This study draws on the Vietnam Multiple Indicator Cluster Survey (MIC) 2010-2011, which was conducted by the General Statistics Office of Vietnam (GSO) with financial and technical support from the United Nations Children’s Fund (UNICEF) and the United Nations Population Fund. Interviews for the MIC were carried out in November and December 2010, and in January 2011. The survey includes 3,678 observations for a module of children under 5 years of age.² The Red River Delta, Northern Midlands and Mountainous Areas, North Central Area and Central Coastal Area, Central Highlands, South East, and Mekong River Delta were the six regions of Vietnam that the sample design for the Vietnam MICS 2011 survey was focused on, producing statistically reliable estimates for. The sampling strata in each of the six regions were designated as urban and rural areas. A multi-stage, stratified cluster sampling method was employed to choose the survey sample. 2,050 households were produced as a result of this exercise, which is the required sample size for each region and results in a total of about 12,000 households. Based on a number of factors, including the design effect, the available budget, and the amount of time required per team to complete one cluster, the average number of households chosen for each cluster for the Vietnam MICS 2011 was determined to be 20 households. By dividing the total number of households by the number of sample households in each cluster, it was decided that 100 sample clusters would need to be chosen in each region. The data contains rich information about age, gender, ethnicity, and other socio-economic characteristics of children and their families.

We also use the 2009 Population and Housing Census. The data are taken from a 15% nationally representative sample of the population. This survey was conducted by the General Statistics Offices of Vietnam and provides information concerning the province where an individual lived 5 years previously. We rely on this information to construct internal migration at the provincial level. Finally, data on provincial GDP per capita are also used, which are taken from the General Statistics Offices of Vietnam.

To investigate the impact of internal migration on children’s well-being, we specify the estimation model as follows:

$Y_{tp}={\alpha }_1+{\alpha }_2{P}_p+{\alpha }_3{X}_{tp}+{\alpha }_4{H}_{Jp}+{\alpha }_{5}R+e_{tp}$ (1)

where Y_ip is outcomes of interest of a child i in province p, including childcare, food, access to books, and the nutritional status of children aged 0 to 5. P_p is migration in province p, defined by a fractionalization index. Specifically, $P_p=1-\sum _{\mathrm{j}=1}^{\mathrm{n}}{\mathrm{s}}_{\mathrm{j}\mathrm{p}}^2$ where n is the number of provinces in Vietnam and s is the proportion of residents in province p who migrated from province j to province p 5 years ago.³ This index ranges from 0 to 1. As the index approaches 0, migration falls, whereas heterogeneity increases as migration approaches one. To calculate migration P_p, we use the 2009 Population and Housing Census and then merge the data with data from the Vietnam Multiple Indicator Cluster Survey 2010-2011 to run regressions. Using migration at the provincial level enables us to cancel out reverse causality, suggesting that the individual-level outcomes of interest cannot affect the province-level migration. Meanwhile, migration lags 1 year because this variable, P_p, is calculated based on the 2009 Population and Housing Census. This implies that migration is relatively exogenous.

The regressions of Equation (1) also control for the characteristics of a child i in a province p, X_ip, including sex of the child, age of the child and age of the child squared. The mother’s level of education can also affect the well-being of children, so we add the mother’s education to the regressions. H_jp contains four dummy variables for the mother’s level of education (i.e., incomplete primary education as a reference group, then primary education, lower secondary education, upper secondary education, and tertiary education), and a dummy variable for the ethnicity of the household head. R is region-level fixed effects, which include five dummy variables for the regions (Red River Delta as a reference group, Northern Midlands and Mountains, North Central and Central Coastal area, Central Highlands, Southeast and Mekong River Delta). The error term is reflected in e_ip. Table A1 of the Appendix provides the descriptive statistics of dependent and independent variables.

Valuable information could be missed if we examine only the mean effect as given in Equation (1). A mean approach using standard linear regression techniques (OLS estimators) examines an average relationship between migration and the well-being of children under 5. Quantile regression enables us to quantify the relationship between the well-being of children and migration across different quantiles of the conditional distribution of the dependent variable – the well-being of children. The magnitude of the coefficient of the effect of migration on the well-being of children increases by quantiles. This means that migration has a greater effect on children with better care.

