Do Remittances Compensate for the Negative Impact of Migration on Children s Schooling?

Transcription

1 Do Remittances Compensate for the Negative Impact of Migration on Children s Schooling? Jozefina Kalaj November 19, 2018 Abstract This paper examines the direct impact of remittances on the school attendance of children ages Drawing data from Albania, the main objective is to understand the schooling choices of migrant households when faced with an abrupt shock to remittance inflows. The proposed methodology consists of two parts. First, to avoid potential bias from selection into migration, the analysis is restricted to households with at least one migrant abroad. Second, identification relies on the unexpected onset of the 2008 global financial crisis and the heterogeneous unemployment shocks which ensued across countries where Albanian migrants were living at the time. Formally, the change in the unemployment rate from 2007 to 2008, which varied markedly between destination countries, is used as an instrument for receipt of remittances. Results indicate that households with migrants in destinations where the unemployment rate increased were less likely to receive remittances and in turn less likely to have children attending school. Specifically, the likelihood of attendance increases by 5-6 percentage points when the probability of remittance transfers increases by 0.1. Results are robust to various specifications. JEL Classification: Keywords: PhD Candidate, Department of Economics, George Washington University 1

2 1 Introduction The number of individuals migrating internationally has increased in recent years. According to the World Bank, from 2000 to 2013 the number of international migrants rose by about 72 million, from 175 to 247 million (World Bank, 2016). This movement is associated with an increase in financial transfers from migrants to those left behind. It is estimated that remittances exceeded $600 billion in 2015, of which roughly 70 percent went to developing countries an amount three times that of development aid (World Bank, 2016). What impact do these transfers have on the family members remaining in the country of origin? Are remittance inflows used for investment in physical and human capital or only used for consumption? These questions have vast social and economic implications for developing countries experiencing out-migration. Although remittances have long been recognized as a potential development mechanism, the more recent availability of individual-level data on migration and remittance receipts has brought new attention to this area of research. This paper adds to the literature by examining the effect of remittances on schooling focusing particularly on secondary school age children and using Albania as a case study. The primary objective is to understand the schooling choices of migrant households when faced with an unexpected shock to remittance inflows. The proposed methodology consists of two parts. First, the main analysis is confined to migrant households a choice made to avoid the potential bias resulting from selection into migration. Second, identification relies on the unexpected onset of the 2008 global financial crisis and the heterogeneous unemployment shocks which ensued across countries where Albanian migrants were living at the time. Specifically, the change in the unemployment rate from 2007 and 2008, which varied markedly between destination countries, is used to predict the probability of remittance inflows at the household level, and in turn estimate the impact of remittances on school attendance. Assessing the impact of remittances on schooling is complicated by the effect of migration. Generally, transfers from abroad are associated with the absence of family members. 1 The outmigration of adults creates a disruption in the household, with psychological and economic costs. In the short run, during the migrant s adjustment period, those left behind have to cover the costs 1 With the exception of those who receive transfers from friends or relatives who are not part of the household. 2

3 associated with the investment. Besides explicit migration costs, the household also experiences a shortage in labor resources, increasing the opportunity cost of time for all members, including children. As a result, children may join or increase time in the labor market, undertake more responsibilities in the family business, or complete additional chores (Hanson & Woodruff, 2003). Furthermore, if the migrant is a role model within the family, such as parent or older sibling, their absence may negatively impact the development of their young family members left behind. Aspirations to follow in the migrant s footsteps, whether stemming from the child or encouraged by the adult migrant, also impact educational outcomes. If the returns to education obtained in the home country are low at destination, the child may be less inclined to remain in school or perform at potential. Chiquiar and Hanson (2005) study the selection of Mexican migrants to the United States and find that returns to education for residents of Mexico are higher than for Mexican immigrants. 2 Legal migration agreements between sending and receiving countries also matter. Limited opportunities for legal migration put formal jobs out of reach, further diminishing incentives to invest in education at home (McKenzie & Rapoport, 2011). Lastly, international migration is daunting and risky especially when going through illegal channels. Having a family member abroad, who is able to provide information and support, mitigates the likelihood of migration failure, making migration an attractive upward mobility alternative to investment in human capital at origin (Kandel & Kao, 2001). Theory suggests that financial gains from remittances dampen the adverse migration effects on children s schooling, as income from abroad relieves liquidity constraints and increases the reservation wage of children. Thus, potentially, child labor decreases and school enrollment increases as the opportunity cost of children s time falls and parents are able to finance education related expenses. 3 Kandel and Kao (2001) remark that higher incomes improve the well-being of children in other dimensions as well, such as health and living conditions, which indirectly affect academic performance in a positive way. On the other hand, Alcaraz, Chiquiar, and Salcedo (2012) note circumstances when the inflow of remittances may actually reduce children s schooling, such as when remittance receipts are used to invest in family enterprises that children are engaged in. 2 While this suggests that the wage premium from migration to the U.S. is greater for the low skilled, Chiquiar and Hanson (2005) show evidence of intermediate selection for Mexican-born men. They attribute their findings to heterogeneity in migration costs across skill levels. 3 See Hanson and Woodruff (2003) for a detailed empirical framework. 3

