Parental Migration and Child Health in Mexico



Parental Migration and Child Growth in Mexico

(an abbreviated version for the Inequality Working Group)

Jenna Nobles*

University of Wisconsin, Madison

ABSTRACT. This study uses nationally-representative, longitudinal data from Mexico to examine the effect of parental migration on child health. I compare children within households who experience parental migration at different key periods of physical development. This comparison provides an estimate of the migration effect net of those family characteristics that are related both to migration and to child health. Using this type of approach addresses the concern that parents who choose to migrate are selected on a number of key characteristics that also affect child well-being. Results suggest that parental migration negatively affects child height-for-age, a long-term measure of child nutritional status and illness.

*jnobles@ssc.wisc.edu. The author would like to thank Elizabeth Frankenberg, Duncan Thomas, Rob Mare, Amar Hamoudi, Rubén Hernández-León, Graciela Teruel, and Luis Rubalcava and participants at seminars at UC-Berkeley, UW-Madison, Yale University, and University Iberoamericana for their very helpful comments on this project. An earlier version of this study was presented at the Population Association of America meetings. The author gratefully acknowledges funding from the National Science Foundation and the Robert Wood Johnson Foundation. All errors and opinions are those of the author.

INTRODUCTION

The relationship between migration and development in sending communities has assumed a central place in social science debates. The more and more visible presence of undocumented migrants in the United States, Western Europe, and Southern Africa and the increasingly large size of remittances they return to sending communities, highlights the possibility for migration to serve as a key pathway to development in resource-constrained settings. One of the focal questions in these debates is whether adult migration improves the well-being of children in the next generation. Traditionally, remittances are considered an important mechanism for improvements in children’s access to education, nutrition, and health care. However, migration introduces other constraints to households by allocating most other household responsibilities, including child care and home maintenance, to the remaining parent. For this reason, it is not a priori clear that parental migration should improve child well-being. I address this question in the setting of Mexico by focusing on the health outcomes of children who stay behind when parents migrate.

Mexico is an important setting in which to answer this question. The proportion of children affected by parental migration is substantial. My own research suggests that 17 percent of Mexican children are expected to experience the migration of a father by the age of 14 (Nobles 2008). Because female migration is becoming more common in Mexico, it is likely that the proportion of children experiencing household migration is actually even larger. Similarly, the size of the remittance flow to communities in Mexico is large. Annually, over twenty billion dollars enter Mexico from international migrants. Internal migrants often remit considerable sums to families as well, though the overall magnitude of these transfers is much smaller (Boucher, Stark and Taylor 2005).

In light of these dramatic figures, several studies have investigated the educational and health outcomes of children in sending communities. Yet, each of these studies faces a common methodological obstacle: migration is a choice made by individuals or families, and it is possible that parents who migrate are different from parents who do not on a number of characteristics that also affect child well-being.

This is particularly problematic for studies set in Mexico. A large literature documents the phenomenon typically referred to as the “healthy migrant paradox” in the United States. Evidence suggests that this paradox, the surprisingly good health of Mexican immigrants given socioeconomic status, is driven by the selective nature of migration; migrants are healthier than non-migrants in Mexico (e.g., Palloni and Morenoff 2001). If children of migrants are in relatively better health than children of non-migrants, then the “effect” of migration on child health estimated through an ordinary least squares regression will be upwardly biased.

My study attempts to address this type of bias by comparing children over time and within households, and thereby sweeping out the fixed inputs to child health at the household level. The approach is commonly used in literature evaluating public health and welfare interventions at the community-level. However, by applying it to families, I can estimate the effect of parental migration on child health, net of fixed characteristics associated with a parent’s choice to migrate.

I focus on children’s stature, a long term measure of health that is particularly responsive to nutrition and illness during the first few years of life. The physiology of child height, and numerous nutrition interventions, demonstrate that stature is largely shaped by inputs accrued by the age of 3 or 4 (Martorell 1995; Schroeder et al. 1995). The age-dependent aspect of this health measure facilitates my comparison of children within households. Specifically, children who are exposed to migration at older ages effectively serve as a “control” group, while children exposed to migration at younger ages will have heights that incorporate the effect of being exposed to parental migration. Child height is also a useful health outcome to explore given previous findings that it predicts later life advantage on a number of measures.

As mentioned, the mechanisms through which migration may affect child health are not limited to remittances. Accordingly, my discussion of these mechanisms draws on the large literature connecting household structure, family transitions, and child well-being as well as literature examining development in the sending regions of migration streams.

FAMILY DISRUPTION, REMITTANCES, AND CHILD WELL-BEING (pulled)

MEASURING THE EFFECTS OF MIGRATION (pulled)

METHODS

Data

This study uses data from two waves of the Mexican Family Life Survey (MxFLS). MxFLS is a longitudinal, nationally-representative household survey collected in Mexico that follows migrants into the United States. The first wave was collected in 2002 and interviewed over 8,300 households in 150 communities across Mexico (Rubalcava and Teruel 2004). The second wave was fielded in 2005, with the intention of reinterviewing all members of the original households. The reinterview process involves tracking adult movers to their new locations, including those moving to the United States and those splitting off from their original household and forming new households.

