Order 625985: human development

P O P U L A T I O N A N D D E V E L O P M E N T R E V I E W 4 0 ( 2 ) : 2 7 3 – 2 9 2 ( J U N E 2 0 1 4 ) 2 7 3

Effects of Parents’ Migration on the Education of Children Left Behind in Rural China

Minhui Zhou Rachel MuRphy Ran Tao

In China in 2012 an estimated 163.4 million rural migrants were living and working outside their hometowns (China National Bureau of Statistics 2012). The majority of these migrants leave their children in the countryside. Ac- cording to figures from China’s 2010 census, more than 61 million children aged between birth and age 17 years were “left behind.” Of these children, 47 percent had two parents working away, 36 percent had a migrant father, and 17 percent had a migrant mother. Left-behind children accounted for 38 percent of all rural children and 22 percent of all children in China (All China Women’s Federation Research Group 2013).

Chinese migrants leave their children behind in the countryside because key features of the country’s social welfare system discourage them from taking them to the cities. First, municipal governments use the household registration or hukou system, a legacy from China’s socialist planning past, to exclude rural migrants and their children from urban-based schooling, health care, housing, and social security (Li and Li 2010; Solinger 1999). Second, the school curriculum varies across administrative districts, so students must take examinations for senior high school and university entrance at their registered province of residence, a requirement that disadvantages those who move across provincial boundaries (Xiang 2007; Ye, Murray, and Wang 2005). Finally, owing to long working hours, it is not feasible for most migrants to resettle with their families or to raise children in the cities (Ye, Murray, and Wang 2005).

While a large body of international and China-specific literature has ex- amined the effects of migration on the well-being of children who accompany their parents, the children who remain in the origin areas have received much

2 7 4 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

less attention even though their numbers are substantially greater (Carling, Menjívar, and Schmalz bauer 2012; Toyota, Yeoh, and Ngyugen 2007). Educa- tional performance is an important aspect of these children’s well-being because it indicates how they are faring in the present and signals their prospects for a good life in the future (Schoon 2006: 6). The quest to earn sufficient funds for children’s schooling is often a principal objective motivating parents’ migration (Dreby 2010; Wan 2009; Yao and Shi 2009).

The positive effects of parental migration arise mostly from the role of remittances in alleviating household financial constraints, thereby improving children’s living conditions and nutrition as well as funding their education (Arguillas and Williams 2010; Bryant 2005; Hanson and Woodruff 2003; Lu and Treiman 2011). Remittances may also mitigate the detrimental parenting behaviors that can occur when parents face stresses associated with poverty (Duncan and Brooks-Gunn 1997; McLoyd 1998). Less has been written about the effect of parental absence itself on children’s education. Therefore, when addressing the negative effects of migration, scholars have turned to a wider literature on family structures that mostly focuses on spousal separation in the United States (Arguillas and Williams 2010; Booth 2003; Kandal and Kao 2001; Lu and Treiman 2011; Wen and Lin 2012). The family structures literature shows that the absence of a parent is associated with children’s lower academic achievement (Garfinkel and McLanahan 1986; Krein and Beller 1988; Seltzer 1994), while the presence of both parents is associated with children’s higher academic achievement (Coleman 1988; for a review see Cunha et al. 2005; Heckman, Stixrud, and Urzua 2006).

The present article draws on a survey of 1,010 children and their guard- ians—both parents and others—living in major labor-exporting regions of Anhui and Jiangxi provinces in China’s agricultural interior. It uses the pro- pensity score matching technique to deal with potential endogeneity, that is, the possibility that some of the observed difference in the educational performance of children with migrant parents and children with at-home parents may be caused by factors influencing the migration decisions of the parents rather than by the effects of the parents’ migration status per se. Whereas most research focuses on the effects of who migrates on children’s educational outcomes, we consider both the effects of who migrates and the effects of who acts as the guardians at home. We also explore the role of chil- dren’s sex in mediating the impact of parental migration and guardianship arrangements on their education, and so cast light on sex differences in the effects of disadvantage on children more generally.

Research questions

Culling from the migration studies literature and the family structures lit- erature, three main factors emerge as fundamental to how parental absence

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 7 5

through migration is likely to affect children’s educational performance. These are: the parents’ migration status, family structure/guardianship arrangements, and the children’s sex. The present state of knowledge on the effects of these three factors underpins our research questions and our analysis of the empirical data. The evidence in relation to these three factors is reviewed in turn.

Our first main research question is: what is the impact of parents’ mi- gration status on children’s educational outcomes? The migration studies literature presents a mixed picture. Some studies find that children whose parents have migrated are advantaged in their education (e.g., Asis 2006), while others find that they are disadvantaged (e.g., McKenzie and Rapoport 2006). Some studies even produce different findings from the same data set. For instance, one study based on the 2006 China Health and Nutrition sur- vey finds that the likelihood of enrollment and of years of schooling is not significantly different for children with both parents at home and for children with one parent who has migrated (Lee 2011). However, children with two migrant parents are found to fare significantly worse than other children. The negative effects are attributed to the severe care deficit that results from the absence of both parents. Another study that uses the same survey data reports the unexpected finding that children left behind by both parents are not necessarily worse off than other children, while children with one migrant parent fare the worst. The former’s relatively good educational attainment is attributed to improvements in family wealth, the actions of the migrant parents in maintaining contact with their left-behind children, and the ac- tions of guardians in compensating for the absence of two parents (Lu 2012).

Other studies aim to disentangle the different implications of maternal and paternal migration for the education of children left behind. At least two such studies pertain to China. One finds that the percentage of children of compulsory school age who drop out of school is highest among left-behind children who live alone (5.4 percent), followed by children whose mothers have migrated and who live with their fathers (4.2 percent), children whose fathers have migrated and who live with their mothers (2.3 percent), and chil- dren with two migrant parents who live with their grandparents (1.9 percent) (Duan and Wu 2009). Another study reports that in families where only the mother has migrated, children’s self-reported school performance is worse than that of children in other migrant families and in non-migrant families (Wen and Lin 2012). Even though these two studies do not isolate the effects of parental migration from the effects of other family circumstances, they suggest that mother-only migrant families may have inherent vulnerabilities.

