While academic and policy circles have given much attention to the assimilatory experiences of Mexican immigrants in the United States, less is known about those who stay behind—an especially unfortunate oversight given the increasing number of Mexican youth with migrant family members. Of the studies on this topic, most have sought to identify the effect that migration has on youths’ migratory and educational aspirations, often using qualitative methods in individual sending communities. The present article supplements this research in two ways: (1) in addition to assessing educational outcomes, the scope of the analysis is expanded to include nonmigrants’ interaction with another homeland institution of upward mobility: the labor market; and (2) using a large demographic data set, statistical techniques are employed to adjust for unobserved selectivity into the migrant family-member population, thus accounting for a potentially serious source of bias. The results suggest that youth in migrant-sending families are less likely to complete the educational transitions leading up to postsecondary school and have a lower probability of participating in the local economy. The results also indicate that unobserved factors play a “nonignorable” role in sorting youth into migrant and nonmigrant families.
The magnitude of Mexican migration into the United States has increased steadily over the past 30 years (Office of Immigration Statistics 2005). The assimilatory experiences and driving forces behind this wave of immigration have generated a large body of scholarship in the family and migration literatures. Until recently, however, academics paid little attention to those for whom migration has a more indirect effect: the family members who stay behind. Analyses of “the other half” have focused predominantly on the macro-level implications of out-migration for economic development, weighing the reduction of productive labor capacity against the aggregate financial gains from remittances (Taylor 1999). Relatively little is known about the social and community-level effects of migration in sending areas, and less still about whether and how efforts abroad shape the individual behaviors and ambitions of family members remaining back home.
In this article, I begin to explore these questions. My principal objective is to determine whether and to what extent immediate exposure to international migration influences how remaining family members—particularly youth—interact with the educational system and local labor markets. This is an especially timely question given the number of people crossing the border from Mexico into the United States—an outflow that totaled nearly 3 million persons between 1995 and 2000 (Passel et al. 2004)—and the amount of resources that these individuals channel back to their places of origin (Sana 2008). It is also a question of considerable scholarly relevance. Although the specific effects of migration are complex and sometimes contentious, there is little doubt that the movement of people and goods reorganizes social, cultural, and economic structures in sending areas—a transformation that may complicate the meanings and value nonmigrants attach to homeland institutions of upward mobility.
The implications of this stance are elaborated most prominently by culture-of-migration theorists (Kandel and Massey 2002; Massey 1986, 1987; Mines and Massey 1985), who connect foreign wage labor with the cross-national transmission of capital, ideas, and culture that combine to promote future migratory behavior. These analyses anticipate a negative correlation between family involvement in migration and the willingness of children to invest in resources associated with upward mobility, particularly schooling. As migratory behavior becomes normalized, the prospect of participating in homeland institutions is eclipsed by more profitable opportunities in foreign markets. Evidence supporting this position has come primarily from fieldworkers in individual communities (Alarcón 1992; Cohen 2004; Levitt 2001; Mines 1981; Reichert 1982; Rouse 1992; Smith 1998; Wiest 1973). To date, only a few quantitative studies have yielded insight into the micro-level implications of the culture-of-migration thesis (Kandel and Kao 2001; Kandel and Massey 2002).
From another point of view, it is plausible to suppose that family member migration might have the opposite effect. Youth in transnational families may approach school and labor as a way to “do their part,” effectively acknowledging through diligence and hard work the sacrifices of migrant family members. This hypothesis has some sway among feminist scholars of migration and the family (Hondagneu-Sotelo and Avila 1997; Parreñas 2001, 2005a). Parreñas (2005a), for example, detected a tendency among youth in transnational families to achieve highly in school as a way of reciprocating the efforts and contributions of remitting relatives. Benefiting from repatriated earnings, there is also reason to imagine that youth in migrant-sending families might fare comparatively better in school because of their ability to offset or otherwise defray the opportunity costs associated with continued enrollment. Edwards and Ureta (2003), López-Córdova (2005), and Taylor (1999) have speculated that this may in fact be the case.
In investigating these hypotheses, this article makes two primary contributions. First, rather than focusing exclusively on the education of nonmigrant children, I extend the analysis to include their labor market activity. In taking a wider view of migration and its implications for nonmigrant children, I am able to evaluate the extent to which the theoretical frameworks described earlier can be generalized to stratifying institutions other than the school system. Second, and perhaps more important, I rely on a series of endogenous switching regression models and Heckman’s index sufficient method to measure and adjust for potential selection bias. The inability of prior studies to address selectivity has, with few exceptions (Hanson and Woodruff 2003; McKenzie and Rapoport 2007; Miranda 2007), led to results that potentially conflate the effect of migration with the effect of unobserved factors—such as economic need and/or a shared sense of ambition and responsibility within the family—that promote migratory behavior in the first place.