MxFLS data include migration histories on both domestic and international migration for all adults. “Permanent” migration histories (moves of more than one year in duration) are collected beginning at age 12. “Temporary” migrations (moves that last more than one month but less than a year) are collected for the two years prior to both survey dates: 2000-2002, 2003-2005.

MxFLS data also include a number of objective health measures, including height, weight, and hemoglobin levels. I focus on height, which is measured directly by trained anthropometrists for all household members. For children under two years of age, height is measured while the child is recumbent.

My sample is children aged 0-8 in 2002 and children aged 0-11 in 2005. The 2005 sample includes the second observation for the children in 2002, as well as those children born between the waves. I limit the sample to children with two parents in the household in 2002 (determined retrospectively for children born between waves) and to children whose parents did not migrate between 1994 and 2002. The pooled sample of 2002 and 2005 data results in 6,707 observations.

There is some missing data on height for children – about 15% in both waves. This level of missing data is similar to a number of other nationally-representative household surveys that collect anthropometry. By limiting the sample to children who have height values and data on parental migration in both waves (except for children born in the interim who, by definition, only have 2005 data), I lose about 38 percent of the randomly drawn sample. In the appendix, I show results from analysis of the degree to which my current analytical sample is selected on various individual and family characteristics.

Measures

I measure child health by using height. Previous research suggests that height is a sensitive indicator of health status for children in Mexico. Many children are at risk for serious illness because of malnutrition. Nationally representative data from 1998 reveal that 18 percent of children under the age of five are stunted, or have a height value below two standard deviations under that of a well-nourished child of the same age. Of those in households characterized by low socioeconomic status, nearly 40 percent of children under age five are stunted (Rivera et al. 2004).

Because height varies systematically with age and gender, I standardize respondents’ measurements against the median values for children of the same age and sex, but from a well-nourished population. Specifically, I assess Mexican children’s height relative to sex- and age-specific height medians of children in the United States, using data from the National Center for Health Statistics (2000).[1] For each child, I compute a z-score that expresses the child’s height as the number of standard deviations that the child is above or below the median for a child of that sex and age in the United States. Most children have two z-score measures, which are from 2002 and are from 2005. Children born after the 2002 survey have one z-score, calculated from the 2005 height measurement.

Children’s exposure to parental migration is measured with a dichotomous variable capturing whether either of the children’s parents migrate outside of the community, without the child, for at least one month between 2002 and 2005. Because of the small number of moves (about 5 percent of this sample of children experiences a parent’s move during this period), I do not distinguish between domestic and international migration.

The Effect of Parental Migration on Child Height-for-Age

Previous research and basic intuition suggests that migration strategies are not random across households, but are instead related to household characteristics, such as socioeconomic status and to community characteristics, such as infrastructure and investment opportunities (Durand et al. 1996; Zahniser 2000). Because these characteristics are likely to affect children’s health, simply comparing children in migrating households with those in non-migrating households will likely produce biased estimates of the effect of parental migration. In a regression framework, a number of characteristics that are associated with both migration and children’s health, such as parental education and income, can be controlled. Characteristics that are unobservable, such as parental preferences for investing in children, or changes in the local economy, are more difficult to control. If the estimated “effects” of parental migration are in fact being driven by an underlying characteristic that affects both child health and parental migration, then any policy aimed at altering migration patterns to improve children’s welfare would not have its intended effect.

I turn to the MxFLS data for evidence of selection. I restrict the data to children in families in which parents have not migrated between the time the children are born and 2002. I then compare the characteristics in 2002 of children whose parents subsequently migrate between 2002 and 2005 with the characteristics of children whose parents do not migrate. These comparisons are presented in Table 1. When I examine the children of migrants and children of non-migrants using the MxFLS data, I find a few key observable differences between the groups. While the children of migrants appear slightly advantaged with respect to health, the difference is not statistically significant. However, children of migrants are younger and have better educated parents than children of non-migrants.

The results suggest that non-migrant families are different from migrant families in a few observable ways, before migration occurs. It is important, then to control for factors such as parental education when relating parental migration to child health. Inclusion of these covariates, however, does not guarantee that I have controlled for all of the ways in which migrants and non-migrants differ, some of which may be unobservable.

I address this issue by exploiting the physiology of child growth. As previously mentioned, the first few years of life are highly significant in determining child height. For this reason, the heights of children who experience parent migration between the ages of 0 and 4 should be more sensitive to the effect of migration than the heights of children who are older than 4 when they experience parent migration.

