A Matter of Gendered Investment: Impacts of Internal Migration on Child Education in Indonesia. Paul Berbée

Master in Economic Development and Growth A Matter of Gendered Investment: Impacts of Internal Migration on Child Education in Indonesia Paul Berbée pa0831be-s@student.lu.se Abstract: This study evaluates investments in education and schooling outcomes of children in households that engage in internal migration. Using panel data from Indonesia, community and household fixed effects are employed to account for unobserved heterogeneity and self-selection of migrants. Maternal migration is found to be associated with an average 30% reduction of educational expenditure and worse schooling outcomes of children as long as the mother is absent. If the mother stays with the children at home and the father engages in labour migration, educational spending remains constant, the children s grades increase significantly, but school attendance is reduced. These are likely results of weaker bargaining power of the absent parent combined with stronger preferences for education among mothers. If mothers exercise decision-making, remittances can foster educational performance of students and social mobility. Key words: Internal migration, Parental absence, Human capital investment, Gender, Intra-household bargaining, Social Mobility, Migrant Selectivity, Fixed effects EKHM52 Master thesis, Second Year (15 credits ECTS) June 2017 Supervisor: Christopher Smith Examiner: Björn Eriksson Word Count: 15,690 Website www.ehl.lu.se

A Matter of Gendered Investment: Impacts of Internal Migration on Child Education in Indonesia Paul Berbée Master in Economic Development and Growth Lund University Contents 1 Introduction 1 2 Literature review 2 2.1 Theoretical background................................... 2 2.1.1 Migration theories.................................. 2 2.1.2 Temporary migration................................ 5 2.1.3 Remittances...................................... 6 2.1.4 Individual vs. family migration........................... 7 2.2 Internal migration in Indonesia............................... 8 2.3 The effects of migration on children left behind...................... 9 3 Data 12 3.1 The dataset.......................................... 12 3.2 Variables........................................... 13 3.2.1 Migration variables.................................. 13 3.2.2 Outcome variables.................................. 15 3.2.3 Other variables.................................... 17 3.3 Descriptive statistics..................................... 19 4 Methods 22 4.1 Methodological challenges.................................. 22 4.2 Empirical strategy...................................... 22 4.3 Variation within and between categories.......................... 24 4.4 Discussion in the context of the study........................... 25 5 Results 26 5.1 Educational expenditure................................... 26 5.2 School attendance...................................... 27 i

5.3 Final Exam grades...................................... 29 5.4 Differences between boys and girls............................. 30 5.5 Limitations.......................................... 31 6 Conclusions 32 A Bibliography 35 B Appendix 38 List of Tables 1 Overview of migration categories.............................. 14 2 Migration status per survey year: Observation numbers................. 38 3 Education outcomes per migration status......................... 38 4 Parental migration details by migration category..................... 39 5 Parental education by migration status........................... 40 6 Household details by migration status........................... 41 7 Between and within community and household variation................. 42 8 Whithin- and between- tabulation of migration status (community level)........ 42 9 Whithin- and between- tabulation of migration status (household level)........ 43 10 OLS: Educational spending................................. 44 11 Community Fixed effects: Educational spending..................... 45 12 OLS: Dependent variable: School attendance....................... 46 13 Community fixed effects: Dependent variable: School attendance............ 47 14 Household fixed effects: Dependent variable: School attendance............. 48 15 OLS: Dependent variable: Final exam score........................ 49 16 Community fixed effects: Final exam score......................... 50 17 Household fixed effects: Final exam score......................... 51 18 Between-Community Estimation.............................. 52 19 Between-Household Estimation............................... 53 20 Gender OLS.......................................... 54 21 Gender CFE......................................... 55 22 Gender HFE......................................... 56 Acknowledgement I would like to thank my supervisor Christopher Smith for his time and good advice and Gregory for his valuable help and useful comments. ii