Other studies, notably several carried out in Southeast Asia, find that among children in migrant families, children in mother-only migrant families have the worst academic attainment, with the detrimental effects increasing the longer the duration of absence (Jampaklay 2006). This research suggests that maternal migration disrupts deeply entrenched gender and generational roles,

2 7 6 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

including the mother’s caring role, causing family stress (Jampaklay 2006; Par- reñas 2005; Yeoh and Lam 2006). By contrast, father-only migration is reported to be the most advantageous to the children because of increases in family wealth, the continuity of the mother in her role as primary child-carer, and the support of the extended family (Jampaklay 2006; Parreñas 2005). However, a study of overseas migration from the Philippines finds that children in mother- only migrant families complete more years of schooling than children from non-migrant families, while children in father-only migrant families complete the same amount of schooling (Arguillas and Williams 2010).

Our second research question pertains to the effect of both paren- tal absence and post-migration guardianship arrangements on children’s educational outcomes. Most of the family structures research that has been consulted by migration studies scholars has examined the effects of single- parent guardianship on children’s educational performance and attainment. These studies report that children who are raised by single parents fare worse than children who are raised by two parents (Astone and McLanahan 1991; Coleman 1988; Garfinkel and McLanahan 1986; Hetherington, Cox, and Cox 1978 cited in Bronfenbrenner 1979: 72–80; Krien and Beller 1988; Seltzer 1994; McLoyd 1998). Explanations include disruptions to parent–child at- tachment, economic deprivation, and emotional and financial stresses that impair parenting quality.

Yet unlike in circumstances of divorce, parental absence in circumstances of migration does not usually signify an adult’s abandonment of familial re- lationships but rather a commitment to them (Nobles 2011). Additionally, parental migration does not usually indicate that the family is economically disadvantaged vis-à-vis the community average (Asis 2006; Bryant 2005). Therefore, even though the mix of parental care and income is important for the educational outcomes of children in circumstances of both divorce and parental migration, some findings in the family structures literature may not be directly relevant to one-parent migrant families.

Even though much of the migration studies literature has examined the effects of parents’ migration status on children’s education, only a few studies describe the effects of post-migration guardianship arrangements (e.g., Asis 2006; Duan and Wu 2009; Pottinger 2005). The family structures literature has also considered the effects of being raised by grandparents or other carers on children’s academic performance. The picture that emerges from this literature is that children who are raised by grandparents or other carers are disadvantaged vis-à-vis children who are raised by their parents (Sawyer and Dubowitz 1994; Sun 2003; Solomon and Marx 1995). In the US setting, where much of this research has been conducted, explanations in- clude the pressures associated with the unavailability of the biological parent and the reasons for this unavailability (Edwards and Daire 2006). Additional explanations pertain to the difficulties that grandparents can face in their

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 7 7

role as surrogate parents, including decreased energy levels, illness, pov- erty, low education levels, and a lack of desire for a parenting role (Minkler 1999, cited in Edwards and Daire 2006). Notably, this latter set of difficulties among grandparent guardians in rural China has been much discussed in the Chinese-language literature on left-behind children (Ding and Sun 2009; Li and Song 2009; Wang and Dai 2009).

Our third research question pertains to the ways in which the effects of parental migration and the effects of post-migration guardianship may vary by the children’s sex. The migration studies literature offers some insights into sex differences in the effect of parental migration on children’s educational outcomes. Studies in Mexico and Thailand report that girls from poor families experience the greatest gains in their years of schooling when remittances are received, because in circumstances of family poverty they are more likely than boys to be deprived of educational investment (Hanson and Woodruff 2003; Jampaklay 2006). By contrast, research from the Philippines observes that boys benefit more from parental remittances, which alleviate the household credit constraints that prevent investment in boys’ education. An explana- tion is that since girls generally contribute more support to their parents later in life, in circumstances of limited resources girls are more likely than boys to receive educational investment (Arguillas and Williams 2010). Other research from Mexico finds that among older boys the potential benefits of remittances are eroded by a “competing alternatives” effect in which they use their migrant parents’ knowledge of labor markets and migrate rather than continue studying (Hanson and Woodruff 2003). Finally, Lee and Park report that in Gansu province in China, the correlation between higher test scores and father-only migration (the only type of migration they consider) is strong and significant only among girls. The authors suggest that the advantage of girls in these families may be explained both by the increase in family income and by the mother’s enhanced ability to create a nurturing environment in which girls can flourish (Lee and Park 2010).

Methodology

Data

Anhui and Jiangxi were selected for this study because both provinces have a high proportion of children classified as left behind. A definition of a left- behind child commonly used by Chinese scholars and policymakers is a child with one or two parents who work outside the county and who therefore do not ordinarily live with the child (All China Women’s Federation Research Group 2013). According to figures from the 2010 census, over half of rural children in Anhui and Jiangxi were left behind, a proportion that is consider- ably higher than the national average for rural areas (ibid.).1

2 7 8 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

Appendix 1 describes our sampling strategy. After excluding cases for which we did not have complete information about test scores, the final sam- ple covered 1,010 children: 538 (53 percent) in Anhui province and 472 (47 percent) in Jiangxi province. In the case of 304 children, both parents were at home; the other 706 children had at least one parent who had migrated. There were 997 primary school students, that is, in grades 4 and 6 (69 per- cent) and 313 junior high school students (31 percent). Boys numbered 553 and comprised 55 percent of the sample, which reflects their higher propor- tion in the general school-age population, while girls numbered 457. Children in the sample were aged between 8 and 17, with an average age of 12 years.

Independent variables

Our main independent variable is parents’ migration status. We designated three main types of migration status: both parents are at home, one parent is a migrant, and both parents are migrants. We also subdivided the migration status of “one parent has migrated” into only the father has migrated and only the mother has migrated. Other independent control variables are child’s age and sex, parents’ age and education level, family characteristics of siblings and income, and a county dummy.