By comparing the difference between younger and older siblings within households, I am able to hold constant time-invariant inputs of parents that may affect both the decision to migrate and children’s health. Specifically, community and household characteristics that affect children’s health and that differ between children of migrants and children of non-migrants, but that are additive and fixed over time, are held constant, regardless of whether they can be measured. The comparison of children within households is accomplished by including a family fixed effect into the regressions predicting child height-for-age. Including fixed effects is conceptually equivalent to including a dummy indicator for each family in the regression.[2]

Additional sources of bias could arise if younger children have systematically higher height-for-age values than older children, or if the 2005 measurements are systematically higher than the 2002 measurements. Making comparisons between younger and older children over two points in time facilitates addressing this type of bias.

Specifically, the last several decades in Mexico have witnessed significant advances in children’s education and health outcomes. Were I to limit my analysis to changes over time for only very young children, I might falsely attribute to migration an effect that arises because of a time trend. The difference in height-for-age values of older children in 2002 and older children in 2005, however, should capture period changes without capturing a migration effect. Accordingly, this difference is subtracted from the comparison of younger children across time points.

Similarly, assessing children’s height relative to age- and sex-specific medians of another population (in this case, the United States) addresses the fact that child age and child height are systematically related. However, in most cases, growth patterns in resource-constrained countries are often different from the growth patterns in well-nourished countries like the United States (Pinstrup-Andersen, Pelletier and Alderman 1995). For this reason, comparing younger children’s height-for-age to that of older children within a household may capture an age effect common to all children, as opposed to a migration effect. Therefore, it is important to compare the difference between older and younger children at the first point in time, 2002, and subtract this difference from the comparison of older and younger children at the second point in time, 2005. Because this difference-in-difference is mathematically equivalent to the difference-in-difference used to remove the period/time effect from the effect of parental migration, no additional subtraction is necessary.

To make these comparisons, I divide children into four birth cohorts, based on their age in years in 2005 (0-2, 3-5, 6-8, 9-11). I pool children’s observations from 2002 and 2005, and estimate an interacted model predicting child height-for-age (θihtc) as a function of dichotomous indicators of birth cohort by year of measurement (CT[pic], where c indicates the 4 cohorts and t indicates 2002 or 2005), an indicator of whether a child’s parents migrated between 2002 and 2005 (Mh), and a family-level fixed effect (μh).[3]

1.

Figure 1 helps to explain this approach. The lexis diagram displays the four birth cohorts in diagonal sections across time (x-axis) and by age (y-axis). The two points of height measurement occur in 2002 and 2005. The horizontal line intersecting the diagram at age four denotes the point after which the effect of migration (or any other event) should have a significantly reduced effect on child height-for-age. I focus specifically on parental migration between 2002 and 2005. To facilitate the comparison of children who experience migration at a young age with those who do not, I restrict the sample to children whose parents did not migrate between the child’s birth and the survey date in 2002. This removes families in which older children were exposed to migration during the critical young ages. Similarly, I restrict the sample to children who have both parents in the household in 2002. For children of divorced or separated parents I cannot determine whether the absent parent was migrating during the 1995-2002 period. I, therefore, exclude children with separated, divorced, or widowed parents in 2002.

Health inputs have the largest impact on height in the period of a child’s life before age four (below the horizontal dashed line in Figure 1). This biological regularity shapes our expectations about the cohorts for whom height should be affected by a parental migration between 2002 and 2005. Cohort 1 has aged entirely out of the critical young years by 2002, so parental migration should not matter for them. Cohort 2 is exposed to migration at a vulnerable age for, at most, the first year of the 2002-2005 period. Cohort 3 is partially exposed during the period, but some of cohort 3’s members reach age 4 by the middle of the period. Only the members of cohort 4 are less than age 4 for the entire period between 2002 and 2005.

Height-for-age of children in cohort 4 reflects the greatest degree of exposure to migration during the ages when children should be the most susceptible to its effects, positive or negative. Thus, estimates for this group have the greatest potential to reveal the effect of migration. Height-for-age of children in cohort 3 reflects some exposure during the susceptible period. I interpret estimates for this group as reflecting a partial effect.[4]

I construct an estimate of migration’s impact on child health by contrasting the height-for-age scores in 2005 of children from migrant households who were less than 3 in 2005 with the height-for-age scores of their same-age counterparts who are also from migrant households but who were measured in 2002, before the migration occurred. Similarly, I construct the estimation of partial impact by contrasting the height-for-age scores in 2005 of children from migrant households who were age 3-5 in 2005 with the height-for-age scores in 2002 of children from migrant households who were age 3-5 in 2002.

The estimates derived from the two comparisons may combine a migration effect and a period effect. If a period effect influences differences between measurements taken in 2002 and 2005, we can obtain an estimate of it by contrasting height-for-age of older children (6-8) in 2005 and older children (6-8) in 2002. Because these children experience parental migration at older ages, their height-for-age values in 2005 should not reflect the effect of migration, and the difference between 2002 and 2005 should reflect only the period effect. I can then subtract this estimate of the period effect from the “migration + time effect” estimates for cohorts 4 and 3, thus purging those estimates of the period effect. This approach also nets out age effects, should they exist. The result is the difference between height for age before and after migration of children measured at young ages and height for age before and after migration for children measured at older ages, i.e. a difference in difference.