1 Introduction An on-going academic debate is focused on the impact of migration on the original regions and households of migrants. A priori, it is not clear in which direction spouses and children that are left behind are affected by a parent s migration; Remittances provide increased material well-being and improved security, but the household may suffer from the absence of a parent or a partner. This study addresses the connections between internal migration, family structure, investments in education and school performance of children. It will not only evaluate the net impact of parental migration on schooling outcomes, but also try to improve the understanding of the underlying mechanisms that include remittances, gender-specific attitudes towards education, changes in intra-household bargaining power, and the emotional burden of parental absence. Particularly, I explore which household compositions and which migration patterns are associated with social mobility or under which circumstances and why they may prioritize educational success. Thereby it will be important to account for different motivations for migration decisions and self-selection into migration. I will look at the specific case of internal migration in Indonesia, one of the largest countries in the world that developed and urbanized quickly during the past decades. Many Indonesians move from rural areas towards the large cities, and large parts of the population are directly affected by internal migration because some family member has moved to another place. In order to address the aforementioned questions, I will use data from the Indonesian Family Life Survey (IFLS), a large on-going panel data set that contains village, household and individual level information from five waves in the period from 1993 to 2014. The IFLS tracks migrants throughout most provinces of the country and has remarkably low attrition rates so that various migration patterns and return migration can be soundly identified. In addition, household expenditures in education, which is a result of intra-household bargaining and a determinant of schooling outcomes, are directly observable. I will estimate the effects of migration on educational spending, school attendance as well as grade averages and employ community and household fixed effects estimates to account for unobserved heterogeneity between migrant groups and non-random self-selection into migration. Generally, the results of this study shed a pessimistic light on the consequences of internal migration on human capital investment and child schooling. I find strong indications that households in which the mother migrates for an extended period of time experience a drop in educational spending by about 30%, but only as long as the mother does not return. This may indicate that mothers attach a higher value to education than fathers and that their bargaining power concerning budget choices is weakened when they are away. Children whose mothers are absent are also up to 22 percentage points less likely to be attending school and achieve systematically lower grades than their peers. Considering the reduction in spending on education, it seems unlikely that this is only a consequence of emotional distress because of the mother s absence. Unlike other forms of migration, fathers labour 1

migration is found to have some positive effects on schooling: Those households with fathers that are labour migrants spend slighly more on education than non-migrant households and their children obtain significantly higher grades as migration takes place. This is consistent with previous research, which suggests that remittances from fathers make an important difference in schooling. Surprisingly, children whose fathers are labour migrants are found to be 7 percentage points less likely to be attending school in spite of higher grades. A possible explanation could be that paternal migration does not necessarily provide incentives to obtain above-average levels of schooling if it is perceived as an example and as an alternative means for achieving social mobility. In short, my results suggest that the absence of the mother is crucial for determining the consequences of migration on schooling because it reduces household investments in education. Monetary returns from migration are only found to be spent on education and result in social mobility if the father migrates and the mother stays home. After this introduction, I will review the relevant literature on migration decisions, the countryspecific context of internal migration in Indonesia and the effects of parental absence on children. The third section introduces the dataset and the sample as well as the most important variables before providing some descriptive statistics. In section four, methodological challenges will be discussed and the empirical strategy of this paper will be explained. Next, section five presents and interprets the results. Finally, some concluding remarks and comments concerning policy implications will be made. 2 Literature review To gain a fundamental understanding of the mechanisms that drive and influence migration in this study, I will first review theories on migration decisions and relate them to phenomena like return migration, remittances and the decision between individual and family migration. This is relevant because different migration patterns probably have very different consequences on both the migrants themselves, and their original households. After that, I would like to shed some light on the patterns of internal migration in Indonesia and the country specific context, before reviewing the scientific literature about the effects of family structure and parental absence on the education of children left behind. 2.1 Theoretical background 2.1.1 Migration theories In the neoclassical migration literature prior to the 1980s, migrants have been assumed to be individual rational agents that maximize their lifetime utility in a setting of perfectly working markets (Sjaastad, 1962; Schultz, 1961). Migration occurs if the expected benefits at the destination minus migration costs are higher than the expected benefits at the place of origin. Migration can be essentially understood as 2