Dependent variables

The principal dependent variable is children’s educational performance, as measured by the average of children’s test scores for 2009 and 2010 in Chi- nese and mathematics. These test scores were transcribed by each student’s homeroom teacher from school records and then standardized by school and class year. Standardization was necessary because, owing to variations across schools in marking scales, the raw data were not amenable to comparison outside of a school’s single grade.

Statistical methods and model specification

When analyzing the impact of parents’ migration on children’s educational performance, it is necessary to address the problem of endogenity because the parents’ decision to migrate may not be random. The decision may be in- fluenced by the families’ attributes, the parents’ attributes, and the children’s attributes. Therefore, the differences between children who are left behind and children who are not left behind may reflect not the impact of being left behind but rather the covariates that determine who is left behind. For example, parental poverty might influence being left behind and simultane- ously influence children’s test scores through avenues that have nothing to do with whether they are left behind. For this reason, regression analysis that simply compares the scores of left-behind children and not-left-behind chil-

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 7 9

dren, such as the OLS method, can lead to bias, with the impact of parental migration on children’s educational performance being either underestimated or overestimated.

In the present study, we would ideally like to observe the differences in test scores for individuals when they have been left behind and again for the same individuals when they have not been left behind. But either each child has been left behind or he has not; there is only one state of the world. It has therefore been necessary to find other ways of dealing with the endogeneity problem. Social scientists commonly select one of the following approaches: instrumental variables, social experiments, or propensity score matching (PSM). There are difficulties in using the instrumental variables approach. This is because the selected instrumental variable must have a consistent re- lationship to the main independent variable but must not explain any of the variation in the model other than through its impact on the instrumented variable. Therefore, when studying the effects of migration, it is necessary to find a variable that potentially affects parents’ propensity to migrate yet is external to the families. This is difficult if not impossible. The social experi- ment approach, which would randomly assign children to left-behind and non-left-behind groups, is clearly infeasible.

PSM, first proposed by Rosenbaum and Rubin (1983) and Heckman, Ichimura, and Todd (1998), is arguably the best available method for dealing with the endogeneity problem. In this method, the probability of each indi- vidual receiving a treatment (being left behind) is estimated by using probit or logit regression on researcher-selected individual and family characteristics. The resulting estimates are known as propensity scores. Each individual in the treated group is subsequently matched to a “nearest neighbor” from the control group on the basis of the independent control variables described above. In this study, the nearest neighbor is the not-left-behind child whose propensity score is most similar to that of the selected left-behind child.

The reasoning is that if members of the treated and control groups are identical to each other in all other respects, the differences between the groups can be attributed to the effects of parental migration (Lu and Treiman 2011: 1131). Results from randomized experiments suggest that the PSM technique can reduce bias by 58 to 96 percent (Shadish, Clark, and Steiner 2008 cited in Lu and Treiman 2011: 1131). The concrete steps for using PSM are explained in Appendix 2.

Analysis

Parental migration and child guardianship arrangements

Of the children in the sample, 43 percent lived in families where both parents were migrants, and nearly a quarter in families with only a migrant father

2 8 0 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

(Table 1). Even though the proportion of children with migrant parents is higher in our sample than in samples reported for other China-based surveys, the breakdown of the parents’ migration status resembles that described in other rural Chinese settings (Chen et al., 2009; Duan and Wu 2009; Lee 2011; Lu 2012; Lee and Park 2010; Wen and Lin 2012). Specifically, two-parent and father-only migrants are the dominant arrangements, while mother-only migrants are in the minority.

The parents’ duration of absence varies depending on which parent is the migrant. The average duration of absence is the longest, around five years, when both parents are migrants. When both parents are migrants, the parents have on average been away from their children for nearly 90 percent of the time since they started school.

When both parents are migrants, care by grandparents is the dominant arrangement. Specifically, 54 percent of children living in two-parent migrant families were cared for by grandparents during the week, while 82 percent of them were cared for by grandparents on the weekend. In families with only one migrant parent, the vast majority of children were cared for by the at- home parent. Additionally, a significant proportion of children in our survey reported looking after themselves during the week. This reflects the avail- ability of facilities for boarding at school from Monday to Friday.

Table 2 shows the variables that need to be controlled in the analysis because they could influence both the parents’ migration behaviors and children’s test scores. Although not shown in the table, it is of interest that parents from migrant families are on average slightly younger than parents from non-migrant families, while their average level of education is slightly higher: this reflects the generally better urban employment prospects of younger and better-educated migrants. Also included as a control variable is per capita income.

Tables 3–5 present the results of both OLS and PSM analysis on the children’s standardized test scores. We present both OLS and PSM results in order to demonstrate the impact of controlling for endogeneity on the analy- sis. In the columns labeled OLS, the figures give the effect on test scores of

TABLE 1 Distribution of children in the survey sample by parents’ migration status

Total sample Anhui Province Jiangxi Province

No. of No. of No. of Parents’ status children Percent children Percent children Percent

Both parents have migrated 437 43.3 180 33.5 257 54.4 Both parents at home 304 30.1 166 30.9 138 29.2 Only the mother has migrated 25 2.5 5 0.9 20 4.2 Only the father has migrated 244 24.2 187 34.8 57 12.1

Total 1,010 538 472

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 8 1

being left at home as estimated by OLS regression of test scores on left-behind status and the independent control variables described above. For example, in Table 3, the 2010 Chinese test scores of children who have at least one migrant parent (the treatment group) are estimated to be higher than the scores of children whose parents are at home, the difference between them being 0.07. The column labeled PSM gives the difference between the means of the PSM-matched control and treatment groups (the average effect of the treatment on the treated, or ATT, as described in Appendix 2). In this case, the 2010 Chinese test scores of the children with at least one migrant parent were, on average, 0.01 higher than those of their nearest neighbors in the control group. Also available from the authors are results obtained from an alternative to the nearest neighbor (NN) matching approach known as the kernel algorithm. It is customary for researchers to use at least two different algorithms as a robustness check on their results (Chen et al. 2009).

