This study explores the effects of assimilation on the health of Hispanics in the United States, using ethnic intermarriage as a metric of acculturation. I exploit a unique data set of linked confidential use birth records in California and Florida from 1970–2009. The confidential data allow me to link mothers giving birth in 1989–2009 to their own birth certificate records in 1970–1985 and to identify second-generation siblings. Thus, I can analyze the relationship between the parental exogamy of second-generation Hispanic women and the birth outcomes of their offspring controlling for grandmother fixed effects as well as indicators for second generation’s birth weight. Despite their higher socioeconomic status, third-generation children of second-generation intermarried Hispanic women are more likely to have poor health at birth, even after I account for second-generation health at birth and employ only within-family variations in the extent of assimilation. I find that a second-generation Hispanic woman married to a non-Hispanic man is 9 % more likely to have a child with low birth weight relative to a second-generation woman married to another Hispanic. These results largely reflect the higher incidence of risky behaviors (e.g., smoking during pregnancy) among intermarried Hispanic women.
Despite being poorer and having lower rates of education and employment, recent Hispanic immigrants are healthier than the average American. They tend to live longer, have lower incidence of heart disease and cancer, and give birth to healthier children than do U.S.-born natives. However, this population’s health advantage erodes with time spent in the United States despite socioeconomic assimilation. Because of the expected positive socioeconomic gradient in health, previous scholars have referred to the initial advantage observed among Hispanics as the Hispanic health paradox (HHP) and to the subsequent deterioration of immigrant health as the immigrant paradox.
A large literature has documented these health trajectories and pointed to immigrant selectivity and sociocultural protection as the main explanations for the immigrants’ initial health advantage and the subsequent health deterioration (Antecol and Bedard 2006; Elder et al. 2013; Giuntella 2013; Jasso et al. 2004; Riosmena et al. 2013; Ullmann et al. 2011). Yet, relatively little is known about the heterogeneity of this assimilation process within the immigrant population because previous studies focused on immigrant-native differences and were based on cross-sectional data with small samples. A better understanding of the health trajectories of immigrant descendants may provide additional insights on how behavioral and socioeconomic factors affect the health of the entire population. This study focuses on second-generation Hispanics and analyzes the relationship between a metric of cultural assimilation, ethnic intermarriage, and the birth outcomes of their offspring, the third generation.
As of 2012, Hispanics accounted for 17 % of the U.S. population and approximately 25 % of all births. Several studies have provided evidence of a positive relationship between health at birth and adult health and socioeconomic outcomes and a negative relationship between infant health and health care costs (Almond et al. 2005; Case et al. 2005; Conley and Bennett 2000; Currie and Moretti 2007). Children of immigrant mothers, despite having a lower socioeconomic status, have better birth outcomes than native children, but second-generation immigrant women bear less-healthy children.1 For these reasons, there is growing concern that the deterioration of birth outcomes in a large and growing segment of the American population may have important consequences on fiscal budgets (Kaushal and Kaestner 2010) and, more generally, on intergenerational mobility (Currie 2011). Furthermore, in light of the enduring debate on whether to enact more or less assimilation-oriented policies, it is relevant to investigate whether the acculturation process—which is usually associated with positive economic outcomes (Bisin et al. 2011; Nekby 2010)—may be deleterious to the health of immigrants and their offspring (Finch et al. 2007). Finally, in accordance with the epidemiological approach (Fernandez 2011), studying the effects of assimilation among immigrant descendants helps to further the understanding of the health risks associated with urbanization, the adoption of sedentary habits, and the Westernization of habits in developing countries that are undergoing major health transitions (Delavari et al. 2013; Marmot and Syme 1976).
Acculturation—usually defined as the process of adjustment in a host country—is typically associated with the acquisition of the host country’s social and cultural norms and the progressive loss of the heritage country’s culture. Cultural assimilation may affect health through the adoption lifestyle behaviors and health practices of the host country as well as the loss of traditional habits and behaviors of the country of origin. In particular, previous scholars have proposed that the adoption of unhealthy behaviors, the loss of social support, and the stress associated with the acculturation process are possible explanations of the positive association between cultural assimilation and adverse birth outcomes (Fox et al. 2015; Lara et al. 2005). Several studies have offered evidence of a protective effect of different metrics of assimilation, such as foreign-born status, ethnic density, maternal age, age at migration, language proficiency, and years since immigration on birth outcomes (Acevedo-Garcia et al. 2005; Arcia et al. 2001; Bates and Teitler 2008; Finch et al. 2007; Guendelman and Abrams 1995; Hummer et al. 2007; Osypuk et al. 2010; Powers 2013; Shaw et al. 2010). However, most of these studies were limited in their scope by sample size, the cross-sectional nature of the data, or the lack of objective and reliable measures on nativity, ethnicity, and health.