I proceed to estimate the same difference in difference for children from non-migrant households by making the same comparisons by age and cohort for children from non-migrant households (however, none of the estimates can be interpreted as reflecting parental migration, because by definition no children experienced a parental migration). With these two difference-in-differences in hand, I can test whether the relationship between younger and older children in migrant households is significantly different from that relationship in non-migrant households, as it should be if parental migration particularly affects the health of children who experience it when they are very young.

Estimating a specification similar in nature to equation 1 on pooled data over time, and comparing groups both across time and between cohorts, is a commonly used method to evaluate the effectiveness of social programs. This is often done at the level of communities or schools (Datar, Mukherji and Sood 2005; Duflo 2001; Frankenberg, Suriastini and Thomas 2005). Yet, by applying this approach to families, I can make use of the same methodological advantage. The selective nature of the “treatment,” in this case, parental migration, can be addressed in part by holding constant the fixed characteristics (measured or unmeasured) of households that may otherwise introduce bias to the results. In this way I am able to isolate the effects of migration.

Within this framework, if some factor other than migration is driving the results, it must be something that changes over time within the household and that affects both parental migration and the health only of the younger children.

I have outlined an approach that attempts to correct for the potential selectivity of migrants relative to non-migrants on characteristics that affect child health. In considering whether such an approach is necessary, it is helpful to compare the estimates that it produces to those that one would obtain using a simpler strategy. Accordingly, before moving to the fixed effects results from the estimation strategy laid out above, I present results obtained by using the sample of children measured in 2002 and regressing their height-for-age values in 2005 on the dichotomous indicator of any parental migration between 2002 and 2005. I include a number of commonly used controls measured in 2002, including child age, sex, and urbanicity at birth; indicators of mother’s and father’s age and education; household per capita expenditures[5]; and an indicator of whether any of the family income comes from agricultural production.

RESULTS

I begin my analysis by using an ordinary least squares estimation to relate children’s parental migration status between 2002 and 2005 to children’s height-for-age in 2005, as described in the last paragraph of the preceding section. I control for a number of potentially confounding characteristics measured in 2002. The fact that the data are longitudinal helps to establish time order; child height is measured after parental migration occurs. The results are shown in Table 2. The coefficient on parental migration is positive and sizeable; it is approximately the same size as the coefficient on having a father with at least some secondary education (7-9 years or 10+ years), relative to a father with completed elementary education or less. However, the coefficient is imprecisely estimated and statistically insignificant. With a larger sample, the result would replicate findings from earlier studies: parental migration is positively associated with subsequent child health outcomes, even in the presence of observable characteristics likely to drive both parental migration and child health.

I then pool observations for children age 0-8 in 2002 and children age 0-11 in 2005 and proceed with the difference-in-difference strategy described above. I estimate an interactive model using indicators of children’s birth cohorts, date of measurement, and whether or not a parent migrated between 2002 and 2005, including family fixed effects. Table 3 presents the coefficients from this approach. Cohort 1 in 2002 is the reference category; cohort 4 is only measured in 2005 and does not have a coefficient estimate in 2002. The top half of the table contains those coefficients for the interaction between the cohort-time period indicators and the migration term; the bottom half of the table contains the coefficients for the first order cohort-time period indicators.

It is worth noting again that the outcome measure, children’s height-for-age, is a standardized measure. The coefficients presented in Table 3 are interpreted as standard deviation differences in child height-for-age. Therefore, coefficients less than 1 are actually still quite substantial in terms of the differences in height-for-age (and underlying nutritional status) that they represent.

Each of the coefficients in the top half of Table 3 are comparisons between the height-for-age of children in migrant households by cohort and year, relative to the oldest cohort measured in 2002, net of the same differences in households in which a parent did not migrate over the period. The coefficients for height-for-age in 2002 are measured before migration occurs and therefore cannot possibly reflect a causal effect of parental migration. Instead, they can be interpreted as cohort-specific differences in growth between children in the households in which a parent subsequently migrates, and households in which a migration does not occur. That these coefficients for the 2002 measures are positive, suggests that the children who have a parent subsequently migrate have different height-for-age values than children whose families remain intact, before the migration occurs. For this reason it is essential to subtract pre-migration differences in height-for-age between children in migrant and non-migrant homes from post-migration comparisons.