4 Apart from the confounded impacts, empirical work is also complicated by the endogeneity of migration and remittance decisions a consequence of unobservable factors which affect migration, remittances and schooling (Alcaraz et al., 2012). For instance, a negative income shock, such as crop failure or job loss, can simultaneously induce out-migration of household members and force children to prematurely leave school for work (Hanson & Woodruff, 2003). Attributes such as ability and ambition shared by household members through genetics and common experiences can also affect both migration behavior and children s schooling (Antman, 2011). Higher ability individuals can identify better migration avenues and be more productive once abroad, increasing the likelihood of remittance transfers. Similarly, their children or siblings in the home country, also of higher ability, may be more likely to succeed academically regardless of remittance inflows. In this particular case, ignoring the potential endogeneity leads to overestimation of the impact of remittance receipts on children s schooling. Reverse causality is also a concern. Migration may be partially determined by children s success in school prompting parents, or family members, to seek job opportunities abroad as a way to finance the academic progression of children at home. Another possible scenario is remittances being sent to the household to reward school attendance and performance. As these examples suggest, the decision to migrate and remit is not random, implying that migrant and non-migrant households, as well as remittance and non-remittance ones, differ from one another, 4 and the inability to control for these differences leads to biased estimates. Previous studies have dealt with these concerns in a number of different ways. In terms of identification, a variety of instrumental variables have been exploited. Meanwhile, separation of the competing effects of remittances and migration has proven more difficult empirically. Given the potential endogeneity of both variables of interest, studies commonly focus on one of the variables and aim to capture net effects. When looking at the overall impact of migration, the treatment group consists of migrant households (both remittance-receiving and non-remittancereceiving) while the counterfactual group consists of non-migrant ones. Origin country characteristics are often used as instruments for migration, such as historical migration rates, distance to destination, and weather patterns. Conclusions are not unanimous, with some studies finding pos- 4 The new economics of labor migration (NELM) argues that migration is a household decision, a mutually beneficial agreement between the migrant and those remaining behind (see Stark & Bloom, 1985). 4

5 itive net migration effects on children s schooling, whereas others negative or insignificant results. In the case of Mexico for example, Hanson and Woodruff (2003) determine that year old girls experience positive net effects in terms of years of accumulated schooling, with results for boys being less conclusive. The authors attribute their findings to the positive effect of remittances, evidenced by the fact that results hold only for households with binding liquidity constraints. On the other hand, McKenzie and Rapoport (2011) find negative effects on completed schooling for year old boys and all year olds. 5 The negative effect for boys is credited to their own migration aspirations; 6 while the observed shift from school to housework for girls is suggested to be a consequence of labor shortages at home following the out-migration of family members. Lastly, Antman (2012) concludes that parental migration from Mexico to the U.S. benefits the educational attainment of girls only. Approaching the problem from the remittance receipt angle is more complicated. Data from various countries show that a substantial fraction of migrant households do not receive remittances. For example, summary statistics from this study indicate that this figure is 40%, while Hanson and Woodruff (2003) note 62%, and Acosta (2011) 22%. Children belonging to these households face the negative consequences of migration without the economic compensation of remittances; hence, not only are they disadvantaged compared to children in remittance-receiving households but even more so relative to children in non-migrant ones (Lu & Treiman, 2007). Grouping migrant nonremittance-receiving households with non-migrant ones can lead to biased results. Consequently, the composition of the counterfactual group needs careful attention when assessing the impact of remittances. Edwards and Ureta (2003), one of the first studies in this area of research, show that remittance receipts positively affect the school retention of 6-24 year olds in El Salvador. The study relies on the assumption that remittances are randomly assigned across households, implying, in other words, equal probability of remittance inflows for both migrant and non-migrant households. Acknowledging the improbability of this assumption, subsequent works have relied on instrumental variables estimation. Acosta (2011), for example, uses migration networks to instrument remittance 5 The study uses the 1997 Encuestra Nacional de la Dinamica Demografica data set, different from the 2000 Mexico Census of Population and Housing used by Hanson and Woodruff (2003). 6 Kandel and Kao (2011) also suggest that this effect dominates the various consequences of migration. They find that while children in migrant households have the necessary resources needed to perform well in school, shown by their superior academic performance, they express less interest in higher education. 5