Exploiting unique administrative birth records data, I analyze the birth outcomes of third-generation Hispanics born in California and Florida, two of the top immigrant destination states in the United States. The large data set and the ability to link the records of two generations allow this study to overcome some of the aforementioned limits. In particular, using longitudinal, intergenerationally linked data allows me to control for second-generation health at birth and exploit within-family variation in metrics of cultural assimilation (e.g., intermarriage and the ethnic content of child’s first name) of second-generation Hispanic women. Specifically, I obtained confidential information on names and dates of birth that allows me to link the birth records of second-generation Hispanic women born in California and Florida between 1970 and 1985 to those of their offspring born between 1989 and 2009. Following an approach similar to that used by Currie and Moretti (2007), I identify second-generation siblings and link their records to those of the third generation. Using within-family variation, I can partially isolate the original migrant’s selectivity and account for background characteristics that are common among sisters, such as genes and family environment (see Fig. S1 in Online Resource 3). Furthermore, having linked the records of third-generation children to the records of their mothers, I can control for second-generation mother’s birth weight and partially account for differences in infant health of second-generation children that may affect their patterns of assimilation and the birth outcomes of the third generation. Finally, one of the significant advantages of this study is that most of the health outcomes considered (pregnancy outcomes and maternal health characteristics) are recorded by medical officials and are therefore not subject to self-reporting bias.
To study the effects of acculturation on the health of immigrant descendants, I use the ethnicity of the father to identify women who are intermarried (Chiswick and Houseworth 2011), and alternatively, the ethnic content of third-generation child’s first name.2 Many social scientists view interethnic intermarriage as the final step in the assimilation process because it is importantly related with the resilience of traditional behaviors and the adoption of the social norms of the host country (Gordon 1964; Qian and Lichter 2007; Sassler 2005). Therefore, intermarriage is often used as a proxy for cultural assimilation. Moreover, marrying a native has direct effects on language abilities, knowledge of the host country’s social norms and labor market conditions, and the ethnic composition of one’s social circle. Social norms and ethnic identity influence the acceptability of risky behaviors in a network (Bethel and Schenker 2005; Fenelon 2013; Unger et al. 2001). Intermarriage may therefore have direct effects on risky behaviors and health by changing the reference group and the relevant social norms affecting individual behavior. For example, U.S.-born individuals with two foreign-born parents are less likely to smoke than children of intermarried couples (Acevedo-Garcia et al. 2005).3
As in any study using siblings, possible systematic differences across siblings may still confound the relationship of interest. Marriage choice is endogenous, and unobserved factors influence whom we marry (Nekby 2010). Individuals who choose to intermarry are not a random sample of second-generation Hispanics, and a natural concern is that the intermarriage decision may be correlated with unobserved determinants of health that may be unrelated to cultural assimilation. To address this concern, I adopt an instrumental variables (IV) approach and use marriage market conditions to instrument for intermarriage. However, the goal of this article is to study the effects of cultural assimilation, to which intermarriage serves as a proxy, on the birth outcomes of third-generation children. The two-stage least squares (2SLS) estimates do not identify the causal effects of intermarriage on birth outcomes because marriage market conditions may directly affect assimilation. Yet, the IV approach allows me to rule out some potential confounding factors that are correlated with interrmarriage but uncorrelated with the IV.
I find that third-generation children of Hispanic women married with non-Hispanic men are 9 % more likely to be of low birth weight (LBW) than children of endogamously married Hispanics. The higher incidence of LBW among children of second-generation Hispanic women in exogamous relationships is surprising because of the positive association between intermarriage and socioeconomic outcomes (Furtado and Trejo 2012; Meng and Gregory 2005; Wang 2012), and because of the positive socioeconomic gradient in health (Case et al. 2002). Second-generation Hispanic women who marry non-Hispanic men live in better neighborhoods, have higher education, and are married to husbands with higher socioeconomic status. As third-generation birth outcomes correlate significantly with quality of care, socioeconomic status, and risky behaviors, we would expect children of intermarried Hispanics to have (if anything) healthier birth outcomes. Indeed, 2SLS estimates confirm the main findings that children of intermarried couples are more likely to be of LBW and, consistent with the existence of a positive socioeconomic gradient in health, are larger than the ordinary least squares (OLS) estimates.
However, I find no evidence of significant effects of intermarriage on birth outcomes when focusing on non-Hispanic native mothers, suggesting that the intermarriage “health penalty” among children of Hispanic mothers is not driven by confounding factors that could be particular to Hispanic fathers. In addition, I find no evidence of significant effects of intermarriage on second-generation Hispanics who intermarried with Hispanics coming from a different country (e.g., women of Mexican origin married to men of Puerto Rican origin). This evidence suggests that the negative effects of intermarriage on third-generation Hispanics are not explained by the reality that mixed-ethnicity marriages may be causing worse pregnancy outcomes for reasons other than assimilation (e.g., stress induced by the collision of two cultural backgrounds).
The health penalty observed among children of exogamously married second-generation Hispanic mothers can largely be attributed to their higher incidence of risk factors (such as higher rates of smoking, alcohol consumption, and hypertension) compared with their endogamously married counterparts. Intermarried second-generation Hispanic women are much less likely to maintain the health-protective behaviors and conditions that characterize the first-generation immigrant Hispanic mothers. The results point in the same direction when alternative metrics of cultural assimilation are used (e.g., the ethnic content of the third-generation child’s first name) and are robust to different model specifications. Overall, these findings suggest that despite its positive effects on economic outcomes, cultural assimilation may accelerate the adoption of riskier behaviors and, therefore, may have negative effects on the health of immigrant descendants.