I represent these comparisons in Table 4; the equations relate expected values of height-for-age of groups of children by whether or not a parent migrates (m), cohort (C1-C4), and year of measurement (02 or 05).[6] I arrive at the effect of parental migration on child height-for-age by subtracting the period effect, (cohort 2 in 2005 – cohort 1 in 2002), from the comparison of 3-6 year olds at both points in time (cohort 3 in 2005 – cohort 2 in 2002). This provides a difference-in-difference that would be the partial effect of parental migration if younger children in non-migrant households are not systematically different in 2005 from all other children. This estimate is the partial effect because children aged 3-6 in 2005 were only aged less than four for part of the period of exposure between 2002 and 2005 (refer to Figure 1). Similarly, I subtract the period effect from a simple comparison of the youngest children in both waves (cohort 1 in 2005 – cohort 3 in 2002). The result of the ensuing difference-in-difference is the effect of parental migration on children’s height, as long as the same difference-in-difference for children in non-migrant households is zero.

The difference-in-difference estimate reflecting the sibling-period comparison in migrant households (Equation IA) is large and negative, but insignificant, suggesting that in migrant households, 3-6 years olds in 2005 do not have systematically different height-for-age values than 3-6 year olds in 2002, controlling for period change. The estimate comparing 0-2 year olds in migrant households (Equation IIA) is larger, also negative, but also not statistically significant. The larger coefficient on the effect of parental migration on the youngest children (0-2 year olds), relative to the coefficient for slightly older children (3-6), is logically consistent with interpreting the later as a partial effect.

I have arrived at these effects by comparing children within migrant households. However, for these effects to reflect parental migration, I must assume that children in non-migrant households have difference-in-difference estimates which are equal to zero. If all younger children, regardless of migration status, were particularly disadvantaged in 2005, then I am overestimating these effects. If all younger children in 2005 were particularly advantaged, then I am underestimating these effects. For this reason, I estimate the same equations (I-II) described above using the coefficients estimated for non-migrant households. Specifically, I test for the presence of some other effect on children’s height-for-age in non-migrant households that may actually apply to all children.

These comparisons are presented in panel (B) of Table 4. Once a period effect is taken into account, younger children in 2005 have height-for-age values that are statistically larger than the height-for-age values of younger children in 2002 (Equation IIB). In other words, young children in non-migrant households are slightly advantaged relative to their age-counterparts at the beginning of the period.

To net out the difference-in-differences estimated for children in households where a parent did not migrate, I subtract the estimates in Table 4, panel B (households in which a parent did not migrate) from Table 4, panel A (households in which a parent migrated). I interpret these difference-in-difference-in-difference estimates, presented in panel C of Table 4, as the actual effects of parental migration on 3-6 year olds (Equation IC) and 0-2 year olds (Equation IIC). The partial effect (-0.35) is negative but imprecisely estimated, whereas the effect for the youngest children (-1.13) is much larger and significant at the 5% level.

In contrast to the results from the OLS estimation in Table 2, these results suggest that parental migration actually causes disadvantage with respect to health for children. This effect is sizeable; in Mexico, a 1.13 standard deviation difference in height-for-age translates into a 3.9 centimeter difference in height for a 3 year old girl, and a 4.0 centimeter difference in height for a 3 year old boy.

DISCUSSION

This study evaluates the effects of parental migration on child health in sending households. I apply an approach developed for program evaluation, but implement it at the household level, in an attempt to account for the selective nature of migration. Specifically, I compare children within households who are exposed to migration at different critical periods of child development. This method holds constant those household-level characteristics that are fixed in time and additive in nature, including those that cannot be measured. The methodological advantage to this type of approach is that it removes the most commonly cited potential sources of bias to more traditional estimation approaches: parental background, and parental preferences. The results suggest that parental migration actually causes declines in children’s health. Younger children have lower values of height-for-age following parental migration than do their older siblings, who, because of the physiology of child growth, have height-for-age values that were largely determined before the migration occurred and should therefore be unaffected by the household transition. The same pattern does not emerge from comparisons of younger and older siblings in non-migrant households.

The direction of this effect differs from that obtained using ordinary least squares regression with the same data. Results from an OLS approach suggests that parental migration has no significant association with child height-for-age, though a sizable positive, albeit imprecisely estimated, coefficient is produced. An estimation which compares children across time and within households changes the direction of this relationship in a significant manner. For this reason, parental, household, or community characteristics appear to be important predictors of both parental migration and child health.

My interpretation of these findings as the causal effect of migration does remain open to a few plausible alternative explanations. First, if the household experienced a major economic setback between 2002 and 2005, it may have created incentive for the household to have one parent migrate, while at the same time constraining nutrition and health care options for the children in the household. If older children’s height-for-age is largely determined before this period, their stature would not reflect the shock, while children in more vulnerable younger years would have height measures which do reflect the negative effects of the shock.[7]

Secondly, the timing of my dates of height measurement coincide with the roll out of a health and education program aimed at reducing poverty in rural Mexico, called PROGRESA, or later, Oportunidades. The program began in 1997 by targeting poor households in rural areas of seven Mexican states (Stecklov et al. 2005). Subsequently, it has been expanded to include poor households in a number of other areas in Mexico. The program has several components; the most relevant for this study is the receipt of nutritional supplements for children under 5, and monetary stipends to the household if age-appropriate children are immunized and attend growth-monitoring and if parents attend health and nutrition-related training (Gertler 2004; Schroeder et al. 1995). If the household became eligible for this program (because of its expansion to new communities) between 2002 and 2005, the receipt of stipends and nutrition supplements for young children would likely have positively affected children’s stature, but may also have reduced the likelihood of parental migration[8], either via the stipulations of stipend receipt or via the improvement in resources for the household.