an investment in human capital in order to increase earnings at a later point in life. In any case, positive returns from migration in the form of wage gaps, differences in employment rates or living standards are a necessary condition for migration in the neoclassical framework. Since returns to education are likely to differ across locations, migrants are assumed to self-select to destinations with different labour market characteristics according to their educational background (Massey, 2008, 20; Lucas, 1997, 735-743). Assuming that most seasonal jobs e.g. in agriculture and construction have relatively little returns to human capital, low-skilled rather than high-skilled workers are predicted to self-select into circular migration. This would explain observations by Deshingkar (2006) who describes circular migration patterns of poor rural workers in Indonesia. Empirical evidence, however, also suggests that the poorest parts of a population usually do not move because they are not able to afford migration costs. While international migration may not be financially feasible for people at the lower end of the income distribution, internal migration may be possible. If the household budget is relaxed as a consequence of returns from migration, this will likely result in higher household consumption and higher investments in the education of children. Since migration can be seen as a form of human capital, also outmigration of children following their parent is conceptually similar to educational investments. However, tables 5 and 6 suggest that in my sample, labour migrants are clearly positively selected when it comes to their education. Therefore, they might have stronger preferences for the schooling of their children regardless of their migration status. So migrant selectivity could be an explanation for a positive association between migration and educational investments. In short, neoclassical migration theories predict that migration to be selective in terms of education and to be associated with higher earnings and improving living standards. More recently, migration researchers have started to argue that many migration decisions are not taken by isolated individuals as assumed by neoclassical theories, but by larger groups such as families, households and communities. It has been suggested that migration cannot only be a means to maximize income but also a response to market failures. The new theoretical approach that evolved from these assumptions is referred to as New Economics of Migration and was inspired by Stark and Levhari (1982) and Stark and Bloom (1985) who point out that migration can be a strategy to collectively minimize risks and diversify income. Accordingly, rural communities that are affected by risks in agricultural production have strong incentives to obtain some kind insurance in case of natural disasters like floods or draughts. Remittances from migrants can be an alternative source of income and serve as insurance as long as they are not correlated with the income volatility at the place of origin (Lucas, 1997, 749). Similarly, circular and in particular seasonal migration can also be regarded as a form of income diversification on the individual level (Lucas, 1997, 729). If migration is a substitute for other forms of insurance, existing insurance mechanisms like social networks at home can hamper migration. Munshi and Rosenzweig (2016) for instance point out that family and castebased networks form significant migration barriers even in the presence of large wage rural-urban wage 3

gaps in India. If access to capital is restricted, savings from temporal migration can also serve as a substitute for formal credit (Rapoport and Docquier, 2006; Lucas, 1997, 748). In this case, risk-averse or poor families that have limited access to insurance or credit may be more likely to self-select into migration. On the other hand, positive selection can take place if more ambitious and entrepreneurial households decide to migrate. Improved access to capital and reductions in income volatility are likely to foster continuous investments in education of migrant households. However, since one can easily think of productive investments other than education and since investments can be made elsewhere in the group, the predicted association between migration and education might be weaker than in neoclassical frameworks. In summary, migration decisions are unlikely to be made purely on the individual level. Therefore, this study will account for heterogeneity between both communities and households. In addition, the split-up of households can be a response to market failures rather than to wage gaps between locations. This strategy may be most appealing for poorer people that work in agriculture and result in persisting connections and remittances flows between migrants and their original households. Empirical studies suggest that access to kinship and other networks at a destination foster the propensity to migrate because they provide valuable information and help upon arrival to gain a foothold in the new environment. In this way, migrant networks act as social capital, lower monetary and psychological migration costs and make migration feasible for individuals who would not have migrated otherwise (Massey, 2008, 42; Lucas, 1997, 743). As more and more migrations take place and networks at destinations grow, migration is likely to become less selective as the costs of migration for less motivated and less educated individuals decrease (Massey, 2008, 48 and 270). Kandel and Kao (2001) and Jampaklay (2006) argue that children from migrant families have better access to networks and relevant information about migration so that they face lower emigration barriers. Accordingly, their respective studies suggest that Mexican and Thai children with migrant parents are more likely to migrate themselves and that their willingness to invest in schooling depends on the returns from education at the expected destination. Summarizing, migrant networks are an important factor on the community level, since areas which many previous migrants presumably face lower migration costs. In addition to that, within a family parental migration can make subsequent migration of other family members more attractive and easier. Altogether, the presented migration theories will in the first place apply to labour migration, which only represents about one-third of all internal migration movements in Indonesia (Deb and Seck, 2009; Berbée, 2017). Nevertheless, the theories yield a couple predictions for the effects of migration on child schooling: According to neoclassical theories, monetary returns from migration will relax budget constraints and foster investments in education, although non-random self-selection into migration could strongly bias the results. New Economics of Migration predict that in cases when schooling is hampered by market failures or income volatility, migration can be a way to compensate 4

