key: cord-0028676-g4duounx
authors: Zhang, Yue; Zheng, Xiaodong
title: Internal migration and child health: An investigation of health disparities between migrant children and left-behind children in China
date: 2022-03-16
journal: PLoS One
DOI: 10.1371/journal.pone.0265407
sha: 9bbadb4226791b9b56e1573dfe419bd5df1cc7ae
doc_id: 28676
cord_uid: g4duounx

Using data from the China Education Panel Survey (CEPS), this study empirically examines the association between internal migration and child health through an investigation of health disparities between migrant children and left-behind children in China. The results show that, in comparison with being left behind, migrating with parents significantly improves children’s self-reported health, height-for-age z-score (HAZ) and BMI-for-age z-score (BAZ), and reduces their frequency of sickness. These findings remain robust to a suite of robustness checks. Furthermore, the health effects of internal migration are more prominent for children with a rural hukou compared with urban ones. Although migrant children are more likely to experience teacher discrimination, they have higher levels of parental care, family relationships, and peer relationships relative to their left-behind counterparts, which indicates possible mechanisms behind the association between children’s migration and health. Our findings underline the importance of policy improvement and evidence-based interventions aiming at reducing involuntary parent-child separation and facilitating the development in health of disadvantaged children in developing countries like China.

Due to the growing scope and impacts of population movements, migration is becoming one of the determining global phenomena of the 21st century [1] . As the largest developing country in the world, the massive internal migration has greatly facilitated China's urbanization and economic development over the past decades. However, the household registration (hukou) system, which is often associated with the receipt of local programs and benefits such as education, health care and social security, has long been the main institutional barrier for Chinese internal migrants to bring their family members (such as their children) to the destination of migration [2] . As a result, a large number of children have either been migrated with their parents or left behind at home while parents moving away for work [3, 4] . According to the statistics released by the Ministry of Education of China, there were 14.27 million migrant children and 13.84 million rural left-behind children in China by 2019 [5] . Migration investigated the health differences between these two groups of children. Theoretically, although migrant children may confront challenges of acculturation of a new living environment and social exclusion, they often have higher levels of parental care and family relationships and are less likely to experience health and behavioral problems triggered by parent-child separation [13, 29] . With parental care and protection, migrating with parents also provides an opportunity for children to broaden their horizons, make new friends, and maintain healthy peer relationships, which are crucial for children's social, behavioral, and health development [20] . Furthermore, it has been reported that official policies highlighted the importance of social integration of migrant households and the equality of developmental opportunity among migrant children in recent years [17] . Consequently, social exclusion and discrimination are not likely to play a major role in health disparities between migrant children and leftbehind children. Hence, we assumed that migrating with parents was linked to better health of children compared with being left behind. In addition, in comparison with urban-to-urban migration, the distance and duration of rural-to-urban migration are often longer due to more inter-provincial movements [16] . Urban children are also more likely to be exposed to resource-rich neighborhoods and schools, social security, and protection (e.g., medical insurance) than rural children [30] . As such, we assumed that the health differences between migrating with parents and being left behind would be more pronounced among rural children.

Using data from the China Educational Panel Survey (CEPS), the purpose of this study was to examine the health consequences of children's migration in China. Different from most past research, this study analyzed the health differences between migrant children and left-behind children, which are the two types of children of internal migrants. Furthermore, we investigated the urban-rural heterogeneity and possible mechanisms of the health disparities between migrant children and left-behind children. Specifically, the potential pathways that were examined in this study included parental care, family relationships, peer relationships, and teacher discrimination. In general, our study contributes to the literature on the health consequences of internal migration and corresponding mechanisms from the perspective of heterogeneity between migrant children and left-behind children. The results can help identify the disadvantaged children more clearly during the process of internal migration in China. Our findings are also beneficial to inform future policies to optimize the allocation of public resources for the development of children of migrants in China and other contexts with a large-scale of migrants. According to our theoretical analysis, we proposed three hypotheses as follows:

Hypothesis 1 (H1): Compared with being left behind, migrating with parents is helpful to improve children's health.

The health disparities between migrant children and left-behind children are more prominent among rural children than urban ones.

Parental care, family relationship, and peer relationship play a major role in the mechanisms underlying the health differences between migrant children and leftbehind children.

Data used in this study were drawn from the China Education Panel Survey (CEPS). The CEPS is a nationally representative longitudinal survey for junior high school students, which applies a stratified and multistage sampling design with probability proportional to size (PPS).