This article makes two important contributions to the extant literature analyzing the relationship between cultural assimilation and infant health. First, this study sheds light on the mechanisms underlying the relationship between acculturation and adverse birth outcomes by showing that intermarried Hispanic women display less healthy behaviors during pregnancy, which in turn leads to poorer health. To the best of my knowledge, this is the first study focusing on the relationship between intermarriage and health. Furthermore, previous studies analyzing the relationship between intermarriage and labor market outcomes focused on the first generation (Furtado and Trejo 2012). Second, this study focuses on differences among U.S.-born second-generation Hispanics, thus isolating the effect of cultural assimilation from generational status and cohort effects. Accounting for initial health endowments, a vast set demographic controls, and time-invariant family characteristics, this study provides support for the hypothesis that the effects of cultural assimilation on health go beyond generational status and time spent in the host country, and cannot be entirely explained by the effects of positive selection and a subsequent process of convergence toward the mean (Giuntella 2013).
The main data used in this study are drawn from the Birth Statistical Master File provided by the Office of Vital Records of the California Department of Health and from the Birth Master Dataset provided by the Bureau of Vital Statistics of the Florida Department of Health.4 These data contain information extracted from the birth certificates of all children born in the years 1970–1985 (1970–1981 in California and 1971–1985 in Florida) and 1989–2009.5 In the sample, I define first generation as all women who were born outside the United States but who delivered in the United States between 1970 and 1985; second generation comprises their children born between 1970 and 1985; and third generation accounts for their grandchildren born between 1989 and 2009. Children are classified based on mother’s country of origin because information on father’s country of origin is not available before 1989 and is often missing thereafter. Thus, the sample of second-generation children also includes children born to first-generation Hispanic women and U.S.-born fathers. Because information on the country of origin is reported in both states only for the major sending countries, I restrict the second-generation sample to children of Cuban, Mexican, and Puerto Rican mothers.6 To match first-generation grandmothers across the different birth certificates of their second generation, I use information on the grandmother’s name, child’s last name, mother’s race, and mother’s state of birth, which implies that children born to the same mother but from different fathers would not be considered in my sample of siblings. I drop individuals for whom the matching variables are missing.7 The matching process is described in Online Resource 1.
Unfortunately, I do not have information on the father’s country of origin that spans the entire period under study and in both states. Thus, as mentioned earlier, the sample includes second-generation children born to Hispanic mothers and U.S.-born fathers. For the same reason, the main variable of interest—intermarriage—is based on the father’s ethnicity, regardless of nativity and marital status, and is defined as the union of a second-generation Hispanic woman with a non-Hispanic man.8 Prior studies (Duncan and Trejo 2007, 2009, 2011, 2012) have assessed the importance of selective ethnic attrition among Mexican Americans and have noted that selective intermarriage and ethnic attrition can produce significant downward bias in tracking the socioeconomic success of U.S. immigrant groups. Spanish surnames, although imperfect, provide a more objective measure of ethnic origin and allow me to identify individuals of Hispanic origin who did not self-report as Hispanic. Therefore, to identify the father’s ethnicity, I use the father’s surname and the latest list of U.S. Census surnames,9 which tabulates surnames that are classified by self-reported race/ethnicity and are based on 270 million individuals with valid surnames on the 2000 U.S. Census (Elliott et al. 2008; Word et al. 2008).10
I restrict the sample of third-generation children to those with nonmissing information on their father’s surname, and whose records were successfully linked to those of their mothers born in California and Florida between 1971 and 1985.11 After these restrictions, the sample consists of 346,962 third-generation children born between January 1989 and December 2009 in California and Florida.
Table 1 reports the descriptive statistics for Hispanic women in exogamous and endogamous relationships. The intermarriage rate in the sample is 18.5 %. One of the typical drawbacks of administrative vital statistics is the lack of information on individual income and occupation. However, the data contain information on parental education, the zip code of residence, and the zip code of the hospital where the birth occurred. Data on zip code sociodemographic and economic characteristics are drawn from the U.S. Census. In particular, I examine the median family income and the poverty rate as of the 1980 census for the zip code of the mother’s birth and grandmother’s residence, and also the median family income and the poverty rate as of the 1990 census for the zip code of the child’s birth and mother’s residence. Consistent with previous studies, I find that intermarried Hispanic mothers are positively selected with respect to socioeconomic status. They are better educated, less likely to live in or give birth in poor zip codes, less likely to live in ethnic enclaves, and more likely to receive adequate prenatal care.12 Hispanics in exogamous relationships are also slightly older and less likely to be unmarried.13 Similarly, non-Hispanic fathers have higher education and are slightly older than Hispanic fathers in the sample.
Despite the higher socioeconomic status, children of second-generation intermarried Hispanics have a substantially higher incidence of LBW (birth weight <2,500 g) and infant mortality14 than children of endogamously married Hispanics. They are also more likely to be macrosomic (birth weight >4,500 g), to present with abnormal conditions at birth, to be born before term (<37 weeks), and to have a low APGAR score.15 Complications during pregnancy are more likely to occur among intermarried Hispanic mothers.