for these failures. Nevertheless, returns from migration might as well be invested for other purposes without affecting schooling. Migrant networks are likely to increase the cost of education relative to migration and decrease incentives for schooling while reducing migrant selectivity. Since different theories with different consequences might apply for different migrant groups, this study will try to differentiate thoroughly between a number of migration patterns and between labour and other reasons for migration. When focusing on effects within households and communities, it must be emphasized that the groups that engage in the different kinds of migration are likely to differ substantially because of self-selection and that comparisons across groups will be problematic. 2.1.2 Temporary migration Many internal movements in Indonesia are only temporary, i.e. migrants do not stay at their destination in the long run. In this study I will refer to migrants who move permanently back to their place of origin as return migrants. By contrast, circular migration is regularly repeated and includes seasonal workers. In the theoretical literature, reasons for return migration can be broadly divided into two (Lucas, 1997, 748): Firstly, migrants may return to their original household if their expectations and hopes have failed to materialize. In that case, returning has not been intended beforehand and return migrants are likely to be negatively selected from the pool of out-migrants (Borjas and Bratsberg, 1996; Gibson et al., 2013). This explanation can reconcile return migration with neoclassical migration theories and might e.g. apply to those with bad luck in a job lottery such as suggested by Harris and Todaro (1970). Also, labour market situations at destinations and origins changing over time may lead to a return of emigrants. Secondly, return migrants can be characterized as target savers who move because of lacking access to capital at their homes and who have planned to eventually return from the very first day. They come back to their homes as soon as a specific investment that they had in mind (including marriage, education of children, housing, land, or savings for retirement) can be realized. Rapoport and Docquier (2006) point out that in the case of target savers return migration and remittances are conceptually similar since it makes little theoretical difference whether savings are repatriated through remittances or through return migration. This second explanation is well in line with the New Economics of Migration that assumes migration to be a consequence of market failures. Successful migrants that experience high earnings are likely to meet their previously set goals sooner and return home earlier. In addition, circular migration can occur seasonally and can be part of a strategy to diversify income that might be wider spread in rural areas and among poor than among wealthier individuals in Indonesia, as it is suggested by Deshingkar (2006) and Deb and Seck (2009). By combining qualitative and quantitative evidence from rural Bangladesh, Kuhn (2000) illustrates that individual circular migration provides an effective solution to market failures and that remittances and circular migration 5

are a strong indication of financial and emotional connections between migrants and their original households. Summarizing, temporary migration can be a strategy to obtain capital for educational investments or other assets that improve the material well-being of households. Also emotional well-being will be improved through the return of a parent. At the same time, return migration may also be harmful to education if it is due to a migration failure and reduces wealth. 2.1.3 Remittances Even though in the past decades a vast literature about the impacts of remittances emerged, most publications focus on international remittance flows towards developing countries. Nevertheless, also internal remittances are substantial and an important source of income for the original households of migrants. As such they represent a potential source of educational investments. In the case of Indonesia, Lu (2010) estimates that as much as 85% of all recent rural-to-urban migrants remit money to their families living apart and that the amount of these monetary transfers on average accounts for almost 50% of the migrants incomes. Concerning the effect of remittances on education, Rapoport and Docquier (2006) underline that migration can have both forward- and backward-linkages across generations, depending on whether they are meant to be the repayment of an educational loan to previous generations or whether parents remittances relax budget constraints and foster investments in their children s education. Several empirical studies from Latin American countries indicate that school attendance is sensitive to shocks in remittance receipts and that remittances have stronger impacts on the age at dropping out of school than other forms of income (Edwards and Ureta, 2003; Hanson and Woodruff, 2003; Alcaraz et al., 2012). This has been interpreted as remittances relaxing credit and investment constraints. Using the same dataset as this study, Adams and Cuecuecha (2010) analyse remittances from overseas migrants in Indonesia and suggest that they are in the first place invested in food, physical assets and housing in particular. Although households receiving remittances spend more at the margin on education, this effect is not statistically significant. A parallel study by Nguyen and Purnamasari (2011) works with the same sample and focuses on the effects of remittances on children. Using instrumental variables to account for selection bias, the authors do not find significant impacts neither of migration of a family member nor of remittances on the school enrollment of children. Vanwey (2004) analyses gender differences in the remitting behaviour of internal migrants in Thailand. Her results suggest that women and migrants from poorer households tend to remit relatively altruistically whereas the men and richer migrants tend to remit in contractual ways. Since contractual motives refer to egoistical reasons such as loan repayment or maintaining inheritance rights, women s and poorer people s remittances might be more likely to foster education in left-behind households. In short, remitting money home is widespread among internal migrants in Indonesia. Although em- 6