Table 3 examines the standardized test scores of children in families where at least one parent is a migrant. The results indicate no statistically significant difference in the scores of children with at least one migrant par-

TABLE 2 Descriptive statistics for the main control variables

Total Anhui Jiangxi Control variable sample Province Province Boys Girls

Children’s age (years) 12.2 11.7 12.8 12.1 12.3 Children’s sex (percent male) 55 57 53 — — Parents’ average age (years) 39.0 38.8 39.3 39.2 38.8 Parents’ average education level (years) 6.1 5.9 6.2 6.0 6.2 Child has a sibling (percent) 62 55 70 51 76 Per capita family income (log value) 8.7 8.8 8.7 8.8 8.6

TABLE 3 The influence of having at least one migrant parent on children’s test scores, OLS regression and nearest neighbor (NN) PSM results, Anhui and Jiangxi provinces, both sexes

NN OLS PSM

2010 Chinese scores 0.07 0.01 (0.08) (0.13)

2010 math scores 0.11 0.07 (0.09) (0.13)

Average of Chinese scores for 2009 and 2010 0.08 –0.08 (0.08) (0.12)

Average of math scores for 2009 and 2010 0.11 0.03 (0.08) (0.11)

NOTE: Treatment group = children with at least one migrant parent.

2 8 2 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

ent and the scores of other children. Table 4 adds guardianship arrangements to the analysis. The table considers the test scores of children who live in at least one-parent migrant families and who usually live with a guardian other than a parent. The results show that these children fare worse in their Chinese scores. Moreover, when the results are disaggregated by children’s sex, being cared for by an adult other than a parent corresponds with significantly lower math scores among boys but not among girls.

TABLE 5 The influence of having two migrant parents on children’s test scores, OLS regression and nearest neighbor (NN) PSM results, Anhui and Jiangxi provinces, both sexes

Both sexes Boys Girls

NN NN NN OLS PSM OLS PSM OLS PSM

2010 Chinese scores –0.10 –0.18** –0.08 –0.32** –0.07 –0.07

(0.08) (0.09) (0.12) (0.13) (0.10) (0.10)

2010 math scores –0.07 –0.16* –0.13 –0.36*** –0.01 0.00

(0.08) (0.09) (0.12) (0.14) (0.12) (0.13)

Average of Chinese scores –0.09 –0.22*** –0.09 –0.29* –0.05 –0.07 for 2009 and 2010 (0.09) (0.09) (0.11) (0.15) (0.10) (0.10)

Average of math scores –0.05 –0.20** –0.11 –0.29* 0.01 0.03

for 2009 and 2010 (0.05) (0.09) (0.12) (0.14) (0.11) (0.13)

*Significant at p < 0.10; **p < 0.05; ***p < 0.01. NOTE: Treatment group = children with two migrant parents.

TABLE 4 The influence of having at least one migrant parent and living with a non-parent guardian on children’s test scores, OLS regression and nearest neighbor (NN) PSM results, Anhui and Jiangxi provinces, both sexes

Both sexes Boys Girls

NN NN NN OLS PSM OLS PSM OLS PSM

2010 Chinese scores 0.01 –0.20** 0.06 –0.16 –0.02 –0.09 (0.07) (0.09) (0.11) (0.15) (0.09) (0.11)

2010 math scores 0.04 –0.14 0.04 –0.28** 0.04 –0.01 (0.08) (0.10) (0.12) (0.13) (0.11) (0.13)

Average of Chinese scores 0.00 –0.23** 0.04 –0.24 –0.01 –0.09 for 2009 and 2010 (0.07) (0.10) (0.11) (0.16) (0.09) (0.10)

Average of math scores 0.07 –0.08 0.07 –0.22* 0.06 0.00 for 2009 and 2010 (0.08) (0.09) (0.11) (0.13) (0.11) (0.12)

*Significant at p < 0.10; **p < 0.05. NOTE: Treatment group = children with at least one migrant parent and care usually provided by a non-parent guardian.

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 8 3

Table 5 analyzes the test scores of children in families where both parents have migrated. Children from two-parent migrant families have lower test scores than other children. Disaggregating the results by children’s sex shows that having two migrant parents has a significantly negative impact on boys’ Chinese and math scores but not girls’.

Comparing the scores of children from families where two parents have migrated to those of children from families where only one parent has mi- grated, we again find a significant disadvantage among boys with two migrant parents (see online Table A).2

Our results indicate that regardless of whether left-behind children live in mother-only or father-only migrant families, there is no significant difference in test scores when compared with children whose parents are at home (see online Table B). However, when the test scores of children in one-parent migrant families are compared with those of children in two- parent migrant families, the higher scores of the former are statistically significant only for children from father-only migrant families. Moreover, children in father-only migrant families have significantly higher scores than children in mother-only migrant families (see online Table B). This may be attributed at least partially to the fact that when only the father migrates, parental care by the mother remains the dominant guardianship arrangement. The higher scores also lend weight to other findings about the importance of maternal care and the difficulties that extended families face in substituting for this care (Jampaklay 2006; Wen and Lin 2012).

The effects of parental absence longer than three years

The preceding discussion demonstrated that children with two migrant parents have lower test scores than other children. To better understand how parental migration may affect children’s educational performance, we consider the effect of the duration of parents’ absence on children’s test scores. Table 6 analyzes data only for children from two-parent migrant families. The treatment group is children whose mother and father have both been absent for three or more years since they began school. Three years was selected as the cutoff because children in the sample are from grades 4, 6, and 8, so their parents could not have been away for more than four years since they started school. Addition- ally, as noted earlier, the average duration of parents’ absence from families with two migrant parents is longer than for other migrant families. The total number of children in the sample with two migrant parents is 437 and, of these, 410 reported an absence of at least three years. The results in Table 6 indicate that children whose parents have been away for more than three years have lower test scores than other children. Moreover, the sex of the child is again important in that the correlation between longer periods of parental absence and poorer educational performance is strong and significant only among boys.