The remainder of this article is organized into five sections. After briefly reviewing the literature on individual- and community-level effects of out-migration, I introduce and describe my data source. Next, I develop two multivariate models to estimate the impact of family migratory behavior on youths’ educational attainment and labor force activity. I then summarize results from these models, focusing in particular on questions related to self-selection. In closing, I recapitulate my main findings and suggest various avenues for further research.
Three Approaches to Studying the Effects of Migration in Sending Areas
As issues surrounding international migration have come to the foreground in recent decades, scholarly orientations have become increasingly diverse. Although the bulk of the literature has sought to predict migratory behavior by using competing theoretical models (Fussell 2004; Fussell and Massey 2004; Massey 1990a, b; Piore 1979; Portes and Walton 1981; Stark 1991; Todaro 1976), or to situate the experiences of contemporary immigrants within debates over assimilation (Alba and Nee 2003; Gans 1992; Hirschman 2001; Portes et al. 2005; Portes and Rumbaut 1996; South et al. 2005; Zhou 1997), an emerging line of research has begun to consider the meaning and impact of migration from the perspective of the homeland.
The existing research on sending areas can be divided roughly into three groups. The first, and most familiar, understands migration as an avenue for capital flow (Delgado Wise and Márquez 2007; Durand et al. 1996; Massey and Parrado 1994; Orozco 2002; Taylor et al. 1996a, b). Conceptually, this approach seeks to isolate and draw conclusions about the mechanisms through which economic changes in migrant-sending areas occur. For instance, Durand et al. (1996) examined the effects of foreign earnings, or “migradollars,” on development in a set of Mexican communities. Using a case study approach, the authors concluded that the financial resources that migration affords have a positive effect on households’ budgets, leading to increased consumer spending and, consequently, additional demand in the labor market for new workers.
Others working in this tradition have taken a decidedly more pessimistic stance, characterizing migration as a sort of self-perpetuating “syndrome” in which sending areas serve as “nurseries and nursing homes” for their mostly migrant labor force (Reichert 1981; Russell 1992; Stuart and Kearney 1981). Rather than triggering economic growth and productive investment, these scholars suggest that a large share of remitted earnings are used for recurrent expenses and consumer goods (Canales 2007), a tendency that increases import demand and fosters an overall climate of economic dependency (Kapur 2005). However, as Taylor (1999) and others have pointed out, this understanding rests on a rather restrictive (and arbitrary) definition of productive spending, which largely ignores the importance of human capital investment. Although it may be true that migrant-sending families rely on remittances to augment consumption, households may also use these monies to finance the continued schooling of younger family members. Recent work by Edwards and Ureta (2003), Taylor and Mora (2006), and López-Córdova (2005) gives some support to this conjecture, showing that remittances relax credit constraints among receiving families, enabling parents to invest more freely and more heavily in their children’s education.
Owing to a web of cross-national social relations, remittance-based economies also have important sociocultural implications for nonmigrants. This observation, which identifies migration as a mode of cultural diffusion within transnational fields, orients the second approach (Guarnizo 1994; Guarnizo and Díaz 1999; Levitt 1998, 2001; Portes and Rumbaut 1996; Roberts et al. 1999). Rather than concentrating on the net economic effect of migration, research rooted in transnationalism calls attention to a more complex interplay between emigrants and their places of origin. For example, in a multi-sited qualitative study of the transnational Dominican community, Levitt (1998) traced streams of “social remittances” from emigrants to their homelands, contending that the micro-level flow of culture—moving in tandem with material goods, money, and people—makes up an important and often overlooked component of global transfers. Social remittances, according to her argument, include normative belief structures, various systems of practice and conduct, and social capital. Together with repatriated earnings, these resources are thought to alter the incentive structure in sending communities as nonmigrants grow increasingly reliant on support from abroad (Levitt 2001).
Massey et al. (1994) reached a similar conclusion when examining Mexican sending communities. As migration gains prevalence, these authors argued, places of origin experience extensive cultural and socioeconomic transformations. With time, the inflow of ideas and customs creates a “culture of migration” (Massey et al. 1998), in which aspirations and conceptions of community, familial relations, and economic responsibility take on new, “transnationalized” meanings. These ideas and customs, in turn, lead nonmigrants to disinvest in traditional institutions of upward mobility with the intention of capitalizing on more attractive alternatives in other countries. Using survey data from the Mexican state of Zacatecas, Kandel and Massey (2002:1002) recently corroborated this hypothesis, concluding that intimate exposure to migratory behavior prompts nonmigrants “to look northward rather than locally for opportunities and social mobility.”