6 receipts, and contrary to Edwards and Ureta (2003) he finds no evidence that the remittances hypothesis holds for year olds in El Salvador. Using a similar approach, Borraz (2005) estimates the impact of remittance inflows on accumulated schooling in Mexico by using historical state migration rates and distance to the U.S. border as instruments. Results suggest that remittances have a positive impact on the schooling of year olds who live in relatively small rural areas and whose mothers have less than three years of education. Borraz (2005) proposes the association between migrant networks and remittance receipts is better job opportunities in the United States for migrants with links to compatriots which results in higher transfers. While the argument is plausible, I reason that migration is the primary link between social networks abroad and remittances. Households in states that historically experienced greater emigration to the U.S. are more likely to be migrant households, and in turn, more likely to receive remittances. Correlation between the instrument and likelihood of migration raises doubts that Acosta (2011) and Borraz (2005) properly separate and assess the impact of remittances. Selection of instrumental variables is another important consideration. Variables originating from the home country are more likely to be correlated with the decision to migrate, making it difficult to capture the direct impact of remittances. To account for potential links between the instrument and migration decisions, Lopez-Cordova (2005) proposes the inclusion of controls that proxy current and historical migration flows. The study uses aggregate level data for Mexico, and instruments the fraction of households receiving remittances at the municipal level with historical rainfall patterns, arguing that areas with more concentrated annual rainfall receive more remittances as a way to smooth consumption. Results suggest no impact of remittances on the likelihood of school attendance for 6-14 year olds, and adverse effects for older children. Specifically, a one percentage point increase in the fraction of households receiving remittances at the municipal level reduces the proportion of year olds attending school by seven percentage points. Lack of clear consensus is not uncommon in this literature. The aforementioned empirical works illustrate the difficulty of determining the causal impact of remittances, both due to paucity of data and suitable instrumental variables. Yang (2008) notes that the ideal experiment would focus only on households with at least one migrant abroad. Moreover, to create an exogenous variation in remittances, migrants should be assigned a random heterogeneous economic shock. The author simulates this experiment by exploiting exchange rate fluctuations during the 1997 Asian financial 6

7 crisis. He finds that children belonging to households that experienced more favorable shocks were more likely to be in school and less likely to be working. Other studies, which effectively isolate the impact of remittances, corroborate the conclusions of Yang (2008). 7 In addition to proper counterfactuals, a commonality among these studies is the use of destination country economic conditions as instruments for receipt of remittances. For example, Amuedo-Dorantes and Pozo (2010) exploit the fraction of non-migrant remittance-receiving households and instrument receipt of remittances with average earnings for personal care service workers and the unemployment rate in locales where Dominicans reside in the United States. According to the authors, the virtue of this strategy is the ability to estimate the sole impact of remittances on schooling since these households experience remittance inflows without the parting effects of migration. They find a positive effect of remittances on school attendance, particularly for higher birth order children, females, and secondary school age children all subgroups with higher risk of underinvesting in human capital. When they reestimate the model using the full sample of children, i.e., adding children from migrant households, the positive effect disappears suggesting that other migration related consequences offset the positive impact of remittances. This paper follows in the same vein as the works noted in the previous paragraph. Data for the analysis are drawn from the Albanian Demographic and Health Survey (ADHS). A key feature of the survey design is that it permits the separation of households into three categories: non-migrant, migrant remittance-receiving, and migrant non-remittance-receiving. Furthermore, the migration module collected data on all household members abroad, including their specific location. These aspects, combined with the timing of the survey, provide suitable conditions for evaluating the impact of remittances on schooling of children. Albania is an interesting case study as it has experienced extensive out-migration since the fall of communism in 1990 a phenomenon which has played an integral part in the livelihood of many households. When the survey was administered at the end of 2008 and beginning of 2009, over one-third of Albanian households had a migrant abroad. Coincidentally, the year before the survey marks the start of the 2008 global economic downturn. Statistics show that the impact of the crisis on the labor market varied significantly across 7 See Alcaraz et al. (2012) for Mexico, Amuedo-Dorantes, Georges, and Pozo (2010) for Haiti, and Amuedo- Dorantes and Pozo (2010) for the Dominican Republic. 7