2 8 4 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

The effects of parental care versus family income

Thus far our analysis of the correlation between parents’ migration status and children’s educational performance has controlled for the effects of annual per capita family income. This has allowed us to concentrate solely on the effects of parental availability on children’s education. Yet, as noted earlier, parental income may also affect children’s test scores. Therefore, it is useful to compare the effects of parental availability with the effects of parental income. Our survey data suggest that families with migrants have higher income than other families. For instance, in 2009 the average per capita income of dif- ferently structured families (including grandparents, parents, and children) was as follows: families with two migrant parents—7,824 yuan; families with one migrant parent—7,600 yuan; and non-migrant families—5,015 yuan. For reference, the annual per capita income of all families in our survey was 6,815 yuan. Our data also show that in 2009 migrant fathers earned more than migrant mothers. In families with two migrant parents, fathers earned, on average, 24,995 yuan per year and mothers 18,873 yuan, while in families with one migrant parent, fathers earned 24,423 yuan and mothers earned 15,231 yuan.

Given the higher income of migrant families, we determined the rela- tive effect of parental income versus parental care on children’s academic performance. (OLS estimates are provided in online Table C.) Although this approach may present endogenity problems, the estimates indicate that re- gardless of whether the control group is children under the care of at least one parent or only one parent, having two migrant parents has negative care effects and positive income effects. Moreover, the income effects are much smaller in scale than the care effects even if we take into account the impact

TABLE 6 Effects of both parents being absent at least three years on children’s educational performance: OLS regression and nearest neighbor (NN) PSM results

Both sexes Boys

NN NN OLS PSM OLS PSM

2010 Chinese scores –0.10 –0.20** –0.11 –0.28* (0.08) (0.09) (0.12) (0.14)

2010 math scores –0.07 –0.26*** –0.16 –0.34** (0.08) (0.09) (0.12) (0.14)

Average of Chinese scores –0.09 –0.25*** –0.11 –0.37** for 2009 and 2010 (0.07) (0.09) (0.12) (0.15)

Average of math scores –0.05 –0.16* –0.13 –0.33** for 2009 and 2010 (0.08) (0.09) (0.12) (0.13)

*Significant at p < 0.10; **p < 0.05. NOTE: Treatment group = both parents have been migrants for at least three years.

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 8 5

of the different units of measurement for the two variables. Specifically, when the control group is (A), children with at least one parent at home, the average effect of care is 5.6 times that of income. When the control group is (B), children with one migrant parent, the average effect of care is 26.7 times that of income. Hence, comparison of B with A suggests that the impact of income on children’s test scores is less than the impact of parental availabil- ity. Therefore, if migrant parents want to offset the negative effect of their absence on their children, they need to earn at least an additional 5,600 yuan per capita per year. In fact, however, families with two migrant parents earn an additional average of only 2,200 yuan more per year than families with one migrant parent.

OLS estimates further show that boys are more adversely affected than girls by the deficit in parental care. (OLS estimates for boys are provided in online Table D.) In the regression on boys in the sample, if the control group is (A), children with at least one parent at home, the average effect of parental absence is 12.8 times that of income, while if the control group is (B), children with one migrant parent, the average effect of parental absence is 128.1 times that of income. Clearly, the income of families with two migrant parents is not high enough to offset the negative effect of parental absence on boys’ educational performance. Yet our finding is not just one of boys’ disadvantage but also one of girls’ advantage because girls whose parents have migrated benefit more than boys from improvements in family income while suffering less from the absence of parental care. (OLS estimates for girls are provided in online Table E.) To reiterate, the combination of changes in both parental availability and family income associated with the migration of two parents has markedly different effects on boys and girls.

Conclusion

This article has analyzed data from a randomized survey conducted in four counties in the major labor-exporting provinces of Anhui and Jiangxi to ex- amine the effect of parental migration on the educational performance of the children left behind. We used propensity score matching, a technique that mitigates endogenity, to conduct a cross-sectional comparison of children’s educational performance by their parents’ migration status. The analysis has shown that the children’s Chinese and mathematics test scores are signifi- cantly lower only when both parents have migrated, while the migration of one parent has little effect. However, when we also consider the daily guard- ianship arrangements, the adverse effects associated with parental absence become more pronounced. Moreover, among children with two migrant parents, the longer the duration of absence, the lower the test scores, a find- ing that is of particular concern given that migrants’ duration of absence has been increasing (Rozelle et al. 1999).

2 8 6 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

We also found that the migration of two parents is only significantly correlated with poorer academic performance among boys. When the child is male, migrant parents’ strategy of arranging for grandparents to care for their children does not adequately compensate for the decline in parental availability. We observed further that when the child is male, parental earn- ings also do not adequately offset the negative effect of the parenting deficit. Although we recognize that migration itself is often aimed at providing for the future of the child (Dreby 2010; Boehm et al. 2011), it is not feasible for most parents to earn the additional income that would be required to mitigate the detrimental impact of their absence on their sons’ educational performance.

Our finding that two-parent migration disproportionately adversely affects the educational performance of boys corresponds to findings from re- search conducted in the vastly different socio-cultural and economic settings of the United States and Europe. Specifically, a large literature on family struc- tures and education suggests that boys are ”more vulnerable to early stressors” than girls and that they are more adversely affected by non-maternal care arrangements (Brooks-Gunn, Han, and Waldfogel 2002). This literature also suggests that girls are better endowed than boys with the protective psycho- social qualities that enhance their resilience in the face of reduced parental input (Bertrand and Pan 2013).

Culturally specific dynamics may also be at work. For instance, research has revealed that parenting customs in Chinese societies can differ by the child’s sex. Some scholars have suggested that boys are raised with an em- phasis on being independent and economically successful in life so that they can later support their families, while girls are raised with an emphasis on being dependent and fulfilling their relational obligations (Bond 1991 cited in Chen and Liu 2012: 487). Parents’ different approaches to raising their sons and daughters may thus contribute to the greater vulnerability of the former and the greater resilience of the latter when a child is confronted with family separation.

To conclude, even though the implications of parental migration for children’s education vary by family structure, guardianship arrangements, and sex, it is clear that the well-being of large numbers of children is sig- nificantly adversely affected.3 We believe that the dominant policy response to the difficulties faced by left-behind children should involve adopting the fundamental and multi-faceted reforms in education, housing, and labor protection that would enable rural families to benefit more fully from urban employment and to settle with their sons and daughters in the cities. In the absence of such reforms, the considerable sacrifice that many rural people make on behalf of their children will impose a further, unintended, sacrifice on the next generation.