Finally, a third framework recasts migration as a behavior that transforms the practices and meanings of family life. Proponents of this approach point out that young people’s actions are often responsive to or constrained by the changing needs, interpersonal dynamics, and composition of migrant-sending families (Hondagneu-Sotelo and Avila 1997; Meza and Pederzini 2008; Parreñas 2001, 2005a). For example, in her case study of the Philippines, Parreñas (2005a) examined the effect of the global “care deficit” on nonmigrant dependents. Although emotional stress and hardship were commonplace in transnational families, children in the study received emotional support from abroad, as well as from parental surrogates within the community. Even more interesting, at least for the purposes of this article, Parreñas (2005a) observed a tendency among youth in migrant-sending families to achieve highly in school as a way of reciprocating the hard work of remitters.
Of course, reciprocating the efforts of family members living abroad need not only (or even primarily) entail improved performance in the classroom; young people may also signal their support by taking on new adult roles and responsibilities. If, for instance, the emigration of an adult earner is followed by a short-term decrease in financial resources, youth may choose to invest more heavily in income-generating activities in an effort to stabilize the household economy (Duryea et al. 2007; Latapí and González de la Rocha 1995; Meza and Pederzini 2008; Skoufias and Parker 2006). Likewise, if the departure of a family member creates an unmet need within the household for chores or other domestic duties, children may shift their time into home production, resulting in interruptions in their schooling and potential declines in educational attainment (Meza and Pederzini 2008; Parreñas 2005b).
Taken together, these three perspectives articulate a general account of international migration and its effect on sending areas. Cross-border movements of capital, people, ideas, norms, and emotional support clearly reorganize the social, cultural, and economic playing field in places of origin. These changes have concrete consequences for individuals—and especially for children. Nevertheless, these literatures contain a number of inconsistencies and opportunities for elaboration. For example, is there a desire to repay efforts abroad by succeeding in school, as Parreñas and other scholars suggest? Or, following the culture-of-migration argument, might young adults opt out of the education system to pursue more profitable opportunities in foreign markets? From a financial perspective, does the migration of a family member—and remittances that ensue—serve to offset the opportunity costs associated with nonmigrants’ continued enrollment? Finally, moving beyond education, to what extent does the likelihood of joining migrant family members in foreign markets alter nonmigrants’ interactions with other homeland institutions, particularly local labor markets?
Although prior research has spoken to some of these questions, the empirical base for doing so has been largely restricted to qualitative studies in specific sending areas. This is problematic for at least two reasons. First, focusing on a limited number of communities in a subset of Mexican states makes it difficult to validate more generic theoretical arguments about migration and its implications for nonmigrants. Although the present research makes no claim to resolving this issue in its entirety, it does represent a deliberate move in that direction. Second, existing studies generally do not distinguish the effect of the treatment (e.g., family member migration) from the effect of unobserved factors that jointly determine migratory behavior and children’s educational attainment and economic activity. Instead, most researchers assume that unmeasured variables—including shared family traits, aspirations, and expectations—do not influence both the outcome of substantive interest and family-level migratory decisions. If this position proves to be unjustified, as a handful of recent studies have suggested (Hanson and Woodruff 2003; McKenzie and Rapoport 2007; Miranda 2007),1 then inferences concerning the effect of family member migration may be biased. In the next two sections, I introduce a data set and analytic framework that will help to better address these concerns.
Description of the Data
Data for this analysis were drawn from the Integrated Public Use Microdata Series-International (IPUMS-International) archive, a collaborative project that houses microdata from Mexico’s XII General Population and Housing Census, 2000 (Minnesota Population Center 2008). The census, administered in February 2000 by the Instituto Nacional de Estadística, Geografía e Informática (INEGI), was the first to ask respondents to indicate the number of family members who left to go live in another country during the five years preceding enumeration. Respondents were to include persons who might have migrated only for a short time or who might have already returned by the time of the census.
The IPUMS-International sample used a stratified cluster design based on enumeration areas and localities. Constructed from 100% of the long-form questionnaires, the data set represents a 10.6% anonymized sample of the total noninstitutionalized population, which amounts to approximately 10 million person records and just over 2.3 million households. The sample was designed to yield representative statistics for all localities with 50,000 inhabitants or more, including all 2,443 of Mexico’s municipalities. As is the case with other IPUMS data products (Ruggles et al. 2008), logical edits and probabilistic hot-deck imputation methods were applied in rare instances where values were either inconsistent or missing entirely (Esteve and Sobek 2003). Weights were provided to inflate the sample to the total population.