8 countries, including those where Albanian migrants were at the time, such as Greece, Italy, United Kingdom, United States, Germany or elsewhere in Europe. For example, between 2007 and 2008, the unemployment rate in Germany and Greece decreased by 1.1 and 0.64 percentage points, respectively. Italy on the other hand experienced an increase of 0.65 percentage points, and the United States was faced with a increase of 1.2 percentage points, marking a rise in the unemployment rate from 4.6% to 5.8%, over the course of a year. Important for the analysis is the fact that fluctuations in the unemployment rate were similar, both in magnitude and direction, across these host countries in the years leading to the crisis (see Figure 4 for details). This study takes advantage of these unexpected heterogeneous labor market shocks that ensued immediately after the start of the crisis. Using instrumental variables (IV) estimation, the change in the unemployment rate from 2007 to 2008 is used as an instrument for the receipt of remittances. The validity of this approach rests on the ability of the instrument to capture economic shocks experienced by migrants abroad, and the correlation of these shocks with receipt of remittances. Furthermore, the instrument should affect the school attendance of children (the outcome variable) only through its association with remittances. While details are discussed later in the paper, I find that the instrument is valid and results are robust to various regression specifications. The primary focus of the study is to measure the direct short-term impact of receipt of remittances on school attendance for children ages a group of particular concern in many developing countries, including Albania. 8 While years of schooling is also an important outcome, it is not suitable for this analysis, given the type of exogenous shock utilized and the cross-sectional nature of the data. Completed years of schooling is inherently a stock variable, one which encompasses a sequence of decisions made in the past decisions which are not observed. The aim, therefore, is to determine if two children, from separate migrant households, with similar observable characteristics aside from the unemployment rate shock experienced by their family members abroad differ post-shock in terms of school attendance. In summary, results show that households with migrants who faced unemployment rate rises were less likely to receive remittances and less likely to have children attending school. Findings also suggest that households in Albania have limited access to credit, since they compensate for the loss in remittance income by taking children out of school. 8 Many developing countries experience secondary school enrollment gaps. 8

9 The main contribution of this study is cleanly identifying the impact of remittances on schooling by eliminating the selection bias due to migration and exploiting a novel exogenous instrument for receipt of remittances. Furthermore, while closely related to Yang (2008) and Amuedo-Dorantes and Pozo (2010), it advances the work of both studies. As noted above, Amuedo-Dorantes and Pozo (2010) use the sample of non-migrant households to examine the remittance hypothesis. The small number of migrant households in the dataset, 140 to be precise, precluded the authors from estimating the impact of remittances using the sample of migrant households. Instrumenting with host country economic conditions becomes difficult when dealing with non-migrant households. In particular, what value does the instrument take for households with no migration history? The authors use Puerto Rico s economic indicators. One can argue, however, that this choice is arbitrary. This study does not face these concerns since all migrant households have at least one member abroad, which enables the direct link between the migrant and the economic conditions at destination. Lastly, while Yang (2008) reports reduced form results only, this paper assesses the impact of remittances using instrumental variables (IV) statistical methods. The paper is organized as follows: Section 2 provides some context on migration, remittances and education in Albania, while Section 3 describes the data used in the analysis. The empirical approach and results are covered in Section 4 and 5, respectively. And lastly, Section 6 concludes with a discussion. 2 Background: Albania 2.1 Migration and remittances Under the communist regime ( ), Albania was one of the most isolated countries in the world. Travel to and from was prohibited, and severe punishment, even by death, faced those who were caught absconding. As for the few who managed to escape, their family members suffered the consequences. In addition to international migration, relocation domestically, in particular between rural and urban areas, was also controlled. These restrictions on mobility for over 40 years, coupled with dire economic conditions, resulted in massive out-migration immediately following the collapse of the regime. External migration continued in the early 1990s, used as a strategy to insure against the eco- 9

10 nomic uncertainty brought on by the transition to a market economy. This first wave of migrants had a significant impact on the well-being of those left behind. Figure 1 shows remittance inflows and their fraction of GDP over the years. In 1992, Albania s GDP was $709 million and migrant transfers were $152 million 9 roughly 21% of GDP. In addition to their impact on consumption, remittances became a source of capital. Financial sector reforms could not keep up with the increasing demand for credit, resulting in the development of an informal credit market funded by remittance inflows (Jarvis, 2000). Alongside this harmless and essential market, pyramid schemes started to form taking advantage of three things: weak institutions, Albanians confidence in the newly created government, and their inexperience with financial markets (Jarvis, 2000). Two thirds of Albanians lost their investments when the pyramid schemes collapsed in It is estimated that the liabilities of the fraudulent deposit-taking firms were about half of GDP in nominal terms (Jarvis, 2000). This event destabilized Albania, violence erupted, the government was overthrown, and citizens became increasingly uncertain about the country s future economic development. Extensive out-migration ensued; the second wave surpassed the first one, with some estimates suggesting a twofold increase in migration (Azzarri & Carletto, 2009). Migrant networks abroad facilitated this outflow which, according to Azzarri and Carletto (2009), reached a peak in 2000 and persisted at a substantial rate for years. By 2010, the stock of Albanian emigrants had grown to 1,438, percent of the population of Albania at the time (World Bank, 2011). 10 Remittance receipts increased as well, reaching a peak in 2007/2008 (see Figure 1). Their importance decreased over time as the Albanian economy grew although they still account for a large percentage of GDP roughly 9% in Certainly the events of 1990 and 1997 served as catalysts for out-migration, but the sustained flows are a consequence of economic disparity between Albania and its neighboring countries, in particular Italy and Greece. From 1991 to 2017, Albania s GDP per capita 11 was, on average, 21% the size of Italy s GDP and 29% the size of Greece s. In addition, the unemployment rate was significantly higher in Albania (see World Bank Development Indicators). Considering these differences and the geographical proximity, it is no surprise that these two countries have the largest stock of Albanian emigrants. Bilateral migration data from 2010 indicate that 47% of emigrants 9 Figures in nominal terms. 10 Population of Albania was 3.2 million in 2009 (World Bank, 2011). 11 PPP adjusted, in constant 2011 international dollars. 10