Fortunately, MxFLS data include information on whether, and when, the household experienced a number of economic setbacks in the 5 years preceding the survey as well as information on household receipt of PROGRESA/Oportunidades at the time of the survey. If either a household economic shock or household uptake of Oportunidades during the period were to bias my results, I would expect to see a positive relationship between either occurrence and whether or not a parent migrated between 2002 and 2005. To investigate this, I use the MxFLS data and regress the migration indicator on whether the household enrolled in Oportunidades over the period (not shown). I then regress the migration indicator separately on series of indicators of economic shock, such as natural disaster, crop failure, or victimization between 2002 and 2005 (not shown). I actually find no evidence that migration is related to either enrollment in Oportunidades or the occurrence of an economic shock, increasing my confidence that my results are not merely capturing either event.

The clear caveat to my finding is that there is a nontrivial amount of missing data. I compare information on children measured in 2002 who are not included in the analytical sample because of missing data on height or parental migration in either wave. Mean values of individual and family-level characteristics are presented in the Appendix table. Most notably, the sample appears to be selected on urbanicity; those included are less likely to be born in an urban area than the children excluded from the analysis. As we would expect, the sample is subsequently also negatively selected on parental education. Importantly, though, the sample does not appear to be selected on parental health endowment; the children who are omitted from the analysis have parental height values that are very similar to children included in the analysis. One way to address the apparent selection on urbanicity and parental education may be to analyze rural and urban children separately.

If in fact the negative effect of migration on the health of very young children holds, it reveals several important insights. There are a number of reasons, including the size of remittance flows, that we may have expected the effect of parental migration on child health to be positive. Yet, the results suggest that the migration process may introduce detriments to child health and nutrition, at least in the short run. These may occur because parental absence from the household makes the provision of child nutrition or access to health care more difficult, given the initial shift in time constraints of the remaining parent or caretaker. In such case, the results are consistent with a large body of family literature that relates household transitions to child disadvantage. Secondly, the finding presents a difficult picture for the outcome of migration behavior for the country as a whole. As mentioned, parental migration in Mexico affects a large number of children. The findings from this study suggest that a considerable proportion of Mexican children face health setbacks because of it.

My finding is consistent with the conclusions of Kanaiaupuni and Donato (1999), who find that migration initially confers a health disadvantage to children in sending regions. Alternatively, the finding stands in contrast to the positive effect of migration found by Hildebrandt and McKenzie (2005). However, the authors focus on international migration, whereas my study pools both internal and international migrants.

This contrast highlights two key limitations of this study. A difference-in-difference approach controls for a number of sources of unobserved potential bias to the estimates. However, it carves the sample into relatively small pieces. Because the percentage of the sample with migrating parents between 2002 and 2005 is small (~5%), I do not further divide this group by other characteristics that may be of interest, such as whether it is the child’s mother or father who moves, the distance of the move from the child’s community, or the duration of the move. Secondly, I can only observe children’s outcomes in the short time period following the migration spell. Given the Kanaiaupuni and Donato (1999) finding, it is possible that the effect shortly after a parent moves away is negative and once the family adjusts and remittances return, the net effect may be positive. It will be possible to measure this process when the MxFLS survey, including wave 3 in 2008, is complete. However, given that stature is largely influenced in children’s early years, a short-term negative effect is still important from a policy perspective. The effect of poor health during early childhood can be mitigated somewhat by later life health improvements, but has measurable long-term health implications regardless of later life behavior (Pinstrup-Andersen, Pelletier and Alderman 1995).

My study also points to a number of questions about migration and family transitions more broadly. I focus here on families in which parents marry, have children, and then migrate. Though, in a number of families, migration likely affects the marital process, as well as fertility behavior. While I do not exclude children of couples that migrate before marriage, I do not directly estimate the impact of these earlier life behaviors on child well-being. The larger question of how migration causally affects child outcomes may well conceptualize a process that spans a larger period of adolescent and adult life. The difficulty in answering these broader questions lies in developing a causal estimation strategy. Instead, this study isolates a period of family life to assess whether an effect of parental migration can be identified.

REFERENCES

Amato, Paul R., and Joan G. Gilbreth. 1999. "Nonresident Fathers and Children's Well-Being: A Meta-Analysis." Journal of Marriage and the Family 61:557-573.

Boucher, Stephen R., Oded Stark, and J. Edward Taylor. 2005. "A Gain with a Drain? Evidence from Rural Mexico on the New Economics of the Brain Drain." in ARE Working Papers: University of California, Davis.