The Mexican census data, like all data, are not without limitations. Relying on a large-scale, cross-sectional sample makes it impossible to fully characterize the dynamic nature of demographic processes such as international migration, and it makes it equally difficult to articulate precise statements with respect to underlying mechanisms. Furthermore, variable availability and the wording of particular census items occasionally serve to complicate interpretation. The indicator used to identify migrant member households, for example, lacks the specificity and detail to ascertain which family member migrated; their familial relationship to the focal child; the exact timing of their departure; and, where applicable, their return. Although (as I describe in more detail later) steps are taken to mitigate some of these limitations, the nature and content of these data nevertheless impose restrictions on the scope of analyses that can be conducted.
These imperfections notwithstanding, the Mexican census data promise at least two attractive advantages. First, the census provides a sample of transnational families that is sufficiently large to allow for complex statistical analyses and, in particular, empirical tests for unobserved selectivity. In all, more than 1.1 million households have at least one international migrant, and nearly 300,000 households have more than one. Second, the demographic, social, and economic characteristics that were collected reflect the age, sex, place of residence, family composition and background, migration history, educational attainment, labor force participation, and various types of earnings (including those derived from foreign remittances) for all members of each household. No other data set, at present, has included all these variables while also using such a large and geographically representative sampling frame.
Sample and Measures
My analytic strategy draws on a series of multivariate models to estimate the role of family member migration in determining (1) youths’ educational attainment, measured in terms of their probability of moving from one educational level to the next; and (2) the likelihood that a youth participates in the labor force. To minimize any complications introduced by youth who themselves migrate or who otherwise leave their family of orientation prior to enumeration,2 I restricted the sample to unmarried adolescents ages 15–18 who were not heads of household and who were living with one or more of their nuclear parents.3 The sample was further limited to include only noninstitutionalized and native-born persons, a technique that yielded a weighted sample of 6,433,326 youth from 4,117,484 families. For each youth in the sample, I matched the child’s record with variables relating to the child’s household, the child’s siblings, the head of the household, and the head of household’s spouse.
To operationalize the first outcome, I separated the Mexican school system into a sequence of successive educational transitions: from primary (elementary school) to lower secondary (middle or junior high school); and from lower secondary to secondary education (high school).4 I defined the dependent variable by a successful transition (1 = those who matriculated; 0 = those who elected to stop), given that the child completed the previous level. Individuals who did not complete a given level were deemed “ineligible” for future transitions and were coded to missing for all subsequent transitions. For instance, lower secondary was coded 1 for respondents who had completed primary and were enrolled in lower secondary, or any higher level of education; 0 for respondents who had completed primary but never enrolled in or completed a year of lower secondary; and “missing” for respondents who did not complete primary. These transitions represent milestones in the schooling process that signify movement across institutional divisions in Mexico’s educational system. As Mare (1980) and others have shown, one of the advantages of this approach over other strategies, which commonly regress years of schooling on social origins, is that background variables may exert different amounts of influence at various points in a student’s scholastic career.
To measure labor force activity, I relied on a binary variable to indicate the child’s employment status during the week preceding the census. Because youth labor encompasses a wide range of activities, which vary considerably in terms of time commitment and intensity, I restricted the definition of employment to remunerated labor of greater than or equal to 20 hours per week. Given the necessarily arbitrary nature of any cutoff, I wanted to be as inclusive as possible in recognizing the economic activities of youth and young adults. In examining the distribution of weekly hours spent working, I found that moving the threshold down to 10 or 15 hours would add only a very small fraction of youth to the universe of economically active. The specific variable of interest, a detailed measure of employment, reports the labor force activity of all respondents, even those whose primary activity was something other than work. The measure is thus more responsive to younger workers, particularly school-age adolescents, who participated only in secondary economic activities.
The main “treatment” variable, a binary indicator of family member migration, was constructed from a question asking respondents to report whether one or more family members had engaged in international migration during the previous five years. To get a sense for how this largely arbitrary time horizon might influence my results, I ran a parallel set of analyses (not shown) in which migrant-member families were identified as those who acknowledged at least one incidence of family member migration within the past five years and/or reported receiving at least one dollar in remittances from family members living abroad. The inclusion of the latter group—which effectively extends the allotted five-year time span outward—added a relatively trivial number of cases to the migrant-member population, and did not alter my findings in a substantively meaningful sense.
To control for determinants of educational attainment and labor force activity, I include measures of children’s demographic characteristics (gender and indigenous group membership), social background, dwelling characteristics, geographic and community-level characteristics, and family composition. Because Mexican youth with more highly educated parents, or with parents who have higher earnings, are likely to spend more time in school and less time working (Levison et al. 2001), the models controlled for social background by using variables for total family income and the maximum of the father’s and mother’s years of completed schooling.5 Dwelling characteristics comprise a variety of proxies for the wealth and standard of living in the household that income might not otherwise capture, including dummy variables for the existence of piped sewage disposal, a private telephone line, and the dwelling’s flooring material (0 = dirt; 1 = concrete).