Information on birth weight is consistently available throughout the period studied and in both states. Given the general consensus that LBW is an important marker of health at birth and is strongly associated with increased mortality and morbidity risk (Conley and Bennett 2000; Currie and Moretti 2007), I focus on the incidence of LBW as the main indicator of fitness at birth. However, in the robustness checks (see Online Resource 2), I consider alternative measures of infant health (see Online Resource 3, Tables S2 and S3).
Administrative records provide only limited information on health behavior during pregnancy and only for more recent years. Information on adult behaviors and health conditions is not available for the California data, whereas the Florida data report tobacco use, alcohol consumption, and weight gain during pregnancy from 1989 onward, and pre-pregnancy weight and height, chronic hypertension, gestational hypertension, and diabetes from 2004 onward. Intermarriage is associated with a higher incidence of risk factors that are known to importantly affect birth outcomes.
Intermarriage and Birth Outcomes
Overall, Table 1 suggests that intermarriage of a second-generation Hispanic woman with a non-Hispanic man is associated with a significant positive socioeconomic premium (Meng and Gregory 2005) but also with a negative health penalty. This section analyzes the effects of intermarriage on the birth outcomes of third-generation immigrants. In a later section, I present the results obtained using grandmother fixed effects and exploiting within-family variation in intermarriage. Following that, I discuss the results obtained using an IV approach, and then move on to discuss the role of father’s selectivity and selection in the marriage market.
To address the concern of selection bias arising from a matching process that selects a sample of women who were both born and have given birth in either California and Florida, I verify the external validity of the results using data from the Natality Detail Data (see Online Resource 2), which contain detailed data on all births in the United States. Robustness checks using alternative models and the ethnic content of child’s first name as a metric of acculturation are presented in Online Resource 2.
where LBW3,t is an indicator for whether the third-generation child was born with a birth weight <2,500 g. INTM2,t is an indicator for whether the father’s ethnicity is non-Hispanic. X2,t includes a set of standard sociodemographic controls: dummy variables for the mother’s and father’s education (less than high school, high school, some college, college or more); indicators for prenatal care adequacy, parity,17 marital status, maternal age dummy variables, a quadratic in father’s age, dummy variables for the interaction of county and year of birth for the second and third generations; indicators for whether mother resided in a zip code in the highest quartile of poverty rate or gave birth in a hospital located in a zip code in the highest quartile of poverty rate; the poverty rate in the zip code of birth of the second-generation; and the share of second-generation Hispanic mothers in the zip code. BW2,t describes a set of indicators for second-generation birth weight (very low birth weight, low birth weight, normal weight, and macrosomic birth). λGM represents grandmother fixed effects.
Table 2 confirms that intermarriage is correlated with a significantly higher risk of LBW. The incidence of LBW for children of intermarried couples is approximately 0.8 percentage points higher (column 1) than that of children of intramarried couples. In column 2, I include controls for second-generation birth weight using indicators for very LBW (birth weight <1,500 g), LBW (birth weight <2,500 g), normal weight (birth weight ∈[2,500 g, 4,500 g]), and macrosomic birth (birth weight >4,500 g). The coefficient diminishes by 12.5 %. Including grandmother fixed effects reduces the coefficient by an additional 25 %, yet the effect remains economically important and statistically significant (+9 % with respect to the mean of the dependent variable). The coefficient is substantially not sensitive to the inclusion of sociodemographic controls and zip code–level characteristics.18 It is noteworthy that this result is not sensitive to the addition of the father’s educational dummy variables. The results point in the same direction when alternative birth outcomes are considered, such as indicators for premature birth, birth weight <1,500 g, abnormal conditions at birth, and infant mortality (see Table S2 in Online Resource 3). There is instead no evidence of nonsignificant effects on birth weight (in grams).19 To gauge the magnitude of the effects, previous research in the United States estimated that smoking mothers are more than two times more likely to have a LBW baby than nonsmoking mothers (Almond et al. 2005; Kramer 1987); thus, the intermarriage coefficient is equivalent to 20 % of the effect of tobacco use on LBW. The coefficient suggests that the risk factors associated with intermarriage can explain approximately 15 % of the observed generational increase in the incidence of LBW between first- and second-generation Hispanics (0.07 percentage points; see Giuntella 2013).20