In the academic year 2014~2015, a total of 9,449 8th grade students in 112 junior high schools were interviewed. We restricted our sample to children of the internal migrants and the final sample size was 2,780. Among them, students (living with their parents) whose hukou registration place was not their residential location were regarded as migrant children (N = 1,348, 48.5%). Left-behind children (N = 1,432, 51.5%) were identified as students who lived in the hukou registration place while their parents moved away to other counties or provinces. Based on the available information regarding health in the CEPS data, this study utilized four indicators to measure child health. First, we used self-reported health (very unhealthy = 1 to very healthy = 5) as an overall health measure. Second, the frequency of sickness, such as cold, fever, cough, diarrhea, in the past year (never = 1, seldom = 2, often = 3) was applied as a proxy indicator of physical health. Third, to assess nutritional status, we used two anthropometric measures including height-for-age z-score (HAZ) and BMI-for-age z-score (BAZ). HAZ and BAZ indicators are widely accepted objective measures of child health [31, 32] . They are often used to determine whether a child is growing "normally" or has a growth problem according to the height and Body Mass Index (BMI) of children and corresponding child growth standards. It has been proved that these child health indicators are important predictors of the short-and long-term health of children [33] . HAZ and BAZ are comparable across different genders and ages. The derivation of the two anthropometric measures is:

where y i is the height (cm) or BMI (kg/m 2 ) of child i in group j. Group j is defined by gender and age (in months). � y _ j and σ j are the mean and standard deviation of the height (cm) or BMI of the reference group j of the same gender and age according to World Health Organization (WHO) child growth standards [34] . Higher values of HAZ and BAZ often indicate a better nutritional status of children.

In addition, we controlled for characteristics of children and their families to reduce potential confounding effects in our empirical analysis. First, given the different growth statuses among different sexes and ages of children, child gender (male = 1) and age (in years) variables were included as controls in our study. Second, existing evidence has shown that children with an urban hukou had significantly better health than children with a rural hukou because China's hukou system determines eligibility for various welfare benefits [35] . Thus, children's hukou status (rural = 1) was also treated as a control variable in our regression analysis. Third, it is reported that the number of siblings is also an important determinant for children's physical development [36] . According to the quantity-quality trade-off theory, only children often have more nutrition intake than children with siblings because of more parental investments and care [37] . As such, the only child (yes = 1) status of children was also controlled when investigating the health disparities between migrant children and left-behind children. Fourth, we also controlled for children's school boarding (yes = 1) status since past research suggested that boarding was often negatively associated with children's nutrition and health [38, 39] . Finally, it is also well-documented that high family socioeconomic status is critical for a good parenting style and healthy living environment, which can subsequently impact child health [40, 41] . Therefore, as two crucial measures of household socioeconomic status (SES), family economic status (poor = 1 to very rich = 4), and parental education (parental average educational years) were also treated as controls in our empirical analysis.

The baseline regressions for our empirical analysis are specified as follows:

where h ij is the ith health outcome for jth child, Migrant j is a dummy variable indicating the migrant status of the jth child, X j represents a vector of control variables, and ε ij denotes the error term. In baseline regressions, we applied ordered probit model for ordinal outcome variables and ordinary least square (OLS) estimation for continuous dependent variables. To account for potential self-selection bias, we further employed propensity score matching (PSM) and inverse probability weighting with regression adjustment (IPWRA). Both the PSM and IPWRA approaches firstly estimate the selection to treatment (Migrant = 1) and predict the probabilities of being treated (propensity scores) for all observations. We used a binary probit model to estimate the treatment model as follows:

Then, the PSM approach matches the treated and control groups and compares the differences in outcomes based on propensity scores directly, while the IPWRA method assigns the inverse of probability for the treated and control groups, and re-estimates Eq (1) using new weights [42] . To address potential omitted variable bias, we conducted two sensitivity tests. First, we performed Oster's bounds analysis to test the sensitivity of OLS estimates to the unobservables based on the selection of observables [43] . The basic idea of Oster's bounding lies in the proportional selection assumption. The proportional selection assumption indicates that determinants of child health consist of both observed and unobserved factors, and none of them could dominate children's health status. Therefore, if the observables are randomly chosen from all determinants of child health, then one can calculate reasonable bounds to examine the coefficient stability on the basis of informativeness of the selected observables, which are proportional to the unobservables [44] . Second, we calculated Rosenbaum bounds for the PSM estimates in the presence of unobserved heterogeneity [45] . The key concept of Rosenbaum bounding analysis is to see how much the estimated treatment effects may vary if the estimated odds of migrating with parents (vs. being left-behind) are manipulated by different levels of hidden bias. Additionally, to alleviate the concern of possible reverse causality problem, we used children's migrant status in 2013~2014 (lagged key independent variable) to identify migrant and left-behind children and re-estimate the baseline regressions. Table 1 presents the summary characteristics of our sample. Overall, the average age of sample children was 14.63 years old. Among them, 48.5% of the respondents were migrant children, 54% were males, 61.9% had a rural hukou, 30% were in a boarding school, and 35% were the only child in the household. compared with left-behind children, migrant children had higher self-reported health, HAZ and BAZ, and a lower frequency of sickness, which preliminarily indicates that migrant children were healthier than left-behind children. In addition, there were also significant differences in some individual and family characteristics between the two groups of children (e.g., boarding status, only child status, family economic status, and parental education), implying that children's migrant status was not random and the self-selection issue should be considered in the empirical analysis. Table 2 presents the estimates of baseline regressions (Panel A) and a series of robustness checks (Panel B to Panel D). In Panel A, the ordered probit and OLS estimates showed that migrating with parents significantly and positively predicted children's self-reported health (β = 0.198, p< 0.01), while it was negatively associated with sick frequency (β = -0.100, p< 0.05).

Holding the control variables fixed, migrant children also significantly outperformed their left-behind counterparts in HAZ and BAZ by 0.155 (p< 0.01) and 0.186 (p< 0.01) standard deviations, respectively. In Panel B, the estimates of the three PSM approaches, including nearest neighbor matching (k = 1, one-to-one matching with no replacement), radius matching (r = 0.05), and kernel matching, consistently indicated significantly positive associations of internal migration with children's self-reported health, HAZ and BAZ, as well as a negative relationship between children's migration and their sick frequency. Fig 1 and Table 3 demonstrate the balance of the characteristics between migrant and left-behind children before and after PSM estimation. The results revealed that some characteristics were significantly different between the two groups of children before matching, such as boarding status and family economic status. Nonetheless, all of the covariates between migrant and left-behind children were no longer significant and the standardized biases across covariates were close to zero after matching, indicating that the observed self-selection bias has been largely corrected. In Panel C, the IPWRA estimates also coincidentally suggested that children's migration positively predicted their health, including higher levels of self-reported health, HAZ and BAZ, and lower frequency of sickness. In addition, we further used children's migrant status in academic year 2013~2014 as the key independent variable and re-estimated the baseline regressions. As shown in Panel D, the results were also robust when using the migrant status in 2013~2014 (lagged independent variable) to measure children's migration, indicating that reverse causality should not be a serious issue in our study.

Tables 4 and 5 demonstrate the results of sensitivity analysis based on Oster (2019)'s bounds for linear regressions and Rosenbaum's bounds for the PSM approach, respectively. Table 4 reports the bounding estimates of Oster's method. Following Botha et al. (2021) [46] , we assumed a R max = 1.3(R 2 ), where R max represents the goodness of fit when controlling for all observed and unobserved factors associated with child health and R 2 is from the linear regressions with all controls. The results showed that the lower bounds (absolute value) of the health disparities between migrant and left-behind children were above zero and only if the selection on unobservables is 3.581 times higher than that on the observables would it be possible to eliminate the observed effects. Such bounding results suggested that we could not reject that the association of children's migration and health indicate at least a partial causal relationship. The Rosenbaum bounds in Table 5 indicated that the health effects of children's migration remained significant until a hidden bias (gamma) or unobserved heterogeneity increased to 1.1 to 1.4 times. Overall, the sensitivity analysis reveals that our findings are robust against unobserved heterogeneity to a certain extent.

Given the differences in socioeconomic development between urban and rural areas, we further conducted a disaggregated analysis by hukou type. As shown in Fig 2, for all the four health outcomes, the health disparities between migrant children and left-behind children with a rural hukou are larger than their urban peers. Specifically, in comparison with their One major possible reason is that rural left-behind children are often more disadvantaged than urban ones in terms of family economic status, parental care, and social support [27, 30] . As such, in comparison with being left behind, rural children benefit more in health when migrating with parents to cities.