The community-level covariates consist of dummy variables indicating the type of locale (urban or rural) and the state of residence. Including these regressors is particularly important given that geographic differences in labor market conditions and educational opportunities are likely to influence the enrollment and employment decisions of youth as well as family-level migratory behaviors. Unfortunately, because of data limitations, it was impossible to determine whether the family lived in its enumerated area for the entire five-year period in question, although preliminary analysis revealed that only 6.1% of families claimed residence in a different state in 1995. Following Blake (1989) and the extensive literature on family disruption and schooling outcomes (see, e.g., Astone and McLanahan 1991), the family composition variables contain measures for the number of dependents in the family and their age structure, as well as an indicator for households with only one parent present. Accounting for these determinants gives a rough control for the interaction between children’s employment and enrollment status, on the one hand, and variability in the care and material needs of the family on the other.
Selection on Unobservables Into the Migrant Pool
One non-experimental approach to resolving this inconsistency makes use of the endogenous switching regression model. The switching model is designed to demonstrate the effects of a categorical variable, such as family member migration, on an outcome when the outcome and observed variable are thought to be jointly determined by some unmeasured factor or set of factors (Winship and Mare 1992). Social scientists have invoked this procedure in a number of settings, including analyses of curriculum tracking and students’ achievement (Gamoran and Mare 1989; Mare and Winship 1988), marital dissolution and women’s economic outcomes (Smock et al. 1999), labor market sector and wages (Sakamoto and Chen 1991a, b), interracial contact and racial attitudes (Powers and Ellison 1995), and earnings among Hispanic migrants (Tienda and Wilson 1992).
The latter two disturbance covariances, σ02 and σ12, are indicative of a common but unmeasured factor that influences both migratory behavior and remaining family members’ educational outcomes. When σ12 > 0, there is evidence for positive and nonignorable selection on unobservables into the migrant family member pool. Such a finding would suggest that youth in migrant-member families are more likely to matriculate (net of observed social, demographic, and economic variables) than randomly selected individuals from the larger population. When σ02 < 0, there is evidence for positive unobserved selection into the no-migrant family member pool because the covariance implies a negative correlation between successful transitions and belonging to a migrant-member family (Mare and Winship 1988). More generally, if either σ02 or σ12 is nonzero, then estimates obtained using conventional approaches to modeling the impact of family member migration would be biased because of unobserved selectivity.
Provided that there are theoretically grounded reasons for doing so, excluding some of the elements in w from the outcome equations can improve the precision with which estimates of σ02, σ12, β0k, β1k, and πk are identified (Mare and Winship 1988).6 For my purposes, findings from prior research suggest at least two plausible restrictions. First, on the basis of studies that stress the salience of social networks and contextual factors to the migratory enterprise (Massey 1986, 1987; Stecklov et al. 2005), let w (but not x) contain municipal- and state-level measures of the percentage of households with at least one migrant family member and the percentage of household receiving remittances. Second, following Durand and Massey (1992) and Root and De Jong (1991), both of whom identified the life-cycle stage of the household as a strong predictor of family migratory behavior, let w (but not x) include indicators of the head of household’s age. Excluding these variables from the outcome equations implies that their effect on youths’ educational transitions and labor force participation is either weak (conditional on the control variables described earlier and contained in x) or indirect, operating via family-member migration. The data support this assumption, yielding extremely slim empirical associations between the additional “pretreatment” variables in w and the outcomes of substantive interest.7
To complete the second stage, Eqs. 8 and 9 are estimated over the subsamples for which y0 and y1 are observed.10 In the next section, I draw on results from these models and from the more naive approach described in Eq. 1 to demonstrate the extent to which unobserved variables influence family member migration and Mexican youths’ interactions with institutions of upward mobility, and to quantify the effect of migration net of measured and unmeasured factors.
Table 1 summarizes the descriptive statistics for each of the variables in my analysis, disaggregated according to family migration history. The first three sections provide gross differences in the control variables. Relative to those in no-migrant families, these tabulations suggest that youth in migrant-sending families have parents with lower levels of educational attainment and income, tend to reside in rural homes with fewer amenities, are more likely to be female, and tend to live in households with only one parent present. With respect to the additional pretreatment variables in the selection equation, which are shown in the fourth panel, it is not surprising that these youth are also likely to reside in municipalities and states with disproportionately high rates of out-migration and remittance receipts.