pirical studies suggest that remittances foster educational investments and child schooling in different parts of the world, no similar evidence exists for Indonesia. 2.1.4 Individual vs. family migration Beside the mere decision whether to migrate or not and whether to move temporarily or permanently, the choice of whether to migrate with the whole family or individually is a third source of non-random migrant selectivity (Gibson et al., 2013). Throughout this study, I will define individual migrants as parents that move without their partner and children, and family migration as a relocation of the whole household, including both parents and their children. Family separation presumably yields high and persisting psychological costs for both the migrant and the original family that increase with the time of absence. Therefore, individual migration seems unlikely if no return is planned (Kuhn, 2000; Lu, 2010). This is the case in neoclassical migration theories that involve a maximization of lifetime earnings, and do not provide explanations for planned return migration. However, in some cases one family member might move first and get the others to join him or her at a later point in time when everything for family life is prepared. On the other hand, according to the New Economics of Migration, a family split-up can be purposeful if migration is adopted as a strategy to diversify income and reduce the vulnerability to external shock (Rosenzweig and Stark, 1989; Lucas, 1997, 752). Also if the migrant intends to return home again soon, e.g. if he or she moved because of a need for capital to start a business, it is likely that most household members stay home to guard the household assets and to save monetary migration costs Rapoport and Docquier (2006). The cost of separation may be higher if it is not an adult or child but the partner that moves away for a longer period. Kuhn (2000) evaluates the propensity to engage in family versus individual migration for Bangladeshi farmers from the Matlab region and the different impacts on the original villages. For that purpose he distinguishes between persons moving without their partner for more than 6 months and cases when both partners move together. Kuhn (2000) argues that family migration is preferred among households that lack productive assets and access to credit so that a simultaneous living at two different places is not feasible. Also the absence of extended family networks that can provide assistance in the care of the conjugal family makes family migration more likely. On the other hand, family migration is also widespread among highly-educated men in urban centres who are economically independent and do not need to rely on financial assistance from insurance networks. Individual migration instead is most common if migration is used as an access to capital for investments at home. Kuhn (2000) emphasizes that in a constellation of a household splitting up, cooperation between the migrant and the original household is usually very close and beneficial for both parties. Even if individual migration may financially pay off, the separation of the family and the absence of a parent will have most probably negative psychological consequences for children that 7

may affect their educational success (see section 2.3). 2.2 Internal migration in Indonesia In terms of population size, Indonesia is the fourth-largest country in the world. It is an archipelago of many islands and ethnically very heterogeneous. During the past three decades, the country urbanized quickly and enjoyed rapid economic growth that came along with improvements in many human development indicators such as income, education and health. Although both the share of the labour force that is employed in agriculture and the share of the rural population have been almost reduced by half since 1980, still one-third of Indonesians earn their livelihood as farmers and 45% live in the countryside (World Bank, 2017). Internal migration in Indonesia is, however, far from a recent phenomenon. In the course of the 20th century, large numbers of people moved for very different motivations, including economic reasons, conflicts and state-led resettlement policies, to other places within the country (van Lottum and Marks, 2012). Along with the economical upsurge in the 1980s, inter-provincial migration increased strongly until the day of today. Census data suggest that a substantial share of the population (about 10%) recently moved to another place and about half of these migrants are women (Deshingkar, 2006). According to Lu (2010) and Deb and Seck (2009) the metropolitan regions of Jakarta, Surabaya and Makassar where new industries have emerged are the most important migrant destinations. Deb and Seck (2009) estimate that in 2000 about half of all Indonesian households had at least one family member with some internal migration experience since 1993 that lasted longer than six month. However, almost 40% of these migrants, with this share being much higher in rural areas, had eventually returned home again. These numbers on return migration are not comparable to those shown e.g. in table 4 since this study only refers to parents with young children at home. Deb and Seck (2009) find that households with internal migrants in Indonesia are generally positively selected in terms of income, education and health. However, this holds only for migration that was persisting until the date of the interview: Higher income and educational levels do not significantly change the probability of having a return migrant in the household. This can be interpreted as an indication that migrants from higher educated households are more likely to move permanently or at least for longer periods compared to members of household with lower education levels. Generally, Deb and Seck (2009) come to the conclusion that people who live in rural communities and who hold productive assets like land are less likely to migrate. In case they migrate, they are found to be more likely to return. Deshingkar (2006) guesses that the phenomenon of negatively-selected return migrants can to some extent be explained by rural labourers from peripheral provinces leaving their homes because they are attracted by informal jobs in construction, labour-intense manufacturing and services in both rural and urban centres. For these workers, pursuing multiple income-earning activities that combine farm and non-farm work through circular migration is reported to be a com- 8