In contrast, the decreasing effect of migration on the well-being of children suggests that migration has a greater effect on children with lower well-being. It is interesting to estimate the heterogeneous effects of migration on the well-being of children under 5. Specifically, our study utilizes the unconditional quantile regression (UQR) estimator developed by Firpo et al. (2009) because it is widely believed that the UQR estimator generates more policy-relevant information than the conditional quantile regression (CQR) estimator (Khanal et al. 2018; Maclean et al. 2014).

While the purpose of OLS regression is to minimize the differences between the observed values and the model’s fitted values, quantile regression weights the differences between the observed values and the predicted values differently and then attempts to minimize the weighted differences (Buchinsky 1994; Koenker 2005; Lê Cook and Manning 2013). Since its introduction, UQR has become a common tool for finding and evaluating distributional effects on outcomes in terms of changes in the observed attributes.

The marginal effects of the unconditional quantile estimator can be determined by averaging the Recentered Influence Function with respect to the change in the distribution of the explanatory variables (Firpo et al. 2009). The coefficients of an unconditional quantile estimator can be interpreted similarly to those of an OLS estimator. The unconditional quantile estimator uses the same explanatory variables in Equation (1).

$\text{Quantile}{\left(Y_{tp}\right)}_{\theta }={\alpha }_{2\theta }+{\alpha }_{2\theta }{P}_p+{\alpha }_{3\theta }{X}_{pp}+{\alpha }_{e\theta }{H}_{jp}+{\alpha }_{5\theta }R+e_{ip}$ (2)

In the 1980s, the Vietnamese government has organized migration programs from more populated to less densely settled regions while discouraging migration from rural to urban areas, especially to major cities like Hanoi, Haiphong, and Ho Chi Minh. The government provided free transportation, housing and basic necessities for migrants at their destinations (Dang 1999). However, this migration policy was not successful due to financial and practical problems (Banister 1993). Since 1989, Vietnam has shifted from a centrally planned to a market economy. Cooperative agricultural land was allocated to individual households. In the 1990s, the markets have been gradually liberalized and in the 2000s, Vietnam got integrated into the world market and attracted a significant inflow of foreign direct investment. These factors have led to an apparent shift from organized to spontaneous migration in Vietnam. The number of inter-district migrants increased by more than half, from about 1.1 million in 1999 to 1.7 million people in 2009 (General Statistics… 2011).

Figure 1 presents a map of migration by province. We find that migration is concentrated in South Vietnam and is the highest in Binh Duong province. Binh Duong province has many industrial parks, attracting people from other provinces to work out there. Ho Chi Minh City also features considerable migration, ranking second in the country. The reason is that Ho Chi Minh City is Vietnam’s economic and financial center, accounting for one-third of the Vietnam’s GDP.

Figure 1.

Provincial migration within the mainland Vietnam. Source: authors’ own calculation from the Vietnam MICS 2010-2011. Notes: The Paracel Islands and Spratly Islands are not included in the map because data are unavailable.

Table 1 provides descriptive statistics, showing that the percentage of children under 5 fed with juice, yogurt and formula is higher for those whose mother has attained higher levels of education. For instance, children who have access to juice account for 2.4%, 9.9%, 21%, 26.8%, and 39.8% of those whose mother’s primary education is incomplete, who have completed primary education, lower secondary education, upper secondary education, and tertiary education, respectively. Also, we find that the nutritional status of children under 5, including the height for age Z-score, weight for age Z-score, and weight for height Z-score, is positively correlated with the level of the mother’s education. A higher percentage of the children of a mother with a higher level of education attend early childhood education programs, including kindergarten or community childcare, and a lower percentage of her children are left in the care of another child. The percentage of children with access to books and the number of books they have are higher for those whose mothers have a better level of education. These results suggest that the well-being of children under 5 is better for those whose mothers have a higher level of education.

Table 2 presents the well-being of children under 5 by degree of migration: low, average and high migration; this categorical variable contains categories corresponding to the three quantiles. We find that a higher percentage of children living in provinces exposed to greater migration have access to juice, yoghurt, and infant formula. In similar vein, children living in provinces with higher exposure to migration have better nutritional status, including height for age Z-score, weight for age Z-score, and weight for height Z-score. Children living in provinces more exposed to migration have a better opportunity to attend early childhood education programs and a lower probability of being left in the care of another child. Finally, the percentage of children with access to books and the number of books they have is higher for those living in provinces with higher migration.