Of more importance for the present analysis are the terms for education and labor force activity. As the fifth section of Table 1 demonstrates, the transition rate among youth in families with migrants is consistently lower than for those with stationary family members. For instance, roughly three-quarters of eligible youth in no-migrant families made the transition from lower secondary to secondary school, compared with only 59.6% of eligible children in the migrant family member group. Such gross differences are virtually nonexistent, however, when considering patterns of youth labor force participation. In fact, rather than decreasing for adolescents in the migrant family member subgroup, as culture of migration theorists might anticipate, rates of labor force activity are actually higher among youth in nonstationary families. Whether these gross differences hold in a multivariate context, and to what extent unmeasured selectivity shapes the results, will occupy my focus in the remainder of this section.
Results From a Conventional Regression Approach
Table 2 reports uncorrected probit estimates of the family member migration effect on educational transitions and participation in the labor force. Here and throughout, I evaluate statistical significance according to the more conservative Bayesian information criterion (BIC). The Bayesian approach, which extracts a penalty proportionate to the sample size (Raftery 1995), is well-suited for the present analysis given the unusually large number of cases involved. To ease interpretation, marginal effects are shown in the third column of each panel, indicating the estimated change in the probability of an outcome for a unit change in one of the predictors, holding all others to the sample mean.
The results are suggestive in two respects. First, the coefficient on family member migration, albeit somewhat small in magnitude, is statistically significant and negative in both of the models predicting educational transitions. The probability of matriculating from lower-secondary to secondary school among youth in families with at least one migrant member, for example, is 4 percentage points less than it is for their counterparts in nonmigrant families, net of individual, familial, and geographic characteristics. Second, if the analyses were to end here, one might reasonably conclude that family member migration has little practical effect on Mexican youths’ labor force participation, as evidenced by the statistically significant but substantively modest coefficient in the rightmost column. Although this finding would be generally consistent with the negligible gross differences observed in Table 1, it says nothing about the role unobserved factors play in “matching” youth to migrant and no-migrant member families.
Testing for Nonrandom Selection on Unobservables
To assess whether unobserved factors bias the inferences drawn from the uncorrected models reported previously, Table 3 provides results from endogenous switching regressions predicting the probability of successfully completing educational transitions. As discussed earlier, these estimates—unlike those in Table 2—are adjusted for nonrandom assignment to migrant and nonmigrant families on both measured and unmeasured variables. The slope coefficients for the former generally behave similarly regardless of migration status and are consistent with expectations. The only indicator that does not fit this description is indigenous group membership, a measure whose sign and significance varies depending on family member migration status.
The coefficients associated with σ02 and σ12, given near the bottom of Table 3, contain information pertaining to unmeasured factors that influence migration and matriculation. Recall that if unobserved determinants of family member migration are unimportant in explaining differences in children’s educational outcomes, these parameters will equal 0. This does not appear to be the case in either transition model. Instead, the estimated error covariances are each positive and statistically significant according to the BIC. Substantively, this implies that youth who actually belong to migrant-sending families are more likely to matriculate from one educational level to the next than would a random sample of children with identical measured characteristics, whereas young people who live in no-migrant families are less likely to do so. Thus, efforts that fail or are otherwise unable to adjust for selection on unmeasured factors—such as the regression analyses reported in Table 2—run the risk of understating the magnitude of the negative migration effect.
A similar set of conclusions emerge from the switching regressions predicting labor force participation, summarized in Table 4. Here, again, the parameter estimates on the control variables for migrants and nonmigrants tend to parallel one another quite closely, suggesting that the net effects of familial, demographic, geographic, and socioeconomic characteristics on youths’ labor force participation are largely homogenous with respect to family member migration. The exceptions include the effect of piped sewage and the presence of young children in the household, both of which obtain significance for one group (youth in nonmigrant families) but not the other.
Turning to the disturbance covariances at the bottom of Table 4, the fact that σ02 and σ12 are statistically significant and show a positive sign implies that there is positive selection into migrant families and negative selection into no-migrant families, net of measured variables. Put differently, if a random sample of Mexican youth with equal individual-level and background characteristics were placed in migrant member families, these youth would be less likely to work than those who actually reside in such families. On the other hand, if the same sample of youth were matched to no-migrant families, they would be more likely to participate in the labor force than the children who in fact live in such households. Either way, these coefficients cast doubt onto the accuracy of the results obtained from the uncorrected probit model and the descriptive analysis presented earlier. Although the exact nature of the selection mechanism cannot be ascertained from these data, it is clear that nonrandom selection on unobserved factors attenuates the negative association that would otherwise obtain between family member migration and youths’ labor force participation.