In light of the previous literature (Furtado and Theodoropoulos 2010; Furtado and Trejo 2012; Meng and Gregory 2005; Wang 2012) that provides evidence of a positive relationship between intermarriage and socioeconomic outcomes, the fact that the children of intermarried couples have worse birth outcomes is particularly striking. Clearly, intermarriage is not an exogenous decision, but the unobservable factors that are usually associated with the likelihood of marrying a non-Hispanic would be likely to downwardly bias the estimated coefficient. Wang (2012) reported that Hispanic–white couples, on average, earn approximately $20,000 more than Hispanic–Hispanic couples. Indeed, in the data, intermarriage is positively correlated with median family income (0.15) in the zip code and with the mother’s education (0.10) and father’s education (0.14), whereas it is negatively correlated with the zip code’s poverty rate (−0.17). Table 3 shows how the positive correlation between intermarriage and socioeconomic status holds even after the inclusion of standard demographic controls (panel A) and grandmother fixed effects (panel B). Intermarriage is associated with higher education (defined as having attended some college, or having a college degree or more), higher average income in the zip code of residence, lower poverty in the zip code of residence and in the hospital zip code, and a higher likelihood of moving to a neighborhood better than the zip code of birth with respect to the quartiles of poverty rate. With grandmother fixed effects included (panel B), the intermarriage coefficient on all the socioeconomic outcomes becomes smaller in absolute value, but it still economically and statistically significant. Moreover, using the data from Florida that contain limited information on risky behaviors, Table S6 (Online Resource 3) shows that tobacco consumption during pregnancy is negatively correlated with high education among Hispanics and non-Hispanics, and results go in the same direction if considering zip code–level economic indicators (e.g., median income or poverty rate). Thus, one may expect that if fixed effects are biased, they are biased toward finding that intermarriage reduces the incidence of LBW.21
Age is defined in nine-year intervals so that for a 25-year-old woman, the marriage market will include all men between the ages of 23 and 31 and all the women between the ages of 21 and 29. The key identification assumption is that the instrument affects third-generation births only through its effect on intermarriage. Immigrants do not locate randomly, and the existence of ethnic enclaves may be importantly correlated with the maintenance of healthy behaviors. However, the inclusion of indicators for second-generation birth weight, the extensive set of controls (in particular, the inclusion of MSA fixed effects), time-varying zip code economic characteristics, and the yearly share of Hispanic mothers in the zip code of residence should mitigate the concern that the exclusion restriction may be violated.
Column 1 of Table 4 presents the first-stage statistics. Standard errors are adjusted to account for MSA-level clustering. A 10 percentage point (about 1 standard deviation; see Table 1) increase in the percentage of first-generation immigrant men from the same country of origin as the woman in the same MSA/age group decreases the probability of the woman marrying a non-Hispanic by approximately 10 percentage points. The F statistic is well above the conventional thresholds for weak instruments. Columns 2 and 3 report the OLS and grandmother fixed-effects estimates presented in Table 3 for comparison.
Second-stage results (column 4) confirm the positive coefficient of intermarriage on LBW incidence. The point estimate is larger than the ones obtained using OLS (column 2) and grandmother fixed effects (column 3), suggesting that Hispanics who marry within the group are, if anything, negatively selected on unobservable characteristics. These results are consistent with the existence of a positive socioeconomic gradient in health and the fact that intermarriage has positive effects on socioeconomic assimilation (Furtado and Trejo 2012; Meng and Gregory 2005). Including sociodemographic controls (column 5) significantly reduces the coefficient, yet the estimate remains significantly larger than the OLS estimate. The result is robust to the inclusion of MSA (column 6) or county fixed effects (column 7). Thus, despite the possibility of differential selection of women across cities over time, the identification strategy is not subjected to omitted variable bias induced by persistent MSA characteristics that may be correlated with both women’s location choices and their health. Exploiting only cohort variation in the share of immigrant men in a woman’s age group residing in her MSA, the coefficient increases in absolute value. Again, these results are consistent with the conjecture that Hispanic women are positively selected into exogamous relationships.
Note that the IV does not allow us to identify the direct effects of intermarriage on birth outcomes because the share of immigrant men in a woman’s age group residing in a given MSA may be related with Hispanic women’s assimilation through other channels. The assimilation of second-generation Hispanic women could be influenced by neighborhood characteristics (Borjas 2000). Because the goal of this article is to estimate the effects of cultural assimilation on birth weight (not intermarriage on birth weight), I am not concerned about bias resulting from selective marriage based on assimilation at the time of marriage. Yet, the 2SLS estimates allow me to rule out some of the other factors that may affect both the marriage patterns of second-generation children and the birth outcomes of third-generation children. Intermarriage may cause LBW through other potential mechanisms that are unrelated to assimilation. For instance, one may think of prenatal stress associated with the collision of two different cultural backgrounds. In the following section, I discuss this alternative mechanism.
The results presented in Table 2 may still reflect the selectivity of migrants on the father’s side. By controlling for grandmother fixed effects and the mother’s birth weight, I can partially account for maternal selectivity, but the intermarriage coefficient might nonetheless be the result of the genetic advantage carried by the Hispanic father. Furthermore, a natural concern is that mixed-ethnicity marriages may be causing worse pregnancy outcomes for reasons other than assimilation. For instance, several recent articles have examined the effect of stressful events that occur in utero on birth outcomes, finding evidence of important adverse effects (Black et al. 2014; Currie and Rossin-Slater 2013; Torche 2011). Mixed-ethnicity marriages may produce more stressful pregnancies if two cultures collide over issues regarding the grandchildren.