Datar, Ashlesha, Arnab Mukherji, and Neeraj Sood. 2005. "Health Infrastructure and Immunization Coverage in Rural India." in Working Paper #294: RAND.

De Vos, Susan M. 1995. Household Composition in Latin America. New York: Plenum Press.

Demuth, Stephen, and Susan L. Brown. 2004. "Family Structure, Family Processes, and Adolescent Delinquency: The Significance of Parental Absence Versus Parental Gender." Journal of Research in Crime and Delinquency 41:58-81.

Duflo, Esther. 2001. "Schooling and Labor Market Consequences of School Construction in Indonesia." American Economic Review 91:795-813.

Duncan, Greg J., and Jeanne Brooks-Gunn (Eds.). 1997. The Consequences of Growing Up Poor. New York: Russell Sage Press.

Durand, Jorge, William Kandel, Emilio A. Parrado, and Douglas S. Massey. 1996. "International Migration and Development in Mexican Communities." Demography 33:249-264.

Fernandez, Leticia. 1998. "Do Fathers Influence Their Children's Health by Migrating? Evidence from Rural Mexico." Unpublished Manuscript.

Frank, Reanne, and Robert A. Hummer. 2002. "The Other Side of the Paradox: The Risk of Low Birth Weight Among Infants of Migrant and Nonmigrant Households within Mexico." International Migration Review 36:746-765.

Frankenberg, Elizabeth, Wayan Suriastini, and Duncan Thomas. 2005. "Can expanding access to basic healthcare improve children's health status? Lessons from Indonesia's 'midwife in the village' programme." Population Studies 59:5-19.

Garcia, Brigida, and Orlandina de Oliveira. 2005. "Fatherhood in Urban Mexico." Journal of Comparative Family Studies 36:306-327.

Gertler, Paul. 2004. "Do Conditional Cash Transfers Improve Child Health? Evidence from PROGRESA's Control Randomized Experiment." American Economic Review 94:336-341.

Hildebrandt, Nicole, and David J. McKenzie. 2005. "The Effects of Migration on Child Health in Mexico." in Policy Research Working Paper #3573: World Bank.

Kanaiaupuni, Shawn Malia. 2000. "Sustaining Families and Communities: Nonmigrant Women and Mexico-U.S. Migration." in CDE Working Paper.

Kanaiaupuni, Shawn Malia, and Katharine M. Donato. 1999. "Migradollars and Mortality: The Effects of Migration on Infant Survival in Mexico." Demography 36:339-53.

Kanaiaupuni, Shawn Malia, Katharine M. Donato, Theresa Thompson-Colon, and Melissa Stainback. 2005. "Counting on Kin: Social Networks, Social Support, and Child Health Status." Social Forces 83:1137-1164.

Martorell, Reynaldo. 1995. "Promoting Healthy Growth: Rationale and Benefits." Pp. 15-31 in Child Growth and Nutrition in Developing Countries, edited by Per Pinstrup Anderson, David Pelletier, and Harold Alderman. Ithaca: Cornell University Press.

Massey, Douglas S., and Emilio A. Parrado. 1994. "Migradollars: The Remittances and Savings of Mexican Migrants to the United States." Population Research and Policy Review 13:3-30.

McLanahan, Sara, and Gary Sandefur. 1994. Growing Up with a Single Parent. Cambridge, Massachusetts: Harvard University Press.

McLoyd. 1990. "The Impact of Economic Hardship on Black Families and Children: Psychological Distress, Parenting, and Socioeconomic Development." Child Development 61:311-346.

National Center for Health Statistics. 2000. "2000 CDC Growth Charts: United States."

Nobles, Jenna. 2008. "The Contribution of Migration to Children's Family Context." Los Angeles: CCPR Working paper.

Palloni, Alberto, and Jeffrey D. Morenoff. 2001. "Interpreting the Paradoxical in the ‘Hispanic Paradox’: Demographic and Epidemiological Approaches." in Population Health and Aging, edited by Maxine Weinstein, Albert I. Hermalin, and Michael A. Stoto. New York: New York Academy of Sciences.

Pinstrup-Andersen, Per, David Pelletier, and Harold Alderman (Eds.). 1995. Child Growth and Nutrition in Developing Countries. Ithaca: Cornell University Press.

Richter, Kerry. 1988. "Union Patterns and Children's Living Arrangements in Latin America." Demography 25:553-566.

Rivera, Juan A., Daniela Sotres-Alvarez, Jean-Pierre Habicht, Teresa Shamah, and Salvador Villalpando. 2004. "Impact of the Mexican Program for Education, Health, and Nutrition (Progresa) on Rates of Growth and Amenia in Infants and Young Children." Journal of the American Medical Association 291:2563-2570.

Rosenzweig, Mark R., and Oded Stark. 1989. "Consumption Smoothing, Migration, and Marriage: Evidence from Rural India." Journal of Political Economy 97:905-926.