Quantifying the Family-Member Migration Effect
Figures 1 and 2 display the effect of family member migration on youths’ likelihood of making educational transitions. On each graph, the horizontal axis gives baseline probabilities of matriculating from one educational level to the next (or various counterfactual scenarios in which there are no migrant family members), and the vertical axis gives the change in probability associated with living in a migrant-sending family. The average treatment effect for the treated is plotted in solid black; the lower and upper bounds on this effect, which were derived using the 95% confidence interval (CI) on the parameter estimates obtained from the switching regression models, are displayed by the dashed lines.
Two patterns in particular stand out. First, both figures show that family member migration is negatively and strongly related to youths’ probability of matriculation, with effect sizes generally ranging from −.10 to −.20 for the transition to lower secondary and −.15 to −.30 for the transition to secondary school. Second, the parabolic shape of the plots implies that the migration effect diminishes gradually as children’s baseline probability of matriculating tends toward 0 or 1. In other words, if a young person’s eventual school continuation decision is a foregone conclusion—that is, if the individual is positioned at either end of the horizontal axis—then the experience of family member migration will have less of a bearing on whether the child transitions than would otherwise be the case.12
Figure 3 presents the analogous graph for nonmigrant youths’ economic activity. The graph shows an inverse relationship between family member migration and youths’ probability of paid employment, a finding that stands in contrast to the descriptive analyses presented earlier. Furthermore, unlike the estimate obtained from the uncorrected probit model, the change in probability associated with family member migration appears to be substantial, falling between –.15 and –.35 for the majority of “treated” youth. To better indicate the magnitude of this effect, consider a hypothetical child whose baseline probability places her on the margin between working and not working. Inspection of the graph reveals that net of measured and unmeasured characteristics, exposure to family member migration would more than halve her probability of labor force participation, from .50 to just more than .20.13
This article has presented analyses of the effects of international migration on Mexicans who remain in their place of origin. Two main results emerged. First, by using nationally representative data from the 2000 Mexican census, I found that the emigration of family members has a large, negative, and statistically significant effect on educational attainment and labor force participation among nonmigrant youth. Second, I found that estimates of these effects—and, in some instances, the inferences that follow—are sensitive to how one deals with selection on unobservables. More specifically, unmeasured heterogeneity works to dampen the estimated effect of family member migration on youths’ educational outcomes and labor force participation.
To what underlying mechanism may we attribute these findings? Overall, the empirical evidence that I have presented would seem consistent with a culture of migration argument, in which the intergenerational transmission of migratory expectations reorients nonmigrants away from homeland institutions of upward mobility and toward opportunities in foreign markets. Rather than prolonging their schooling as a result of relaxed financial constraints, immediate exposure to migratory behavior seems to encourage nonmigrant youth to take on a transnationalized view of the opportunity structure. Like Levitt (2001) and others have argued, the path to social mobility for these individuals is not tethered to their ancestral homes and communities, and this reality appears to have an immediate and substantial influence on the educational and economic strategies they enact.
This is not to deny the possibility that other mechanisms are also operating. That family member migration decreases the likelihood that youth participate in local labor markets could, in part, be symptomatic of the ways in which roles and responsibilities are distributed within migrant-sending families. A reduction in paid labor among nonmigrant youth, for example, might reflect an increased demand for housework, childcare, or other (unpaid) domestic duties, and, at the same time, a decreased demand for adolescent earnings. Unfortunately, data constraints prevent me from exploring this possibility further. To remedy this situation, an obvious next step would be to replicate my analysis using data that feature a more detailed indicator of youth labor, preferably one that is sensitive to variation in the intensity and type of work activities (both paid and unpaid) in which young people engage.
Several other aspects of this study also suggest opportunities for further research. First, the cross-sectional nature of the census data used in the foregoing analysis makes it difficult to establish a clear temporal ordering, and thus precludes an outright claim to causality. Nevertheless, insofar as it is unlikely that the outcome variables under consideration have a direct effect on whether family members migrate, the findings are at the very least suggestive that family member migration causally influences—in the classic, counterfactual sense of the word—nonmigrant youths’ educational outcomes and economic activity. Subsequent work would do well to investigate this issue further, presumably using a longitudinal research design.
Future research also stands to benefit from incorporating more refined measures of municipal-level characteristics, particularly those pertaining to labor market conditions and the availability of educational opportunities. This would be useful not only as an additional check on the robustness of my results but also in illuminating whether, to what extent, and in what ways contextual factors moderate the association between family-member migration and nonmigrants’ work behaviors and schooling decisions. Does the promise of gainful employment in the local labor market, for instance, attenuate the negative migrant family-member effect on youths’ economic activity? Likewise, how might the quality, quantity, and accessibility of local schools and educational services change young people’s calculus concerning migration, formal education, and socioeconomic mobility?