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 8 7

Appendix 1: Sampling strategy

Our survey sample was drawn using a multi-stage stratified design and involved a random sampling procedure at each of five stages. In the first stage, we selected Anhui and Jiangxi, two provinces with a high proportion of children with migrant parents. In the second stage, we randomly chose 2 out of 16 prefectures in Anhui and 2 of 11 prefectures in Jiangxi. Within each prefecture, we randomly chose one county. In the third stage, we used a list of all townships ranked by per capita income to randomly select a richer township and a poorer township from within each county. This pro- duced a total of eight townships.

In the fourth stage we selected the schools and grades 4, 6, and 8. Children in grades 4 and 6 are usually in primary school while children in grade 8 are in junior high school. However, in one township in Jiangxi province the grade 6 students were in junior high school. In Jiangxi province, owing to school mergers, each township had only one complete primary school, that is, a school that offered the full range of grades, so these schools were included in the survey. Every township in Anhui province had several complete primary schools. Given the possibility of differences in the backgrounds of children in the central primary school located in the township and the schools located in the villages, we randomly selected one or more village schools as well as the main township school in order to form the dataset. The final sample comes from 24 schools: 8 middle schools, 8 central primary schools, and 8 village primary schools.

In the final stage of sampling, 120 students were randomly selected in each of 7 out of 8 townships, namely, 40 students each from grades 4, 6, and 8. Sampling was facilitated by a government policy to “Care for Left Behind Children,” which required schools to keep a record of which children’s parents had migrated. This requirement enabled the survey team to calculate the numbers of students from migrant families and non-migrant families for each grade for inclusion in the sample. The survey team then randomly selected students from the respective name lists of the left-behind children and the other children. In one township, however, 180 students rather than 120 students were randomly selected. This is because we initially carried out our survey in Jiangxi province and subsequently assumed that the conditions in Anhui province would be the same. Hence, in the first township that we visited in Anhui, we drew a proportionate sample of left-behind children and other children from the central school. However, after talking with teachers we learned that the distribution of schools in Anhui was different from that in Jiangxi. With the help of the county education bureau we determined the ratio of left-behind children to other children for the whole township rather than just for the central school and then used this as the basis for sampling. Next we added cases from village schools to the sample already drawn from the central school, which resulted in the inclusion of 60 addi- tional questionnaires for that township. We then determined our sampling strategy for the three remaining townships located in Anhui. Specifically, a sampling quota of 40 children for each grade was distributed among the central and village schools based on the ratio of the number of students in the central school to the number of children in schools in each township, and then the students were randomly selected from the respective name lists.

2 8 8 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

Appendix 2: Propensity score matching

The concrete steps for using propensity score matching are as follows. First, a probit or logit model (in our study we use a probit model) is used to estimate the propensity score p(X) of every child where X is a set of researcher-selected variables. In our study, the propensity score is the probability of becoming a left-behind child. In determining the propensity score, the researcher controls for the variables that could indepen- dently influence both (1) whether or not an individual receives the treatment and (2) the outcome variable, such that the treatment variable—in this case the child’s left-behind status—functions as an exogenous independent variable (D, equal to 1 if the child is left behind and 0 if not). In our study, X includes four kinds of variables: (1) children’s age and sex (Chen et al, 2009); (2) parents’ age and education level; (3) the family characteristics of siblings (Steelman and Mercy 1980) and income (Blau 1999); (4) a county dummy to control for county characteristics such as local migra- tion networks, educational quality, and so on.

Next, the researcher chooses a matching algorithm to compute the PSM estimator. We have used two PSM approaches, the nearest neighbor approach and, as a check on robustness, the kernel approach (results for the latter available from the authors). To implement the nearest neighbor approach, each untreated (non-left-behind, D=0) child is matched to a treated (left-behind, D=1) child whose propensity score comes nearest. The process is repeated until all left-behind children have been matched. Each child is, in effect, assigned a doppelgänger, as close as possible in all respects except for having received or not received the treatment. The difference between the test score of the treated individual (Score 1) and the test score of an untreated individual (Score 0) is computed. The mean of the differences between the matched nearest neighbors is calculated and serves as an estimate of the expected value of the effect of parental migration on the children’s test scores. This mean, conventionally denoted τ, is referred to as the average effect of the treatment on the treated, or ATT. Using standard terminology,

τ ATT PSM = E

p(X) | D=1 [E(Score

1 | D=1, p(X)) – E(Score

0 | D=0, p(X))]

Notes

This project was funded by a British Academy Career Development Grant, and we gratefully acknowledge this support that made our sur- vey and collaboration possible. We are also grateful for supplementary funding from an Oxford University John Fell Fund Grant.

1 It is not always easy to classify an indi- vidual as either a migrant or a non-migrant. For instance, when we asked children and guardians “who in your family works out- side?,” some respondents replied according to who was working outside at the time of the survey, while others replied according to

what was the usual situation. In 170 cases in our sample, children and their guardians gave different replies about who was absent. To determine a classification in these cases, we considered children’s answers to a question about who usually looks after them, in order to ascertain which parent or parents are usu- ally absent. We also used information from the guardians’ questionnaires to calculate how long the parent had been away from home since the child started school, and used “more than 50 percent of the time” as the benchmark to be classified as a migrant. Hence some

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 8 9

children were counted as a having a migrant parent even though the parent may have been at home at the time of the survey.

2 Online tables are available at http://www. ccsp.ox.ac.uk/sites/sias/files/ documents/Zhou_ Murphy_Tao_PDR_2014_OnlineTables.pdf.