To rule out the idea that unobservable traits of Hispanic fathers are inherently good for a child’s birth weight as well as the reality that intermarriage may affect birth outcomes through channels other than assimilation (e.g., stress), I examine the effect of intermarriage on U.S.-born non-Hispanic white women (born in the same years as second-generation Hispanic women analyzed so far) and focus on the birth outcomes of their offspring. If the intermarriage coefficient captures the effect of unobservable traits of Hispanic fathers that are inherently good for a child’s birth weight, one should expect to find a protective role of having a Hispanic father even when analyzing the effects of intermarriage among children of non-Hispanic white women. Similarly, if intermarriage had a causal affect unrelated to assimilation, this effect should also affect white non-Hispanic women who marry Hispanics.23 When conducting this test among non-Hispanic white women, I find that marrying a non-Hispanic (Hispanic) man does not have significant effects on the risk of LBW—and if anything, is associated with a lower (higher) incidence of LBW (see columns 3 and 4 in Table 5).
Perhaps other unobserved factors affect both selection in the marriage market and birth outcomes and unrelated with acculturation. However, the inclusion of indicators for second generation’s birth weight, of grandmother fixed effects, and the rich set of sociodemographic controls reduce the concern that these confounding factors could significantly alter the primary finding.24
Selection in the Marriage Market
Another relevant concern is that selection in the marriage market might be substantially different for men and women. However, despite significant gender differences in intermarriage rates between blacks and whites, Wang (2012) found no significant gender differences in the intermarriage rates of Hispanics and whites: white men who married Hispanic women are not less educated than those who married white women. In particular, 32.3 % of white men married to white women completed college education, compared with 33.1 % of white men who married Hispanic women. One may still be worried about the selection of non-Hispanic men on unobservable characteristics (e.g., behaviors) that may not necessarily be correlated with socioeconomic status. Given the limited information on paternal behaviors, I cannot directly test whether the intermarriage health penalty is driven by the negative selection of non-Hispanic men married to Hispanic women along other dimensions. Yet, although the incidence of risky behaviors (e.g., smoking) differs markedly between Hispanic and non-Hispanic white women, the differences are less marked and are nonsignificant among men (Agaku et al. 2014). Furthermore, using the Tobacco Use Supplement of the Current Population Survey (TUS-CPS) and controlling for standard sociodemographic characteristics (e.g., age, education, and state of residence), I find no significant differences in the prevalence of tobacco use between non-Hispanic men married to Hispanic women and non-Hispanic men in endogamous relationships.25
Other Robustness Checks
Table 6 considers the ethnic content of child’s first name and the share of Hispanic mothers in the zip code as alternative metrics of cultural assimilation (see Online Resource 2). Finally, to verify the external validity of the results obtained using the data from California and Florida, I conduct a similar analysis using the Natality Detail Data (see Table 7 and Online Resource 2).
Intermarriage may directly affect the adoption of risky behaviors by changing the reference group and facilitating the adoption of the social norms prevailing among U.S.-born non-Hispanic women. To further investigate the possible mechanisms underlying the negative (positive) relationship between intermarriage and third-generation birth outcomes (LBW), I use the Natality Detail Data to exploit information on maternal behaviors and risk factors that is available in the birth records of all U.S. states.
Table 8 illustrates the incidence of these risk factors among non-Hispanic white natives, non-Hispanic blacks, and immigrants of Hispanic origin. First-generation immigrants have a substantially lower incidence of risk factors compared with non-Hispanic white natives. Hispanic descendants show some convergence toward the less-healthy behaviors and higher incidence of risk factors of non-Hispanic white natives, but they retain a fairly sizable health advantage. Yet, for both the first and second generation, intermarriage is associated with a higher incidence of risk factors and less distance to the native behavioral norm.26
Table 9 focuses on smoking during pregnancy, which has been widely recognized as the most modifiable risk factor for LBW (Almond et al. 2005; Currie and Schmieder 2009) and has been shown to be one of the major factors explaining the life expectancy advantage of first-generation Mexican immigrants in the United States (Fenelon 2013) and, more generally, immigrant health trajectories (Leung 2014). The results presented in Table 9 confirm that intermarriage is importantly associated with smoking during pregnancy and that this correlation holds up when sociodemographic controls are included (columns 1–3). Using the data from Florida that contain limited information on risky behaviors, I investigate whether the correlation between intermarriage and tobacco consumption during pregnancy is robust to the inclusion of grandmother fixed effects (columns 4–7). Columns 6 and 7 show that when grandmother fixed effects are included, the coefficient decreases from 2 percentage points to 1.3 percentage points, but the effect remains economically important and statistically significant. Being intermarried is associated with a 60 % higher likelihood of tobacco use during pregnancy. The results for other risk factors, such as alcohol use and gestational hypertension, tend toward the same direction. However, in these cases, the magnitude and the significance of the coefficient are less robust to the inclusion of grandmother fixed effects and sociodemographic controls.27