11 reside in Greece and 36% in Italy. 12 The impact of international migration is far-reaching in Albania, with an estimated one in three households having a migrant abroad (Azzarri & Carletto, 2009; World Bank, 2007). Consequently, migration effects will continue to be broad stressing the need for research to help guide policy. A number of studies have analyzed the impact of migration and remittances on various outcomes for those left behind in Albania. Poverty reduction over the years has been attributed, in part, to remittance inflows as evidenced by the largest decrease in poverty occurring in areas which experienced the largest increase in the fraction of households receiving remittances (World Bank, 2007). McCarthy et al. (2009) and Miluka et al. (2007) examine the impact of international migration on agricultural production. Results indicate a reduction in total and per capita labor dedicated to farming. Despite this decrease in labor, no difference is observed in agricultural income between migrant and nonmigrant households, while migrant households have higher total income, on average. The authors suggest that migrant households are able to maintain the same level of agricultural income due to increased specialization and/or utilization of less labor intensive inputs. Higher total incomes could be explained by reallocation of labor towards non-agricultural family enterprises. Konica and Filer (2009), using data from 1996 (one of the earliest household surveys conducted in Albania with an extensive migration module), note that 20% of remittance inflows were invested in family businesses. In terms of labor supply, they find no impact of migration and remittances for men and a negative impact for women. Potentially, this is due to men working in family businesses funded by remittances, and women increasing their time in home production. The study, however, does not address possible endogeneity issues and thus it is unclear whether the decrease in labor supply is a consequence of migration and remittances, or vice versa. 2.2 Schooling Formal education in Albania consists of primary (grades 1-5), lower secondary (grades 6-9), and upper secondary (grades 10-12). Children are eligible to enroll once they turn six and compulsory 12 Data was obtained from the World Bank Migration and Remittances Data webpage. The Bilateral Migration Matrix 2010 data file indicates that figures are Ratha and Shaw (2007) updated with additional data for 71 destination countries as described in the Migration and Remittances Factbook

12 schooling includes the primary and lower secondary sequence, in other words, up to ninth grade. 13 The transition to a market economy, the pyramid scheme collapse, and the massive post-communism out-migration impacted the education system in a negative way. Although improvements have been made since the early 2000s, many issues still persist. Albania has long struggled to achieve satisfactory upper secondary school enrollment rates. Data suggest that a large fraction of students do not make the transfer from lower to upper secondary school, and those who do have a high probability of dropping out. Results from the 2005 Multiple Indicator Cluster Survey (MICS) show that 96% of 13 year olds attended school, compared to 83% of 14 year olds and 58% of 17 year olds. In terms of gender, girls start school earlier 14 but appear to have a higher probability of dropping out during upper secondary school with 16 being the critical age (Albanian National Institute of Statistics, 2007). Other disparities in attendance rates exist between urban-rural areas and socioeconomic status the difference is particularity stark between children in the poorest and richest wealth quintile (Albanian National Institute of Statistics, 2007). 15 In terms of schooling outcomes, despite substantial progress, Albania continues to lag behind countries of comparable economic standing. According to data from the Program for International Student Assessment (PISA), which measures performance on mathematics, science, and reading for 15 year olds enrolled in school, Albania had the lowest scores in 2012 in all of Europe and Central Asia (World Bank, 2014). Compared to Bulgaria, Romania, and Serbia, students in Albania are behind by one school year; while the difference is larger (about 2.5 years) when compared to OECD member countries (World Bank, 2014). Domestically, urban areas outperform rural ones by one year of education in all fields (World Bank, 2014). This urban-rural gap, while not unique to Albania, is larger than the average gap in Europe and Central Asia. Hence, Albanian children in rural areas are not only less likely to attend school but even when they attend, they perform worse than those in urban areas. As for gender, girls perform better than boys in all three subjects but the difference is particularly large in reading equivalent to 1.5 years of schooling and the largest 13 Prior to the 2008/09 academic year, compulsory schooling was 8 years and the education structure was different, consisting of primary (grades 1-4), lower secondary (grades 5-8), and secondary (grades 9-12). 14 It is not clear why in Albania girls start school at a younger age than boys, on average. 15 MICS 2005 uses principal component analysis on household assets to construct wealth quintiles. Assets used were: number of persons per sleeping room, type of floor, roof and wall material, type of cooking fuel, household ownership of television, mobile telephone, fixed line telephone, refrigerator, washing machine, watch, bicycle, motorbike, car, type of drinking water, type of sanitary facility. 12