The estimation results of migration on food and nutritional status of children under 5 are reported in Table 3 and are based on Equation (1). We establish that a 10-percentage point increase in migration increases the probability of children having access to juice by 7.2%, yogurt by 9.6%, and infant formula by 2.7% (Columns 1 to 3). Likewise, migration is statistically highly significant for the height for age Z-score, weight for age Z-score and weight for height Z-score at the 1% level (Columns 4 to 6). A 10- percentage point increase in migration raises the height for age Z-score by 0.14 standard deviation, weight for age Z-score by 0.18 standard deviation, and weight for height Z-score by 0.18 standard deviation. Those estimation results suggest that migration increases the probability of access to food and improves nutritional status of children under 5 years.

We also find that mother’s education plays an important role in improving access to food and nutritional status of children. Specifically, children whose mothers have a higher level of education are more likely to have access to foods like juice, yogurt, and infant formula (Columns 1, 2 and 3). Also, a mother’s education is positively associated with the nutritional status of children, such as the height for age Z-score, weight for age Z-score, and weight for height Z-score. Children living in households whose heads belong to Kinh/Chinese groups have better access to food and better nutritional status compared to those living in households whose heads belong to other ethnic groups.

Table 4 provides estimates of the effect of migration on childcare and access to books. The results show that migration increases the likelihood of children attending early childhood education programs, including kindergarten or community childcare. The result remains consistent when we use the hours a child attends early education as a dependent variable. Putting it differently, a 10-percentage point increase in migration results in an increase in the number of hours a child attends early education by 1.2 (Column 2). We also find that migration is negatively associated with the probability of a child being left in the care of another child. A 10-percentage point increase in migration reduces the likelihood of a child being left in the care of another child by 2.8% (Column 3). Similarly, higher migration decreases the number of days a child is left for more than an hour with another child of less than 10 years (Column 4). Finally, we find that higher migration increases access to books for a child and increases the numbers of books available to them. Those findings are consistent with the ones of Binci and Giannelli (2012) who show that migration increases schooling and reduces child labor through remittances in Vietnam.

The estimation results show that a mother’s education improves childcare and access to books. A mother with a higher level of education increases the probability of her children attending early childhood education programs and the hours of their attendance in such programs. We find that a mother’s education is negatively associated with the likelihood of children being left in the care of another child and with the number of days they are left. Children living in households whose heads belong to the Kinh/Chinese group spend more hours attending early education programs, are less likely to be left in the care of another child, and spend fewer days left with another child than those living in households whose heads belong to another ethnic group.

To run unconditional quantile regressions, we need a dependent variable with continuous values. Accordingly, we use the nutritional status of children, including height for age Z-score, weight for age Z-score, and weight for height Z-score to analyze the impact of migration across different quantiles. Table 5 reports on the estimation results for unconditional quantile regressions, showing that the effects of migration on the nutritional status of children under 5 are heterogeneous across quantiles and tend to be higher for children with better nutritional status.

Table 5 shows that holding all other factors constant, a 10-percentage point increase in migration leads to a 0.141, 0.142, and 0.194 rise in standard deviation in the height for age Z-score at the 25th, 50th and 75th quantiles, respectively. The results are not statistically significant for the height for age Z-score at the 10th and 95th quantiles. Similarly, we find that a 10-percentage point increase in migration increases the weight for age Z-score by 0.124 standard deviation at the 50th quantile, 0.196 standard deviation at the 75th quantile, and 0.443 standard deviation at the 95th quantile. We find no evidence of the impact of migration on the weight for age Z-score at the 10th and 25th quantiles. Also, a 10-percentage point increase in migration raises the weight for height Z-score by 0.127, 0.322 and 0.529 standard deviation at the 50th, 75th, and 95th quantiles, respectively. The estimation results are not statistically significant for the weight for height Z-score at the 10th and 25th quantiles. These findings suggest that migration plays a greater role for children with better nutritional status, which can be explained by the fact that better-off households have more resources and benefit more from migration.