mon practice. In addition, it is claimed that temporary migration of unmarried women that work in export sector factories in east Java to financially support their original families has become common and socially accepted. The observation that permanent migrants are positively- and return migrants negatively-selected could be explained by different motivations: Well-educated people might move permanently in order to increase their long-term income perspectives in cities where they experience high returns to their education, as it is predicted by neoclassical theories and observed in Indonesia by Massey (2008, 181). These people presumably do not depend on social networks, prefer to move with their whole family and do not have incentives to return Kuhn (2000). For poorer Indonesians in rural areas, repeated and circular migration may be perceived as strategies to diversify their incomes as suggested by the New Economics of Migration. In the expectation to return anyway, they have incentives to move individually, to maintain strong a relationship with their original households and to send remittances home. Lu (2010) estimates that as much as 85% of all recent rural-to-urban migrants remit money to their families and that these monetary transfers on average amount to almost 50% of the migrants incomes. Evidently, the consequences of both described migration patterns are likely to have different consequences for family relations, income and child education. A study by Lu (2008) supports the finding that migrants are positively-selected in the first place and experience significant income gains, even though this holds only in the case of young labour migrants. No selectivity and no positive returns from migration can be observed among people that move for family reasons. These findings can be explained by neoclassical migration theory that predicts both migrant selectivity and monetary returns from migration, but only applies to labour migration. Because of the positive effects on income, it is straightforward to assume that labour migration has different effects on child education than other forms of migration. Against that background, it is important to keep in mind that only about one third of all recent internal migrant movements are estimated to be made for work-related reasons (Berbée, 2017). The assumption that neoclassical migration theories fail to explain overall internal migration streams in Indonesia at a macro-level is supported by the results of van Lottum and Marks (2012) who suggest that income disparities within the country have not been significant for explaining past migration movements. In short, internal migration in Indonesia in the first place occurs towards urban centres and is common for both men and women. Educated and relatively wealthy Indonesians tend to self-select in permanent migration, whereas temporary individual migration is more common among less educated people from the countryside with lower incomes. In addition, motivations for internal migration are found to be heterogeneous and only a minority of migrants to move because of work. 2.3 The effects of migration on children left behind As Antman (2013) points out, the net effects of parental migration on children left behind are a priori unclear because several counteracting mechanisms are at work. On the one hand households with a 9

migrant are likely to receive remittances and to be economically better off, which relaxes their budget constraints and allows for higher educational and health-related expenses and leads to a decrease in demand for child labour. On the other hand, the absence of parents can have negative effects on child development. Lacking a parent is likely to be a psychological burden and result in less family members engaging in childcare. Additionaolly, children may need to take over domestic chores or other labour, which can them out of school. At a closer look, the effects of parental absence on the education of children left behind depend on several different factors: First, the duration of the absence of a parent matters. The underlying mechanisms are discussed in sections 2.1.1 and 2.1.2 about migration theories and temporary migration: Depending on the reason for return migration, it affects child schooling in different ways. For instance, Jampaklay (2006) uses longitudinal data from Thailand to argue that especially long-term absence (longer than 2 years) of a parent has more persistent negative effects on the educational chances of children left behind than shorter trips. Also Kandel and Kao (2001) suggest that higher numbers of working trips to the US of Mexican fathers reduce the schooling years completed by their children. However, they find slightly positive effects of repeated migration on grades and attribute this to improved economic well-being through remittances. The finding that material wealth may be more important for explaining grades than schooling years is also supported by Asis and Ruiz-Marave (2013) who use a sample of primary school children from the Philippines and generally find positive effects of migration on school pacing and achievements (measured by grades). However, their results indicate that the variables that affect both outcomes slightly differ: School pacing is strongly negatively influenced by variables that can cause distress or poor academic support like poverty, rural residence, low education of the mother and psychological difficulties of the child. In contrast, medium and high household wealth seems only to have an independent statistically significant effect on school achievements. Assuming that long-term migration of a parent is more likely to improve family income, it will positively influence educational spending and school performance, but only as long as psychological well-being and good care is ensured. For this purpose and for educational success, constellations with a migrant father and the mother staying with the children are found to be most beneficial. Second, it makes a difference for children s education, whether the father or the mother migrates. Especially younger children are likely to be emotionally stronger attached to their mothers who are usually more involved in child care than fathers so that their role in the family cannot be easily replaced. As a consequence, the costs of maternal migration ar probably higher from the child s perspective regardless of possible benefits from migration. Higher emotional costs of maternal absence can explain results by Cortes (2015) and Kandel and Kao (2001) who find that differently to the migration of mothers, the migration of the fathers has positive effects on the school pacing of Philippine children, even when controlling for differences in remittances. However, results by Battistella and Conaco (1998) suggest that the presence of an extended family (that is aunts, uncles, grandparents or cousins) in 10