13 achievement gap between boys and girls of any country in the region (World Bank, 2014). 16 PISA data permit the identification of Albanian students abroad, both native-born and those with Albanian born parents. Using PISA scores to draw comparisons, on average, Albanian students abroad perform worse than their peers in the host country, but above students in Albania (World Bank, 2014). This discrepancy between Albanian students at home and abroad could be due to a number of factors, such as school quality, teaching methods, returns to education, or positive selection of migrants. The World Bank (2014) report examines the socioeconomic status of migrants and finds that Albanian students in Greece do not appear to be different from those in Albania, but those in Switzerland tend to fare better. 17 This suggests that positive selection of migrants may exist in some cases but not all. A few studies have examined the impact of international migration on the schooling of children left behind in Albania. In particular, Giannelli and Mangiavacchi (2010) focus on the consequences of parental separation. They conclude that parental migration negatively impacts children s schooling in the long run, by increasing the probability of falling behind and dropping out. A concern with this study is that the authors do not address the endogeneity of migration. Moreover, children with former migrant parents are treated the same as those with parents still abroad. In a more recent study, Mastrorillo and Fagiolo (2015) use instrumental variables estimation and determine that migration of family members negatively affects school enrollment, especially for non-compulsory school age children. The authors use current migration rates as an instrument for migration. It is plausible, however, that current migration rates affect school enrollment not only through migration but also through contemporaneous peer effects, thus jeopardizing the exclusion restriction (Antman, 2011). While these studies suggest that children in migrant households are less likely to attend school or attend without delay, neither considers the direct impact of remittances, the main focus of this paper. 16 This difference persists even after controlling for grade and other child characteristics. 17 PISA s Index of Economic, Social and Cultural Status (ESCS) is constructed using variables such as parental education, parental occupation and family home possessions. 13

14 3 Data and Descriptive Statistics Data for this study are drawn from the Albanian Demographic and Health Survey (ADHS), administered by the Albanian Institute of Statistics (INSTAT) and the Institute of Public Health (IPH) at the end of 2008 and beginning of Using a nationally representative sample, the survey was designed to capture key characteristics of the population. In addition to a central focus on health outcomes, the study collected extensive data on migration. Two modules were devised on migration, one pertaining to present members of the household and the other to those abroad at the time of survey. The head of the household (or respondent) was asked to provide information on all usual members of the household, from year 1990 or later, who were residing outside of Albania at the time of the survey. Country of residence, year of migration, gender, age, marital status, relationship to the head of household, and education at time of migration, were among the recorded characteristics. The remittance question, with a reference period of 12 months, was inquired for each international migrant. In other words, the survey respondent was asked to recall whether the migrant sent money or goods to the household during the year preceding the survey. Information on the use and value of transfers was unfortunately not obtained. In this study, a household with at least one member abroad in 2007 is considered a migrant household. A remittance household is a migrant household that received remittances. Non-migrant households were not administered the remittance question. While it is possible that these households received remittances from non-household members abroad, this lack of information does not impact the main results of the paper since they are based on migrant households only. In terms of elapsed time since migration, a few restrictions are made. Migrants who left Albania before 1990, i.e., during communism, are excluded from the analysis, since pre-regime and post-regime migration patterns differ. Further, given that the identification strategy relies on the unexpected onset of the global financial crisis specifically exploiting the change in the unemployment rate from 2007 to 2008 as an instrument for remittances it is necessary for migrants to have been at the destination both in 2007 and Theoretically, migrants who left Albania post-shock were aware of the change in the unemployment rate in each country, and this information could have influenced their choice of destination. To avoid this bias, all migrants who left Albania in