To provide insight into our main findings, we explored possible channels behind the health effects of children's migration, including parental care, family relationships, peer relationships, and teacher discrimination (see details of the indicators in Table 6 ). In terms of parental care, the IPWRA estimates suggested that children's migration positively predicted parental life care (β = 0.199, p< 0.01), study care (β = 0.174, p< 0.01) and mind care (β = 0.116, p< 0.01). For family relationships, the results demonstrated significant and positive associations of internal Notes: Robust standard errors in parentheses are clustered at the school level. The lower bound and upper bound of bounds analyses are shown in brackets. The bounds analysis assumes R max = 1.3(R 2 ), where R 2 is from the OLS regressions with all controls. The lower bound β 1 is estimated on the basis that the proportional amount of selection on unobservables to selection on observables is 0 (i.e., δ = 0). The lower bound β 0 is calculated when δ = 1, that is, the proportional degree of selection on unobservables is equal to selection on observables (δ = 1). The estimated δ required for β = 0 indicates that relative to selection on observables, there should be δ times the amount of selection on unobservables to become insignificant. � p < 0.1, �� p < 0.05, migration with mother-child relationship (β = 0.099, p< 0.01), father-child relationship (β = 0.111, p< 0.01) and relationship between parents (β = 0.140, p< 0.01). The results in Table 7 also showed that migrant children suffered from more significant teacher discrimination (β = 0.040, p< 0.05) compared with their left-behind counterparts. Nevertheless, they had significantly better classmate relationship (β = 0.094, p< 0.01) and schoolmate relationship (β = 0.099, p< 0.01). To sum up, although migrant children were more likely to face the challenge of teacher discrimination in comparison with their left-behind peers, the higher levels of parental care, family relationships, and peer relationships indicate candidate pathways underlying the positive association between internal migration and health among migrant and leftbehind children.

Although a larger number of previous studies have examined the association between parental migration and child health, empirical evidence regarding the link between children's internal migration and health remains scarce. Using data from the China Education Panel Survey (CEPS) in the academic year 2014~2015, this study examined the disparities in self-reported health, sick frequency, HAZ, and BAZ between migrant children and left-behind children in China. In our empirical analysis, we employed the ordered probit model and OLS estimation as our benchmark regression approaches. Taking possible self-selection bias into account, we used inverse probability weighting with regression adjustment (IPWRA) and propensity score matching (PSM) methods as robustness checks for our main findings. To address the concern of unobserved confounding factors, we applied Oster's bounds analysis for linear regressions and Rosenbaum bounds analysis for PSM estimation. Besides, we used children's migrant No one helped = 1; One or two days = 2; Three or four days = 3; Almost every day = 4.

How often do your parents discuss things happened at school? Never = 1; Sometimes = 2; Often = 3.

How is the general relationship between you and your mother?

Not close = 1; Not too close nor too far = 2; Very close = 3.

How is the general relationship between you and your father? 

Do you think the school teachers are prejudiced against students from non-local county/district? Not prejudiced at all = 1; A little bit prejudiced = 2; Somewhat prejudiced = 3; Very prejudiced = 4.

Notes: All items were reported by students except for teacher discrimination, which was reported by students' parents due to the lack of information about the corresponding student-reported question.

https://doi.org/10.1371/journal.pone.0265407.t006 status in the previous survey (lagged independent variable) as the key independent variable to address the concern of possible reverse causality problem. Our study also investigated the urban-rural heterogeneity of the health effects of internal migration by children's hukou type.

In addition, we further explored the possible channels underlying the associations of internal migration with health among migrant and left-behind children.

The empirical results suggest that migrating with parents significantly improves children's health status (higher self-reported health, HAZ, BAZ, and lower frequency of sickness) in comparison with being left behind. These results were robust after accounting for potential selfselection bias, unobserved heterogeneity, and reverse causality issues. These findings were also in line with the first hypothesis of this study. Further, the health consequences of children's migration were shown to be more salient for rural children than urban ones, which was consistent with our second hypothesis and some past research regarding the urban-rural socioeconomic gap [47, 48] . In terms of mechanism analysis, despite experiencing increased teacher discrimination, migrant children had higher levels of parental care, family relationships, and peer relationships relative to left-behind children. The results revealed that the third hypothesis proposed by this study cannot be rejected, which suggested that improved parental care, family relationships, and peer relationships could be major reasons that internal migration was positively associated with child health.