3 We believe that the long-term policy approach for addressing the difficulties faced by spatially separated families does not lie, as some Chinese scholars and policymakers propose, in expanding rural boarding facili- ties (Chen et al. 2009; Wan 2009; Xie 2009; Yan and Zhu 2006; Yao and Shi 2009; Zhou

2007) because such an approach could argu- ably encourage two-parent migration. This conjecture is informed by two observations from our fieldwork. First, two-parent migra- tion was more common in our survey counties in Jiangxi where boarding, particularly at the primary school level, was widespread. Second, some parents told us that the presence of boarding facilities enables them to feel more at ease with the idea of leaving their children behind, because teachers would watch over them during the week and the burden on grandparents would be reduced.

References

All China Women’s Federation Research Group. 2013. “Research report into the situation of rural left behind children and rural to urban migrant children,” China Women’s Movement (6): 30–34. (In Chinese.)

Arguillas, Marie J.B. and Lindy Williams. 2010. “The impacts of parents’ overseas employ- ment on educational outcomes of Filipino children,” International Migration Review 44(2): 300–319.

Asis, Maruja. 2006. “Living with migration: Experiences of left-behind children in the Philip- pines,” Asian Population Studies 2(1): 45–67.

Astone, Nan Marie and Sara S. McLanahan. 1991. “Family structure, parental practices and high school completion,” American Sociological Review 56(3): 309–320.

Bertrand, Marianne and Jessica Pan. 2013. “The trouble with boys: Social influences and gen- der gap in disruptive behaviour,” American Economic Journal: Applied Economics 5(1): 32–64.

Blau, David M. 1999. “The effect of income on child development,” The Review of Economics and Statistics 81(2): 261–276.

Boehm, Deborah A., Julia Meredith Hess, Cati Coe, Heather Rae-Espinoza, and Rachel R. Reynolds. 2011. “Introduction: Children, youth and the everyday ruptures of migration,” in Cati Coe, Rachel R. Reynolds, Deborah A. Boehm, Julia Meredith Hess, and Heather Rae-Espinoza (eds.), Everyday Ruptures: Children, Youth and Migration in Global Perspective. Nashville: Vanderbilt University Press, pp. 1–22.

Bond, Michael H. 1991. Beyond the Chinese Face: Insights from Psychology, Oxford: Oxford Uni- versity Press.

Booth, Margaret Z. 2003. “The impact of parental availability on Swazi students’ school achievement: A nine-year longitudinal study,” International Journal of Educational Devel- opment 23: 257–274.

Bronfenbrenner, Urie. 1979. The Ecology of Human Development: Experiments by Nature and Design. Cambridge: Harvard University Press.

Brooks-Gunn, Jeanne, Wen-Jui Han, and Jane Waldfogel. 2002. “Maternal employment and child cognitive outcomes in the first three years of life: The NICHD study of early child care,” Child Development 73(4): 1052–1072.

Bryant, John. 2005. “Children of international migrants in Indonesia, Thailand and the Philip- pines: A review of evidence and policies,” Innocenti Working Paper No 05-2005, Florence, UNICEF Innocenti Research Centre.

Carling, Jøregen, Cecilia Menjívar, and Leah Schmalzbauer. 2012. “Central themes in the study of transnational parenthood,” Journal of Ethnic and Migration Studies 38(2): 191–217.

2 9 0 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

Chen, Jennifer Jun-Li and Xiaodong Liu. 2012. “The mediating role of perceived parental warmth and parental punishment in the psychological wellbeing of children in rural China,” Social Indicators Research 107: 483–508.

Chen, Xinxin, Qiuqiong Huang, Scott Rozelle, Yaojiang Shi, and Linxiu Zhang. 2009. “Effect of migration on children’s academic performance in rural China,” Comparative Economic Studies 51(3): 323–343.

China National Bureau of Statistics. 2012. “2012 Report on the Monitoring Survey of the Na- tion’s Peasant Workers (nongmingong).” http://www.gov.cn/gzdt/2013-05/27/content_ 2411923.htm. Accessed 1 May 2014.

Coleman, James S. 1988. “Social capital in the creation of human capital,” American Journal of Sociology 94: S95–S120.

Cunha, Flavio, James J. Heckman, Lance Lochner, and Dimitriy V. Masterov. 2005. “Interpret- ing the evidence on lifecycle skill formation,” Working Paper 11331 (May), National Bureau of Economic Research, Cambridge, Massachusetts.

Ding, Tongfang and Deyu Sun. 2009. “Status investigating on family education of children left in Quanjiao Anhui,” Journal of Chuzhou University 11(4): 75–77. (In Chinese.)

Dreby, Joanna. 2010. Divided by Borders: Mexican Migrants and Their Children, Berkeley: University of California Press.

Duan, Chengrong and Lili Wu. 2009. “The status quo of rural left behind children in China and its analysis,” Journal of Chongqing Technology and Business University 26(1) (February): 24–30. (In Chinese.)

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

Edwards, Oliver W. and Andrew P. Daire. 2006. “School aged children raised by their grand- parents: Problems and solutions,” Journal of Instructional Psychology, 33(2): 113–119.

Garfinkel, Irwin and Sara S. McLanahan. 1986. Single Mothers and Their Children: A New American Dilemma. Washington, DC: Urban Institute Press.

Hanson, Gordon H. and Christopher Woodruff. 2003. “Emigration and educational attainment in Mexico.” University of California at San Diego. http://irps.ucsd.edu/assets/022/8772.pdf.

Heckman, James J., Hidehiko Ichimura, and Petra Todd. 1998. “Matching as an econometric evaluation estimator,” Review of Economic Studies 65: 261–294.

Heckman, James, Jora Stixrud, and Sergio Urzua. 2006. “The effects of cognitive and non- cognitive abilities on labor market outcomes and social behaviour,” Working Paper 12006 (January), National Bureau of Economic Research, Cambridge, MA.

Hetherington, E. Mavis, Martha Cox, and Roger Cox. 1978. “The aftermath of divorce,” in Joseph H. Stevens and Marilyn Mathews (eds.), Mother/Child, Father/Child Relations. Wash- ington, DC: National Association for the Education of Young Children.

Jampaklay, Aree. 2006. “Parental absence and children’s school enrolment: Evidence from a longitudinal study in Kanchanaburi, Thailand,” Asian Population Studies 2(1): 93–110.