People born in different countries or regions have been exposed to diverse life experiences, schools, and value systems. As a result, they form a variety of perspectives that allow them to perceive and solve problems differently. Previous studies show that migration leads to innovation and creativity (Alesina & La Ferrara 2005), increases the probability of introducing new product innovations (Nathan and Lee 2013), and enhances entrepreneurship (Audretsch et al. 2010; Cheng and Li 2012; Marino et al. 2012) and labor productivity (Trax et al. 2015; Parrotta et al. 2014; Buchholz 2021). Obviously, better innovation, entrepreneurship and labor productivity due to migration would increase workers’ wages. Indeed, current studies find that migration is positively associated with workers’ wages (Ottaviano and Peri 2006; Kemeny 2012; Suedekum et al. 2014; Elias and Paradies 2016). Zhao (2020) also show that rural-urban migrants are positively associated with the wages of urban natives in China. These findings suggest that migration may increase the income of native people. Parents with better income improve the well-being of their children under 5. Therefore, we hypothesize that migration and ethnical and cultural diversity associated with migration increase the well-being of children under 5 through the channel of income.

To test this hypothesis, we rerun regressions of Tables 3 and 4, adding the wealth index score as a control variable. We have no information on the income or expenditures of households, so we use the wealth index score as a proxy for household income (Sahn and Stifel 2003). We expect that the household wealth index score will capture the impact of migration on the well-being of children under 5 years of age. This means that the magnitude of the coefficient of migration on the well-being of children would be much greater in regressions without controlling for the wealth index score than in regressions that do control for the wealth index score, or we find no statistically significant effect of migration on the well-being of children.

Table 6 reports on the estimation results for the effect of migration on food and nutritional status of children under 5, controlling for the household wealth index score. We find that the magnitude of the migration coefficient on food and nutritional status of children drops considerably, and the statistically significant level also drastically declines. For example, a 10-percentage point increase in migration increases the likelihood of children having access to juice by 7.2% in the regression without controlling for the household wealth index score, and by 4.5% in the regression, controlling for the household wealth index score. A 10-percentage point increase in migration increases the weight for age Z-score by 0.18 standard deviation in the regression without adding the wealth index score, and by 0.1 standard deviation in the regression when the wealth index score is added. The estimation result is not statistically significant for the height for age Z-score in the regression, when the wealth index score is added, while it is statistically significant for the height for age Z-score in the regression when the wealth index score is not added.

Table 7 presents the estimates of the effects of migration on childcare and access to books, controlling for the household wealth index score. This time, the estimation results are not statistically significant for the dummy variable for early childhood education, hours a child attends an early education program, the dummy variable for access to books, and the number of books. The results are still statistically significant for the dummy variable for being left in the care of another child and the number of days left with another child.

The findings in Tables 6 and 7 suggest that migration affects the well-being of children under 5 through the channel of income. Specifically, migration improves workers’ wages. This leads to higher incomes for households living in provinces more affected by migration. Obviously, parents with better incomes are more likely to buy food and books for their children, thereby improving the nutritional status of their children. Similarly, other findings in China reveal that household income is a major determinant of children’s academic performance, (Liu et al. 2020) while evidence from other developing countries shows that family income plays a key role in ensuring food security for children (Bhandari 2017). It is assumed by most parents that sending their children to formal school is better than keeping them at home, the opportunity to earn better wages would encourage families to send their children to kindergarten. All these factors improve the well-being of children.

We expect that migration leads to an increase in employment opportunities for women. Children under two can attend an early childcare center, but access to early childcare centers is limited in Vietnam. The Vietnamese education system is largely public, with 90% of children aged 3-5 attending public preschools (Dang et al. 2022). Dang et al. (2022) also show sizable, positive effects of childcare on women’s employment opportunities, and these effects are greater for younger children. This means that the impact of migration on the well-being of children may be greater for mothers with children over 2 years than for those with children under the age of 2. To test this hypothesis, we use nutritional status as a dependent variable. Children of various ages have different needs, so it is appropriate to investigate the effect of migration on the height for age Z-score, weight for age Z-score, and height for age Z-score for children of various ages. Specifically, we divide the sample into two sub-samples — children under the age of 2 and those over 2.

Table 8 reports on the estimation results of the effect of migration on nutritional status by age. We find that the magnitude of the coefficient of the effect of migration on the height for age Z-score, weight for age Z-score and weight for height Z-score is larger for children aged over 2 than those under the age of 2. A 10-percentage point increase in migration increases the height for age Z-score by 0.14 and 0.17 standard deviation for children under the age of 2 and children over 2, respectively. Similarly, a 10-percentage point increase in migration raises the weight for age Z-score by 0.12 and 0.31 standard deviations for children under the age of 2 and children over 2, respectively. Finally, a 10-percentage point increase in migration is associated with a 0.12 standard deviation increase in the weight for height Z-score for children under the age of 2, and 0.3 for children over 2. These findings corroborate the impact of migration on nutritional status through the channel of income. Mothers with children over 2 years of age may have greater opportunity to participate in the labor market than those with children under the age of 2. This suggests that mothers with children over 2 may have a better income due to migration than those with children under the age of 2. Mothers with better income would spend more on their children, leading to a better nutritional status.