Finally, researchers could usefully explore the types of questions posed here from a comparative-historical perspective, using similar census microdata sets that are currently available for other countries, such as Colombia, Ecuador, Panama, and South Africa (Minnesota Population Center 2008). Doing so may help to shed light on whether the migration effects documented earlier are site specific or amenable to more generic theoretical applications across space and time, as well as contribute further insights into how and under what conditions the cross-border movement of people influences the behaviors and future orientations of individuals who remain in their places of origin. Either way, I hope that the results and analytic techniques described herein will help to move this important line of work forward.
Earlier versions of this paper were presented at the 2006 meetings of the American Sociological Association and the 2007 meetings of the Research Committee on Social Stratification and Mobility (RC28). I am grateful to John Robert Warren, Scott Eliason, Jennifer C. Lee, Elaine M. Hernandez, Chris Uggen, and Teresa Swartz for their helpful comments and suggestions; and to the Minnesota Population Center for its invaluable research support. All errors and omissions, however, are solely my responsibility.
Hanson and Woodruff (2003), for example, found that family member migration is influenced by some of the same hard-to-observe factors that contribute to the educational attainments of nonmigrant children. It is unclear whether the same is true for youths’ decisions concerning labor force participation.
In theory, for a youth to count as a migrant and enter the analysis sample, he or she would have to (1) temporarily out-migrate; (2) return to Mexico prior to enumeration; and (3) do so before turning age 19. Although the Mexican census data offer no way to formally assess how often this situation occurs, other analyses using different data sources suggest that it is infrequent. Durand et al. (2001), for example, found that more than 90% of temporary Mexican migrants are older than age 18 upon their initial departure. Thus, if 10% of temporary Mexican migrants are 18 and younger (thereby meeting the first criterion), one can presume that an even smaller fraction of temporary Mexican migrants satisfy the first and second criteria.
Auxiliary analyses indicated that only a small proportion of Mexican youth start their own household by age 18.
Again, because individuals are increasingly likely to themselves engage in migratory behavior as they approach adulthood, these analyses are not able to assess the final transition in Mexico’s education system (e.g., the move from secondary to university-level studies). This is a regrettable but necessary restriction in scope.
The four levels of total family income—a variable that does not include earnings that children themselves contribute, nor income derived from foreign remittances—correspond to quartiles in the income distribution.
The elements in x and w need not be entirely disjoint. Thus, in addition to specifying exclusion restrictions, the right-hand side of the selection equation also contains indicators of parental education, age structure and number of siblings, urban-rural status, dwelling characteristics and amenities, and indigenous group membership. See supplementary appendix table for more details.
In addition to examining empirical associations between the outcomes of interest and the pretreatment variables at the bivariate level, I regressed each of my dependent variables onto the socioeconomic, demographic, geographic, and family compositional variables included in x, as well as the additional predictors in w. In all three models, the resulting parameter estimates for the pretreatment variables were substantively trivial in magnitude. These results are available upon request.
By construction, z and λ in Eq. 7 are inversely related, such that a larger estimated value for z implies a smaller λ.
A Huber-White sandwich estimator was used in the second stage to correct for heteroskedasticity (White 1980).
The standard errors generated by the switching regressions may be deflated because of the presence of an estimated quantity (the inverse Mills’ ratio) in the second-stage probit models. Because of the unusually large sample size, however, it is unlikely that any such bias would meaningfully alter my substantive conclusions.
I evaluated Eq. 10 by using the mean values of the k characteristics in x and w.
In supplementary analyses, I estimated average treatment effects on the treated separately for boys and girls. For both educational transitions, the resulting estimates did not differ significantly by gender. Although not completely unsurprising, this result is consistent with at least some previous work (Kandel 2003; Kandel and Kao 2001), and may reflect, at least in part, Mexico’s closing gender gap in compulsory education.
Given what is known about the gender composition of the Mexican labor force, one might reasonably surmise that this pattern reflects heterogeneity in the relationship between family member migration and youths’ tendency to engage in paid labor, most notably between nonmigrant boys and girls. That is, females would seem more likely to be situated near the origin of the horizontal axis, and thus be relatively less susceptible to the influence of family member migration. An auxiliary by-gender analysis (not shown) bore out this speculation. Although the family member migration effect was negative for both boys and girls, the index sufficient estimate obtained for the female subsample was smaller in magnitude, resulting in a flatter and less pronounced curve. As mentioned, this finding is not entirely unexpected, particularly given the low rates at which females participate in the Mexican economy and the large differences between boys and girls in terms of their domestic roles and responsibilities (Levison et al. 2001).