect), transfer of behavioral norms through migrants’ experience from the destination to the host country, and—especially for teenagers—decision to delay pregnancy in response of a decreased cost of migration.

Women’s migration is expected to exert a stronger influence on the diffusion of contraceptive use and general health knowledge than men’s migration (Lindstrom and Saucedo 2002). I investigate whether the gender of the migrant, precisely whether the existence of at least one sister who migrates to the United States, has an impact on the outcomes of interest. Nonetheless, households with female migrants comprise only 20 % of my migrant sample and few observations can be considered in the analysis. The results are no longer statistically significant.³³

5 Conclusion

Mexican migration to the United States is a long-standing phenomenon which has seen increasing flows of migrants over time and has strongly affected Mexican society. In this paper I investigated the impact of this migration on occurrence of pregnancy and child health care for those who remain in the country of origin. The main econometric challenge is the endogeneity of migrant status, which is solved by the use of an instrumental variable approach. Potential migration is used as instrument for current household migration to the United States. The principal channels through which migration affects fertility and child health care are wealth, transfer of behavioral norms from receiving to sending country communities and cost of migration. My data do not allow to provide a complete decomposition of these channels. However, I show some evidence of the increased health knowledge—measured as knowledge of contraceptive methods—among teenagers in households that experienced migration to U.S. and I argue that these girls may be better endowed to process information (wealth effect), decide to delay pregnancy (reduced cost of migration) and are more exposed to behavioral norms of the host country of their household members (knowledge transmission to the sending countries). Overall, the results show that having at least one household member who migrated to the United States decreases the occurrence of pregnancy among teenagers. Evidence of the increased knowledge of contraceptive methods which can partially drive this phenomenon is found. Being in migrant households increases the knowledge of contraceptive methods among teenagers: they likely practice active birth control behaviors. Less significant results are obtained for child care outcomes: being in a migrant household only raises the likelihood of the child to be delivered by a doctor in the sample of all women, especially in rural areas.

A part of the impact I find is undoubtedly related to the characteristics of the context I study. The wealth effect and the reduced cost of migration are widely documented in the literature as channels through which migration affects fertility and child health care. The transfer of behavioral norms from receiving to sending communities applies especially to those countries experiencing temporary and recurrent migration, as Mexico. More detailed data providing additional information on women’ future migration and sources of knowledge of birth control methods would help to better understand the relative weight of these channels. Further investigation in this direction is encouraged and can integrate what I document in this paper.

Appendix

1.1 A.1: Use of at least one contraceptive method

Table 7 Use of at least one contraceptive method

1.2 A.2: Further child-bearing for women with two or more children

Table 8 Probability of further child-bearing for women with two or more children

1.3 A.3: Wealth

ENADID collects direct information on remittances and household income, but there are many missing values and inconsistent amounts.³⁴ In order to measure the impact of migration on household income, I therefore create a wealth index: wealth is proxied by household durables and utilities and is defined through the first principal component analysis.³⁵ I measure the impact of international migration on wealth index by using the same specification reported in the main analysis. Table 9 reports the estimates.

Table 9 Wealth index

The results show that migration leads to a higher wealth index: on a scale between $-$8.3891 and 6.0967, being in a migrant household increases wealth by 5.5 index points for all women and 2.66 index points for teenagers. Migrant households are richer and can plausibly spend more resources on human capital accumulation. Higher educated women can thus better process information and acquire more health knowledge with consequences on fertility and child health care. There is a differential impact due to the location of the household: both all women and teenagers in urban migrant households have a higher wealth index than their peers in non-migrant households.