the household reduces the negative effects of an absent parent and can at least partly substitute the mother s or father s role in child care and emotional well-being. Nevertheless, as long as the mother is present in the household, also remittances matter as an explanation for school pacing, as a study by Kuhn (2006) suggests. He observes that only international migration of fathers in Bangladesh has significant positive impacts on child schooling whereas internal migration is statistically insignificant. He attributes this to differences in remittances. The same seems to hold in Mexico (Antman, 2012): Whereas domestic migration of fathers has no significant impact on educational attainment of children, children whose fathers works in the US stay significantly longer in school, suggesting that the paternal absence per se is relatively unimportant for child education, but remitted dollars make a difference. Jampaklay (2006) points out that the migration of fathers increases not only the propensity of children to stay in school, but also the propensity to out-migrate themselves. Both can be interpreted as ways to achieve social mobility and could be an indication that migrant fathers are not only breadwinners, but also examples. If the mother migrates instead, children are more likely to neither move nor achieve higher education. The observation that the absence of mothers has worse consequences on child education than the absence of fathers might also partly be due to changes in the intra-household bargaining power. Nguyen and Purnamasari (2011) and Antman (2012) point out that the absence of one parent may impact schooling outcomes if parents have different preferences over the allocation of resources devoted to the education of their children. Usually it is assumed that mothers have stronger preferences in favour of education than fathers. Strong detrimental effects of maternal absence (Jampaklay, 2006; Cortes, 2015) could at least partly be interpreted as mothers losing control over schooling and investment decisions. Unfortunately it remains unclear in how far unobserved differences between households whose mother or father migrates bias the results of these studies. Third, the effects of parental migration depend on child characteristics such as gender and age. Fathers and mothers might not only have different preferences concerning education in general but also whether to allocate resources to boys or to girls. The results of Antman (2015) suggest that while a male household head is away, the share of resources spent on girls compared to boys increases, which is attributed to a greater decision-making power of the spouse. Once the migrant returns, boys are found to be favoured again instead. With a different data set, Antman (2012) suggest that remittances from fathers are more beneficial for girls that for boys education, which may be because of the same reason. Apart from that it has been suggested that boys are emotionally less stable and struggle more to cope with parental absence. Arguillas and Williams (2010) for instance finds that boys school attendance is more strongly affected by many familiar background variables, including also the parents international migration, than girls. For the case of Indonesia, Nguyen and Purnamasari (2011) is the only study of my knowledge that evaluates the effects of parental migration on children left behind. However, it entirely focuses on international migrants and does not find significant net impacts of remittances on school enrollment 11

of children, although the direction of the effects is opposite between male and female migrants. 3 Data 3.1 The dataset The Indonesian Family Life Survey (IFLS) is an on-going longitudinal survey that is carried out by the RAND Corporation together with several Indonesian and American partner universities that provide support in the review of methodology and survey design (Strauss et al., 2016). The data collection took place in five waves that were fielded in 1993/1994, 1997, 2000, 2007/2008 and 2014/2015. The initial 1993 sample contains roughly 7,200 households with more than 30,000 individuals. It was drawn from 13 out of 27 Indonesian provinces, but it is representative of about 83% of the Indonesian population at that time. The IFLS aims at providing a broad range of information about the socio-economic, demographic and health conditions of the Indonesian population. In addition to questionnaires for individuals and household heads, the IFLS also provides a large number of community and township details such as community organization as well as infrastructure such as transportation facilities, schools and access to health care. The IFLS tracks all respondents who migrate, join other households or establish new ones. Major efforts to re-contact all respondents result in a remarkable success rate of over 90% (Strauss et al., 2016; Deb and Seck, 2009). Migrants are tracked on their movements within the IFLS provinces and complete information is collected about all members of existing or newly established households they are living in, regardless of whether these members have been interviewed before or not. Because the original households split up, individuals from the sample join other households and children are born, continuously new households and new individuals enter the sample. For that reason the number of households interviewed in 2014 has roughly doubled compared to 1993 and is about 15,000 with 64,000 individuals. Information about migration movements and the presence of temporary migrants in their original household is twofold: Firstly, it can be observed whether a previous household member is absent at the time of an interview. If he or she has left the household, the household head is asked about the reasons, timing and the destination of migration. Secondly, respondents that are 15 years or older are retrospectively interviewed about their own migration history and whether they moved temporarily to another place since the previous interview. In this case migration is defined as living outside the locality of residence during more than six months. Again, detailed information about the reason, the duration and the places of each migration movement are provided. That way, information about the living conditions and activities of both, the migrant and the original household is available and not only on-going but also temporary and circular migration can be identified. For the purpose of this study, I combine information on migration from the household roster that is based on answers 12