Previous studies indicated that children of Chinese internal migrants were more disadvantaged in education and health in comparison with native children living with their parents [14] . The findings of our study further suggested that left-behind children were more prone to nutrition and health risks than migrant children. In this regard, our study highlights the adverse health consequences of parental absence in China. Policymakers should take actions to reduce the occurrence of involuntary parent-child separation due to institutional constraints. Evidence-based policy interventions such as equalization of health resources and services should also be implemented to foster the development of disadvantaged children.

This study has several limitations. First, due to the data constraints, the observations of our study were junior high school students rather than children of all ages, which suggested that the findings of this study should be interpreted and generalized with caution. Second, although we used four health indicators to measure children's health outcomes, they may not be able to cover all health dimensions, especially for psychological health measures such as emotional stability and depression. Third, the potential mediators we proposed and empirically examined may not fully explain the associations of internal migration with health among migrant and left-behind children. More relevant studies are needed to contribute to the comprehensive evaluation of the impacts of migration on children's well-being in China as well as other developing countries.

Conceptualization: Xiaodong Zheng. 

Hukou system, mechanisms, and health stratification across the life course in rural and urban China. Health and Place

Does wealth matter? An assessment of China's rural-urban migration on the education of left-behind children

Differentiated childhoods: Impacts of rural labor migration on left-behind children in China

Overview of China's education: the development of national education in 2019

The well-being of China's rural to urban migrant children: Dual impact of discriminatory abuse and poverty

Migration and child health: Understanding the coverage of child immunization among migrants across different socio-economic groups in India

Well-being of migrant and left-behind children in China: Education, health, parenting, and personal values

The impact of rural-urban migration on the health of the left-behind parents

Labour Migration and Health of Left-Behind Children in

The health of left-behind children in rural China

The status of care and nutrition of 774 left-behind children in rural areas in China

China's left-behind children: Impact of parental migration on health, nutrition, and educational outcomes

The psychological and behavioral outcomes of migrant and left-behind children in China

Access to public schools and the education of migrant children in China

Quality of migrant schools in China: Evidence from a longitudinal study in Shanghai

Choices or constraints, education of migrant children in China

Identity crisis among rural-to-urban migrant children in China: a proposal for school and government interventions

The impact of perceived discrimination on mental health among Chinese migrant and left-behind children: A meta-analysis

The Causal Effects of Rural-to-Urban Migration on Children's Well-being in China

The mental health of children of migrant workers in Beijing: The protective role of public school attendance

Economic Stress, Social Relationships and School Satisfaction Among Migrant Children in Urban China

The impact of migration on family left behind. In International Handbook on the Economics of Migration

Effects of Parental Migration on Mental Health of Left-behind Children: Evidence from Northwestern China

Where did all the remittances go? Understanding the impact of remittances on consumption patterns in rural China

Labor migration and time use patterns of the left-behind children and elderly in rural China

Victimization of Children Left Behind in Rural China

Solution or isolation: Is boarding school a good solution for left-behind children in rural China?

Childhood Left-Behind Experience and Employment Quality of New-Generation Migrants in China

Health status of children left behind in rural areas of Sichuan Province of China: A cross-sectional study

BMI-for-age Z-score distribution shifts among Chinese children: Gender disparity

Family income and child health in China

Social Pensions and Child Health in Rural China

Development of a WHO growth reference for school-aged children and adolescents

Urban-rural disparities in child nutrition-related health outcomes in China: The role of hukou policy

Influence of siblings on child health behaviors and obesity: a systematic review

Do siblings take your food away? Using China's One-Child Policy to test for child quantity-quality trade-offs

Malnutrition in China's rural boarding schools: The case of primary schools in Shaanxi Province

Impacts of boarding on primary school students' mental health outcomes-Instrumental-Variable evidence from rural northwestern China

Socioeconomic status and child health: why is the relationship stronger for older children

How important is parental education for child nutrition

Inverse probability weighted estimation for general missing data problems

Unobservable Selection and Coefficient Stability: Theory and Evidence

Early-life exposure to parental mental distress and adulthood depression among middle-aged and elderly Chinese

Overt bias in observational studies

Implications of COVID-19 labour market shocks for inequality in financial wellbeing

Victimization of Children Left Behind in Rural China

Rural-urban migration and mental health of Chinese migrant children: Systematic review and meta-analysis

��� p < 0.01 https://doi.org/10.1371/journal.pone.0265407.t004