15 or 2009 are excluded. Additionally, to capture working age migrants, only migrants between the ages of 15 and 65 are included. The final sample of migrants is For further details regarding the sample selection please see Table A1 in Appendix A. 3.1 Migrant and remittance households A comparison of migrants and non-migrants from the same age group indicates that migrants are younger on average by about six years, disproportionately male, and equally likely to be married (see Table 1). The survey collected information on the highest level of schooling that migrants attended prior to leaving Albania. Estimates show that migrants are less likely to drop out of school before fourth grade but more likely to do so during upper primary and secondary school. Consequently, while 12.3% of non-migrants have attended college, only 7.5% of migrants have done so. This suggests an intermediate selection in terms of education. It is important to note that these results may understate the schooling of migrants since some may have continued their education after moving abroad. To see whether this is the case those who went abroad primarily to study are excluded (243 migrants). Differences in educational achievement do not change significantly (see last panel in Table 1). Table 2 shows that a large fraction of migrants reside in Greece (48%), followed by Italy (36%), United Kingdom (5%), United States (4%), Germany (2%), and other European countries (4%). Bilateral migration data from the World Bank suggest a similar dispersion of Albanians abroad. Outflows by year of migration are displayed in Figure 2. Impacts from the 1997 collapse of the pyramid schemes can be seen by the record number of departures in 1998 and the substantial outflows that persisted, peaking again in Using data from the 2005 Albanian Living Standards Measurement Survey (ALSMS), Azzarri and Carletto (2009) note similar trends. Out of 7810 households, 3058 (39%) are migrant households. 18 This analysis focuses on the subset of households with year old children living at home at the time of survey. A total of 2499 households fit this definition, and 638 (26%) are migrant households. Table 3 highlights mean household characteristics by migration and remittance status for this select sample. 19 Results indicate that migrant households are on average smaller in size and older an expected result since 18 For more information on the sample of households please see Section A1 and A2 in Appendix A. 19 Table B1 in Appendix B provides the corresponding analysis for the full sample. Conclusions are similar, suggesting that the subset of households with children ages do not differ markedly from the rest. 15

16 migrants are younger compared to non-migrants. In terms of wealth, 20 migrant households are underrepresented in the top quintiles. This relates to the schooling of the household head, with figures showing that heads of migrant households have, on average, one less year of education. Lastly, migrant households appear to be located predominantly in rural areas. Specifically, only 37% of migrant households live in urban locales compared to 53% of non-migrant ones. These results suggest that migrant and non-migrant households differ from one another in terms of observable characteristics. A comparison of remittance and non-remittance households indicates that some of these differences carry over. Remittance households appear to be less wealthy than their nonremittance counterparts and less likely to live in urban areas. 3.2 Children ages 14 to 18 About 25% of year olds dwell in migrant households, of which 60% belong to remittancereceiving households. It is worth noting some differences between children of these household types, both in terms of schooling and other characteristics. Panel A of Table 4 shows that while attendance drops for all children as they get older, the decline for those in migrant households is striking. Results indicate that 16 year olds in migrant households are 17.6 percentage points less likely to attend school. The difference is also substantial for 17 and 18 year olds, about 8 and 10 percentage points, respectively. In the case of 16 year olds, the disparity in attendance can be partly explained by children from migrant non-remittance households, who (compared to children in non-migrant households) are 26 percentage points less likely to attend school. Generally, current attendance is also affected by past enrollment patterns. The variable overage for intended grade in 2007 denotes if the child had fallen behind in school prior to the academic year. In Albania, a child must be six years old by the start of the academic year in September to enroll into first grade. The over-age variable is calculated by taking the age of the child and subtracting their completed grades and the official first grade enrollment age of six. 21 So, a child who turns six by the first of September and enrolls into first grade would be zero years over-age, i.e., the student will have no gaps in years of education. If the child does not enroll, they will be one year behind or one year over-age for their intended grade the following academic year 20 The wealth index was constructed by ADHS using principal component analysis of household assets, such television, car, bicycle, dwelling characteristics, etc. 21 Specifically: over-age = age - completed grades - 6. Escobal, Saavedra and Suarez (2005) use a similar measure. 16