Kandal, William and Grace Kao. 2001. “The impact of temporary labor migration on Mexican children’s educational aspirations and performance,” International Migration Review 35(4): 1205–1231.

Krien, Sheila Fitzgerald and Andrea H. Beller. 1988. “Educational attainment of children from single parent families: Differences by exposure, gender and race,” Demography 25(2): 221–234.

Lee, Leng and Albert Park. 2010. “Parental migration and child development in China,” Work- ing Paper, Gansu Survey of Children and Families.

Lee, Ming-Hsuan. 2011. “Migration and children’s welfare in China: The schooling and health of left behind children,” The Journal of Developing Areas 44(2): 165–182.

Li, Limei and Si-ming Li. 2010. “The impact of variations in urban registration within cities,” in Martin King Whyte (ed.), One Country, Two Societies: Rural-Urban Inequality in Contemporary China. Cambridge: Harvard University Press, pp. 188–215.

M i n h u i Z h o u / r a C h E l M u r p h y / r a n t a o 2 9 1

Li, Rui and Nan Song. 2009. “A study of the protection of rights and interests of rural left be- hind children,” Journal of Beijing University of Agriculture 24: 70–73. (In Chinese.)

Lu, Yao. 2012. “The education of children left behind in rural China,” Journal of Marriage and Family 74(2): 328–341.

Lu, Yao and Donald Treiman. 2011. “Migration, remittances and educational stratifica- tion among blacks in apartheid and post-apartheid South Africa,” Social Forces 89(4): 1119–1144.

McKenzie, David and Hillel Rapoport. 2006. “Can migration reduce educational attainment? Evidence from Mexico,” World Bank Working Paper No. 3952.

McLoyd, Vonnie C. 1998. “Socioeconomic disadvantage and child development,” American Psychologist 53(2): 185–204.

Minkler, Meredith. 1999. “Intergenerational households headed by grandparents: Contexts, realities and implications for policy,” Journal of Aging Studies 13: 199–218.

Nobles, Jenna. 2011. “Parenting from abroad: Migration, non-resident father involvement and children’s education in Mexico,” Journal of Marriage and Family 73(4): 729–746.

Parreñas, Rhacel Salazar. 2005. Children of Global Migration. Stanford, CA: Stanford University Press.

Pottinger, Audrey, M. 2005. “Children’s experience of loss by parental migration in inner city Jamaica,” American Journal of Orthopsychiatry 75(4): 485–496.

Rosenbaum, Paul R. and Donald B. Rubin. 1983. “The central role of the propensity score in observational studies for causal effects,” Biometrika 70(1): 41–55.

Rozelle, Scott, Li Guo, Minggao Shen, Amelia Hughart, and John Giles. 1999. “Leaving China’s farms: Survey results of new paths and remaining hurdles to rural migration,” The China Quarterly 158: 367–393.

Sawyer, Richard J. and Howard Dubowitz. 1994. “School performance of children in kinship care,” Child Abuse and Neglect 18: 587–597.

Schoon, Ingrid. 2006. Risk and Resilience: Adaptations in Changing Times, Cambridge: Cambridge University Press.

Seltzer, Judith A. 1994. “Consequences of marital dissolution for children,” Annual Review of Sociology 20: 235–266.

Shadish, William R., M.H. Clark, and Peter M. Steiner. 2008. “Can nonrandomized experiments yield accurate answers? A randomized experiment comparing random and nonrandom assignments,” Journal of American Statistical Association 103(484): 1334–1344.

Solinger, Dorothy. 1999. Contesting Citizenship in Urban China: Peasant Migrants, the State and the Logic of the Market. Berkeley: University of California Press.

Solomon, Jennifer C. and Jonathan Marx. 1995. “To grandmother’s house we go: Health and school adjustment of children raised solely by grandparents,” Gerontologist 35: 386–394.

Steelman, Lala Carr and James A. Mercy. 1980. “Unconfounding the confluence model: A test of sibship size and birth order effects on intelligence,” American Sociological Review 45(4): 571–582.

Sun, Yongmin. 2003. “The wellbeing of adolescents in households with no biological parents,” Journal of Marriage and Family 65: 894–909.

Toyota, Mika, Brenda S.A. Yeoh, and Liem Ngyugen. 2007. “Editorial introduction: Bringing the left behind back into view in Asia,” Population, Space and Place 13(3): 157–161.

Wan, Meng. 2009. “Statistical analysis of survey of left-behind children in rural areas in Jiangxi province,” Technological Forum 6: 31–33. (In Chinese.)

Wang, Zhanghua and Lichao Dai. 2009. “Education of the rural home—Left behind children and intervention of social work,” Journal of Hebei Normal University 11: 125–129. (In Chinese.)

Wen, Ming and Danhua Lin. 2012. “Child development in rural China: Children left behind by their migrant parents and children of nonmigrant families,” Child Development 83(1): 120-136.

Xiang, Biao. 2007. “How far left behind are the left behind?: A preliminary study in rural China,” Population, Place and Space 13(3): 179–191.

2 9 2 E d u C a t i o n o f C h i l d r E n l E f t B E h i n d i n r u r a l C h i n a

Xie, Jianwu. 2009. “Problems and countermeasures for the education of left behind children,” Education Research Monthly 8: 79–81. (In Chinese.)

Yan, Honghe and Xiatao Zhu. 2006. “Survey of the influence of the boarding school system on the education of left behind children in the countryside,” Modern Middle School and Primary School Education 143: 4–6. (In Chinese.)

Yao, Bianfang and Ya Shi. 2009. “Analyses on the management policy of left-behind children education,” Education and Teaching Research 23: 18–22. (In Chinese.)

Yeoh, Brenda and Theodora Lam. 2006. “The costs of (im)mobility: Children left behind and children who migrate with a parent,” in Perspectives on Gender and Migration. Bangkok: United Nations Economic and Social Commission for Asia and the Pacific, pp. 120–149.

Zhou, Xiangqian. 2007. “Brief discussion of making boarding schools the second home of left behind children,” Research into Basic Education 7: 15–16. (In Chinese.)