Table 10 shows that when this set of risky behaviors is controlled for, the coefficient of intermarriage on LBW reduces substantially. Panel B focuses on smoking during pregnancy, and panel C includes smoking, alcohol consumption during pregnancy, and gestational hypertension.28 When observable risk factors are controlled for (panel C), the OLS estimate becomes nonsignificant, while the 2SLS coefficient shrinks by approximately 32 %. The estimates do not identify causal effects of tobacco use (or of the other observed behaviors) because there may be unobserved behaviors correlated with the observed ones. Yet, the similarity of panels B and C, and the fact that the coefficient diminishes but remains positive and significant when alcohol use and gestational hypertension are controlled for separately (see Table S8 in Online Resource 3) suggest that smoking during pregnancy likely explains an important part of the negative effect of cultural assimilation on health outcomes. This finding is consistent with recent studies on the role of smoking in explaining the healthy immigrant advantage (Fenelon 2013; Leung 2014), The unexplained part of the intermarriage coefficient is likely to be related to other types of behavior, such as dietary habits, for which I do not have data but that are known to significantly affect birth outcomes and to deteriorate with time spent in the United States and across generations.29
This article analyzes the effects of second-generation assimilation on the birth outcomes of third-generation Hispanics in the United States. Among second-generation Hispanics, intramarried couples exhibit higher resilience in terms of healthy behaviors, health conditions, and birth outcomes. Using ethnic intermarriage as a metric of cultural assimilation, I show that third-generation children of intermarried Hispanic couples are 9 % more likely to be of LBW than children of intramarried couples. This finding holds true even after I account for potential confounding factors, focusing on a subsample of second-generation siblings and controlling for grandmother fixed effects. Furthermore, the results do not appear to be driven by unobservable traits associated with the father’s ethnicity or by the fact that intermarriage may affect birth outcomes through channels other than assimilation. The father’s Hispanic ethnicity has no protective effect for children of non-Hispanic women, nor does it have a significant effect on Hispanic women intermarried with Hispanic men of different origin. Instead, the evidence suggests that these differences are explained by the higher prevalence of risky behaviors (e.g., smoking) among second-generation mothers in exogamous relationships. Similar results are obtained with alternative metrics of cultural assimilation based on the ethnic content of third-generation child’s first name and on the ethnic composition of the neighborhood.
Overall, these results highlight the importance of policies that encourage the maintenance of healthy behaviors among immigrant descendants. Educational and prevention programs targeting Hispanic mothers may have important effects, particularly on those who are highly acculturated, and thus have a significant potential to reduce these risks. These policies may partially counteract the tobacco industry’s increasing efforts to market cigarettes to Hispanic immigrants in the United States (Acevedo-Garcia et al. 2004; Osypuk and Acevedo-Garcia 2010).
Finally, these results are also relevant to sending countries, such as Mexico, that are undergoing important epidemiological transitions. As health transitions occur more rapidly in a migrating population, the health trajectories observed among immigrants can help predict the risks associated with the typical public health problems related to urbanization, the diffusion of sedentary habits, and substance abuse in developing countries.
I am grateful to Francisca Antman, Randall Ellis, Delia Furtado, Rania Gihleb, Kevin Lang, Claudia Olivetti, and Daniele Paserman for their comments and useful suggestions. I am also thankful to Yasmine Serrano (Florida Department of Health) and Julie Turner (California Department of Public Health), who were extremely helpful with the data collection process. I would like to thank all the participants at the AlpPop 2015 Conference, the 2015 annual meeting of the Population Association of America, as well as all seminar attendees at the University of Manchester and the University of Oxford. Any errors are my own. The project was made possible by generous funding from the Boston University Institute for Economic Development.
In a companion study (Giuntella 2013), I showed that modest positive selection on health at the time of migration can account for the initial advantage in birth outcomes of second-generation Mexicans. Moreover, a simple process of regression toward the mean reverses the apparent paradox and predicts a greater deterioration than that observed in the data. Yet, the convergence would have occurred more quickly if immigrants had not socioeconomically assimilated or if they had fully assimilated in the incidence of risky behaviors during pregnancy.
Throughout the article, I use the term “intermarriage” loosely because I use only the father’s ethnicity regardless of the actual marital status.
Economists have shown growing interest in studying the determinants of intermarriage (Bisin and Verdier 2000; Bisin et al. 2004) and the effect of family structure on the economic assimilation of immigrants (Baker and Benjamin 1997; Eckstein and Weiss 2002; Furtado and Trejo 2012; Meng and Gregory 2005). Most studies have focused on first-generation immigrants, suggesting that marrying a native is positively associated with the success of immigrants in the labor market. These studies have provided evidence that intermarriage may improve job prospects and increase the rate of economic assimilation by facilitating the adoption of the host country’s customs, improving language proficiency, and expanding social connections (Furtado and Theodoropoulos 2010).
To gain access to the data and conduct this study, I obtained approval from the Florida Department of Health Institutional Review Board, the California Health and Human Service’s Committee for the Protection of Human Subjects, and the Boston University Institutional Review Board, where I conducted the analysis.
I obtained data from the California Department of Public Health for the years 1970–1981 and 1989–2009.
However, when I considered individuals whose mothers were born abroad and reported Hispanic origin, the results are substantially unchanged. In addition, only 0.06 % of second-generation Hispanic women in the sample reported black race. Excluding them from the analysis did not change the results.