17 ( = 1). Since exceptions are sometimes made for children with September birthdays when parents request permission for children to enroll in school before they turn six, this paper considers all children with an over-age of one year or less to be on track. Results in Panel B of Table 4 show that children in migrant households were more likely to have gaps in schooling prior to the academic year. Specifically, 86% of children in non-migrant households were on schedule compared to 81% of those in migrant households. The majority of children who are not on track appear to be 2-3 years behind. It is interesting to point out that significant differences are not observed between remittance and non-remittance children. Findings show that 16 year olds in remittance and non-remittance households were equally likely to be on track at the start of the academic year. However, during the academic year non-remittance receiving children were 14.2 percentage points less likely to be attending school. These results suggest that a short-run shock affected enrollment decisions. In terms of other characteristics, mothers of children in migrant households have completed fewer years of education. This difference in schooling is not observed between children living in remittance and non-remittance households. 4 Empirical Methodology The impact of the receipt of remittances on school attendance is modeled by the following regression equation: Attend schoolih = y ih = α + βr ih + X ih λ + e ih (1) y ih = I(yih > 0) where y ih is the latent likelihood of attending school by child i in household h. The indicator function I(yih > 0) is equal to one if y ih > 0, and zero otherwise. R ih is a binary variable noting the remittance status of the household, X ih is a set of individual child and household characteristics considered to be important predictors of school attendance, and e ih is the error term. As discussed earlier, from an econometric point of view, estimation of the parameter β is complicated by the endogeneity of remittances and the confounded migration impact. To illustrate the confounding effects, while not addressing the endogeneity issue, Table 5 columns (1) and (2) examine the impact of family migration on school attendance for children ages

18 omitting the variable for receipt of remittances (R ih ). The household migration status is captured by a binary variable (Migrant HH ih ) which is included in the X ih vector of individual child and household characteristics. For ease of interpretation, the functional form for the conditional probability of attendance is assumed to the linear and equation (1) is estimated using Ordinary Least Squares (OLS). The regression specification in column (1) includes a number of child characteristics, such as age dummies, gender, indicators for gaps in schooling prior to the academic year, and maternal education. Head of household characteristics (education, marital status, age, age squared) and household level variables such as wealth, household composition (denoted by number of children 0-6, number of school-age children, number of working age adults, and those over 65) are also added as covariates. To account for location differences, region indicators and their interaction with an indicator for urban location are included. Lastly, an indicator noting the survey year is added, as some households were interviewed at the end of 2008 while others at the beginning of In column (1) of Table 5 the migration coefficient indicates that children in migrant households are 3.8 percentage points less likely to attend school than those in non-migrant households. At the time of survey, household members were asked about prior migration experiences. To account for children and households with past migration episodes, variables indicating the return migration status of the child and household are added in column (2). While these variables are potentially endogenous, their inclusion does not significantly change the estimated migration effect which remains significant and suggests that having a migrant abroad decreases the probability of attendance by 3.6 percentage points. As expected, a migration episode which occurred when the child was 15 or older has a large and negative effect on attendance, while migration spells before age 15 have no impact. One can think of migration at a young age happening jointly with parents, thus, less likely for work reasons. Furthermore, a temporary gap in education is easier to overcome when it occurs during primary school than secondary school, making the transition back to school easier after returning to Albania. Other covariates have the expected effect on attendance. As age increases, the likelihood of attending school decreases. 22 Prior gaps in school negatively affect current attendance as does the presence of other children in the household, in particular those six or younger. This effect could 22 Results not shown but available on request. 18

19 be attributed to the caretaker role that older siblings often occupy. However, having grandparents at home potentially lessens the caretaker responsibilities and increases the likelihood of school attendance. The education of both the mother and the head of household are positively associated with attendance. Theory suggests that educated parents consider the schooling of children both a consumption and an investment good. The level of education in the household is also related to wealth, which is shown to increase the likelihood of attendance. In particular, being in the top wealth quintile (fifth), compared to the bottom one (first), increases the probability of attendance by 17 percentage points. In columns (3) and (4), the binary variable noting the remittance receipt status of the household (R ih ) is added. The confounding effect of migration and remittances is manifested by these estimations. Results in column (3) suggest that being in a migrant household has a large negative impact on schooling. Specifically, it decreases the likelihood of attendance by 6.9 percentage points. However, if the household receives remittances the negative effect diminishes. It is important to stress that both variables are potentially endogenous and thus a causal relationship is not identified. Nevertheless, this estimation emphasizes the importance of the counterfactual when assessing the impact of remittances. The strategy used in this paper is to focus on migrant households only. This approach eliminates the endogeneity resulting from selection into migration. Moreover, conditional on identification, children in migrant non-remittance-receiving households are a suitable counterfactual for those in migrant remittance-receiving ones. 4.1 Identification strategy It is likely that there is selection into the decision to remit. In other words, migrant remittancereceiving households differ in terms of unobservables from migrant non-remittance-receiving ones. For example, characteristics such as ability, ambition, and shocks to household income affect both receipt of remittances and school attendance, resulting in correlation between the error term (e ih ) and remittances (R ih ). Consistency of least squares estimators relies on the assumption that the error term is unrelated with the regressors. The breakdown of this fundamental assumption means that the OLS estimator is inconsistent. For example, if higher ability migrants are more successful abroad, and so are their relatives at home, the OLS coefficient of the remittance variable would be positively biased, as the effect of ability on schooling is attributed to remittances. On the other 19