Rubalcava, Luis, and Graciela Teruel. 2004. "The Mexican Family Life Survey Project (MxFLS): Study Design and Baseline Results." in CIDE & UIA Working Paper.

Salgado de Snyder, V. Nelly. 1993. "Family Life Across the Border: Mexican Wives Left Behind." Hispanic Journal of Behavioral Sciences 15:391-401.

Schroeder, Dirk G., Reynaldo Martorell, Juan A. Rivera, Marie T. Ruel, and Jean-Pierre Habicht. 1995. "Age Differences in the Impact of Nutritional Supplement on Growth." Journal of Nutrition 125:1051S-1059S.

Seltzer, Judith A. 1994. "Consequences of Marital Dissolution for Children." Annual Review of Sociology 20:235-266.

Stark, Oded, and Robert E. B. Lucas. 1988. "Migration, Remittances, and the Family." Economic Development and Cultural Change 36:465-481.

Stecklov, Guy, Paul Winters, Marco Stampini, and Benjamin Davis. 2005. "Do Conditional Cash Transfers Influence Migration? A Study Using Experimental Data From the Mexican PROGRESA Program." Demography 42:769-790.

Strohschein, Lisa. 2005. "Parental Divorce and Child Mental Health Trajectories." Journal of Marriage and Family 67:1286-1300.

Teachman, Jay, Randal Day, Kathleen Paasch, Karen Carver, and Vaughn Call. 1998. "Sibling Resemblence in Behavioral and Cognitive Outcomes: The Role of Father Presence." Journal of Marriage and the Family 60:835-848.

Van Hook, Jennifer, and Jennifer Glick. 2005. "Mexican Migration to the United States and Extended Family Living Arrangements." in Unpublished manuscript.

Zahniser, Steven S. 2000. "One Border, Two Crossings: Mexican Migration to the United States as a Two-Way Process." Pp. 242-276 in Immigration Research for a New Century, edited by Nancy Foner, Ruben G. Rumbaut, and Steven J. Gold. New York: Russell Sage Foundation.

Figure 1. Lexis diagram of children’s age, year, and points of measurement

[pic]

Table 1. Descriptive Characteristics by Family Migration Status

| |Of children with two parents in the household in 2002, |

| |neither having migrated since the child’s birth |

|  |  |  |

|Characteristics in 2002 |Parent did not move away |Parent moved away at least once |

| |between 2002-2005 |between 2002-2005 |

| | | |

| |Mean |

| | | |

|Child height-for-age z-score |-0.65 |-0.45 |

|Child age in years |4.2 |3.9 |

|Child born in an urban area |30% |35% |

| | | |

|Father's age in years |35.2 |32.8 |

|Father's education in years |7.3 |8.7* |

|Mother's age in years |31.7 |30.0 |

|Mother's education in years |6.7 |7.9* |

| | | |

|Household ln(pc expenditures) |6.1 |6.1 |

|Household uses land for income |19% |14% |

|N |2,519 |112 |

Note: Sample: children 0-8 in 2002. Source: Mexican Family Life Survey

*Mean value (or percentage) is significantly different between groups at p < 0.05

Table 2. Results from OLS Regression of Child Height-for-Age in 2005

on Parental Migration Status between 2002 and 2005, controlling

for child, parental, and household characteristics.

| |Child Height-for-Age |

| |(measured in 2005) |

|Parent migrated between 2002 and 2005 |0.21 |

| |(0.14) |

|Controls (measured in 2002) | |

|Child's age in years |-0.05 |

| |(0.01)** |

|Child is male |-0.15 |

| |(0.05)** |

|Child born in urban area |0.13 |

| |(0.07) |

|Father's age: 15-29 years |-- |

| | |

| 30-39 years |0.08 |

| |(0.08) |

| 40+ years |0.14 |

| |(0.11) |

|Father's education: 0-6 years |-- |

| | |

| 7-9 years |0.19 |

| |(0.07)** |

| 10+ years |0.22 |

| |(0.09)* |

|Mother's age: 15-25 years |-- |

| | |

| 26-35 years |0.10 |

| |(0.09) |

| 36+ years |0.19 |

| |(0.12) |

|Mother's education: 0-6 years |-- |

| | |

| 7-9 years |0.30 |

| |(0.08)** |

| 10+ years |0.37 |

| |(0.10)** |

|Ln (HH per capital expenditures) |0.18 |

| |(0.03)** |

|Household uses land for income |-0.33 |

| |(0.07)** |

|Constant |-1.46 |

| |(0.23)** |

|N |2,914 |

Note: Sample: Children age 0-8 in 2002. Standard errors (in parentheses) adjusted for clustering at the household level. The omitted category of father’s age is 15-29 years; mother’s age is 15-25 years; father’s and mother’s education is 0-6 years. An F-test concludes that father’s education and mother’s education is jointly significant at the 5% level, but father’s age and mother’s age are not. Source: Mexican Family Life Survey

* p ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download