Regarding the matching of mothers to grandmothers, in California, I matched only one daughter in 84 % of the cases, I matched two daughters in 12 % of the cases, and I matched three or more daughters to each grandmother in 4 % of the cases. In Florida, I matched only one daughter in 80 % of the cases, I matched two daughters in 17 % of the cases, and I matched three or more daughters to each grandmother in approximately 3 % of the cases. Over the entire sample, the average number of children matched to each mother is 1.91; the average number of grandchildren linked to each grandmother is 2.50, and this figure would be 4.20 if it were conditioned on linking at least two second-generation sisters to their offspring.
Because the information on father’s race is missing for 40 % of the sample, I do not use it in the main analysis. For those observations for which information on father’s race is available, only 2.37 % of second-generation Hispanic women had a black partner. Similarly using national data from the Natality Detail data (see the Possible Mechanisms section), I find that approximately 96 % of the non-Hispanic fathers of children born to Hispanic mothers report a white race, 3 % report black race, and 1 % report other races (American Indian, Alaskan, Asian, and Pacific Islander). Among those for whom information on father’s country of birth is available, 98 % of non-Hispanics were born in the United States. Controlling for father’s race or excluding black fathers from the analysis does not affect the main results.
Data are available online (http://www.census.gov/topics/population/genealogy.html).
Using the 2000 U.S. Census Bureaus surname list, I consider as Hispanic any surname with a posterior probability of being associated with Hispanic ethnicity higher than .7, but the results are robust to the use of different thresholds. Furthermore, using self-reported ethnicity to define intermarriage yields similar findings.
Information on father’s surname is missing for approximately 13 % of third-generation births. However, information on self-reported ethnicity is available in 99 % of the sample. In Online Resource 3, I show that using father’s self-reported ethnicity to define intermarriage yields substantially identical results (see Table S8).
Adequate prenatal care is defined as beginning prenatal care in the first trimester of pregnancy. Access to first-trimester care is a conventional way to assess the access to care for pregnant women (http://www.hrsa.gov/quality/toolbox/introduction/coreclinical/index.html).
Note that the information on marital status is available only for a subsample of the observations and relies on apparent marital status.
For this measure, I consider children who were reported dead within a year. However, most of the children who died in their first year of birth died in their first week.
The APGAR scale is determined by evaluating the newborn baby on five criteria (appearance, pulse, grimace, activity, and respiration) on a 10-point scale.
The number of times a woman has given birth.
When the estimates are broken down by country of origin, the coefficient on intermarriage is positive and significant for Mexicans (0.004; SE = 0.002), although it is positive but not statistically significant for Cubans (0.006; SE = 0.007) and Puerto Ricans (0.003; SE = 0.012). Using information on father’s country of origin to define intermarriage and excluding U.S.-born Hispanic men results are similar (0.009; SE = 0.009), but the results are not precisely estimated given that we have information on father’s country of origin for only a subsample of the data.
Because LBW due to shortened gestational age has different causes and different short- and long-run consequences than those cases due to intrauterine growth restriction, I estimate the baseline specification of column 4 controlling nonlinearly for gestational weeks. The estimated effect is only marginally smaller (0.004; SE = 0.002).
Restricting the analysis to third-generation children born after 1995 (2000), the point estimates are larger but not statistically different from the main results illustrated in the article (see Online Resource 3).
Similar results are obtained when analyzing alcohol consumption, although the differences are not significant for gestational hypertension. Results are available upon request.
Using only first-generation immigrants or including all individuals reporting a given ethnic origin, which includes second and later generations, yields similar results.
However, it is possible that a pregnant Hispanic woman married to a non-Hispanic man feels additional stress from her child’s loss of ethnicity (i.e., many third-generation Hispanics do not speak Spanish).
The result on white non-Hispanic women is robust not only to the addition of the father’s and mother’s education but also to the separate analysis of women who married equally, more-, or less-educated Hispanic men (results available upon request). In addition, when controlling for grandmother fixed effects and the previous generation’s birth weight, I find that marrying a high school dropout has no significant effect on the risk of LBW among children born to non-Hispanic U.S.-born mothers.
The TUS-CPS is an NCI-sponsored survey of tobacco use and policy information that has been administered as part of the Current Population Survey (CPS) since 1992. The CPS is a monthly survey conducted by the U.S. Census Bureau for the U.S. Department of Labor, Bureau of Labor Statistics. Results are available upon request.
Although the prevalence of smoking among Hispanics is lower than the prevalence among non-Hispanic whites, smoking prevalence among Hispanic men is substantially higher than among Hispanic women and is not statistically different from the prevalence observed among non-Hispanic whites (Agaku et al. 2014). Hispanic women are also much more likely than non-Hispanic white women to quit smoking during pregnancy (Osterman et al. 2013).
These results are available upon request.
Including information on maternal weight gain and maternal diabetes leaves the estimates practically unchanged. The results are available upon request.
For instance, maternal dietary practices are significant determinants of health at birth, and there is evidence of unhealthy assimilation with respect to dietary habits (Guendelman and Abrams 1995). Using the American Time Use Survey, I find evidence that among immigrants of Hispanic origin, intramarriage is positively associated with time spent on food preparation and consumption.