We are concerned that migration is correlated with provincial gross domestic product (GDP) per capita. In other words, provinces with higher GDP per capita will have a larger inflow of migrants, and thereby experience greater diversity associated with migration. Thus, our estimation results may be driven by provincial GDP per capita. As a robustness check, we rerun regressions of Tables 3 and 4, adding provincial GDP per capita as a control variable, and report the results in Tables A2 and A3. The results in Table A2 remain unchanged and almost the same as those in Table 3, that do not control for provincial GDP per capita. The same results are also found in Table A3 and Table 4. This confirms that our results are robust, even after controlling for the level of provincial GDP. Once again, these results affirm a positive contribution of migration on the well-being of children.

Using the Vietnam Multiple Indicator Cluster Survey (MIC) 2010-2011 and the 2009 Population and Housing Census, this study documents the impact of migration, defined as a fractionalization index, on the well-being of children under 5. We find that migration is positively associated with the well-being of children under 5, including access to food, improved nutritional status, and childcare. A 10-percentage point increase in migration increases the likelihood of children having access to juice by 7.2%, yogurt by 9.6% and infant formula by 2.7%. Similarly, the estimation results show that migration is positively associated with the height for age Z-score, weight for age Z-score and weight for height Z-score. Also, migration has a positive effect on the probability and hours of children attending early childhood education programs. Children living in provinces more exposed to migration are less likely to be left in the care of another child and have a higher probability of access to books.

The results also show that migration increases the probability of children having access to early education programs and books and decreases the likelihood of children being left in the care of another child. Interestingly, the effects of migration on the nutritional status of children under 5 are heterogeneous across quantiles and tend to be larger for children with better nutritional status. For instance, a 10-percentage point increase in migration increases the weight for age Z-score by 0.124 standard deviation at the 50th quantile, 0.196 standard deviation at the 75th quantile, and 0.443 standard deviation at the 95th quantile. We also document income as a channel through which migration affects the well-being of children under 5. Migration may improve wages and employment opportunities for parents, thereby leading to better well-being for children under 5 years. Non-monetary factors may also be channels through which migration influences the well-being of children under 5.

There is a substantial body of evidence in economics that shows that malnutrition in early life affects human capital accumulation, health, and socio-economic status in adulthood. Therefore, it is of the utmost importance for both economists and policy makers to identify measures to improve the well-being of children. The government can intervene to counter child malnutrition by making cash transfers. However, the market can play an important role in enhancing the well-being of children without any government interventions. Obviously, migration helps allocate redundant labor efficiently, thereby increasing economic prosperity. Also, the study findings on the relationship between migration and the well-being of children suggest that encouraging migration can be a good way to improve the well-being of children. Given that the rate of inter-province migration remained unchanged between 1999 and 2019, our policy implications may be applicable to the current situation. Also, such implications can be used for other countries with similar socio-economic contexts (General Statistics… 2020). Meanwhile, Vietnam is still maintaining a household registration system which restricts the labor movement from rural to urban areas and between cities. Household registration is regarded as a tool to place restrictions on people’s movement and the household registration book limits access to vital services as public school and health insurance for young children (World Bank… 2016). The findings of this study provide empirical evidence for policy makers to consider cancelling the household registration system.

The limitation of this study is that migration may face confounding factors, which can affect our estimation results, although this study attempts to address this issue by using migration at the provincial level with the variable lagging for over 1 year. The traditional method for dealing with confounding factors is to use instrumental variables. However, it is not easy to find a good instrumental variable that directly affects migration and does not affect the dependent variable and the error terms. The second way of addressing confounding factors is to use the difference-in-differences method. This method requires policy shocks, which affect one group and do not affect the others. Obviously, identifying such policy shocks is a thorny issue. It is also interesting to consider the effects of migration on children in terms of the risk of being abused, especially sexually abused, environments for living and schooling, and the effect of difficulties parents may face at the beginning of migration on a child (physical and mental health). However, the dataset is not available for this information. We suggest that future research should consider these issues.