of the household head and the questionnaires of the migrants themselves. In case of contradictions, I prioritized the information provided by the migrants. I restrict the sample to individuals between ages 6 and 22 at the time of the interview, because I am only interested in outcomes for young people of schooling age. In addition, I exclude persons whose parents have died because the psychological consequences of parental death are probably very different from migration or other reasons for absence. In practice, I will only look at individuals who can be assigned to any of the migration categories that are defined in section 3.2.1. In practice this means that I drop observations if either not sufficient information about parents is available or if the migration patterns of the household do not fit in any of the categories. The former may be the case if the father of a child is unknown or if a parent lives permanently separated from his or her children. The latter occurs if several different migration movements take place in a household, e.g. both the household and an individual parent move or both parents eventually move away for some time. By restricting my sample to individuals with a valid migration status, I drop 18,240 of the remaining 41,824 observations from the 5 waves of the survey so that I end up with 23,584 observations from 6,663 different households in 1,074 different communities. This may bias the remaining sample, but has the advantage of precisely defined and clearly distinguishable treatment variables. An overview about the remaining sample sizes by migration status and observation year can be found in table 2. Table 2 about here. 3.2 Variables 3.2.1 Migration variables In the IFLS, migration is defined as moving to another place outside the location of residence and living there longer than six months. For this study I define seven different migration categories based on the migration experiences of the respondent s biological parents. Table 1 provides a formalized overview. The categories are mutually exclusive, so that one observation can only adopt one single migration status at a time. First, I define individuals with no migration experience as those whose parents have been permanently living in the same household and never engaged in migration since the child was born. This also implies that the entire household including the child has never moved to another locality. Because parental absence and any own migration can be ruled out, this group will serve as the main comparison group in this study. It will be labelled as category 1 in table 1 and other tables. Second, I identify students whose parents are individual migrants. This means that one parent moves to another place, but the original household, including the child and the other parent, stays at its original location. Thus, individual migration implies that a parent has been absent from the child s household for at least 6 months. Since the consequences of paternal and maternal absence may 13

Table 1: Overview of migration categories 1. No migration experience Household remains at original location; Neither of them has ever migrated; Both parents are present 2. Labour migration of father Household remains at original location; Father migrates for work; Remaining family stays at original location; Mother is present and has never migrated 3. Other migration of father Household remains at original location; Father migrates for other reasons than work; Remaining family stays at original location; Mother is present and has never migrated 4. Labour migration of mother Household remains at original location; Mother migrates for work; Remaining family stays at original location; Father is present and has never migrated 5. Other migration of mother Household remains at original location; Mother migrates for other reasons than work; Remaining family stays at original location; Father is present and has never migrated 6. Labour family migration The entire household moves outside the original location; Either mother or father migrates for work; Both parents are present 7. Other family migration The entire household moves outside the original location; Neither mother nor father migrate for work; Both parents are present differ, I distinguish between a mother s or a father s individual migration. There are many reasons for parents living separated from their original household that are not related to labour migration and to which traditional migration theories do not apply, such as divorce or family duties. According to the literature, the effects of labour migration differ from migration for other reasons and is for instance associated with higher remittances. For that reason, I also distinguish between individual migration that takes place because of work-related and because of other reasons. Categories 2-5 refer to this type of migration. Third, I identify family migration. In this case the whole household moves to another place outside its original community including the child and both parents. Again, I distinguish between work-related and other reasons. By doing so, I make sure that e.g. families who just relocate their house to a new area are not wrongly assumed to be labour migrants. Family migrants might be an interesting comparison group because they presumably enjoy positive effects of migration, but do not experience parental absence. They will be labelled as categories 6 and 7. Apart from the different migration categories, dummy variables for return migration are added to the dataset. These take the value one if a parent is labelled as an individual migrant, but is currently living with the original household again. 14