Educational Selectivity of Migrants and Current School Enrollment of Children Left-Behind: Analyses in Three African Countries

Abstract

Migration of household members is often undertaken to improve the well-being of individuals remaining in the household. Despite this, research has demonstrated inconsistent associations between migration and children’s well-being across sending areas and types of migration. To understand the degree to which different types of migration and migrants are associated with schooling, we analyze comparable data across three African countries differing in prevalence, type, and selectivity of migration. Results suggest that recent migration is differentially associated with left-behind children’s school enrollment across settings. When analyses are restricted to migrant-sending households, however, migrant selectivity is positively associated with school enrollment.

Introduction

In 2015, the United Nations adopted the Sustainable Development Goals to improve human development around the globe (UNDP 2015b). Goal 4 sets a target of achieving universal primary and secondary education for all children by 2030. To reach such a macro-level goal, it is necessary to understand the individual- and family-level factors that impede and facilitate sending children to school, keeping them in school, and achieving while in school, particularly in contexts where overall education levels are historically low.

Labor migration can be an important strategy for families to increase resources and invest in children’s well-being (Stark 1991). Economic resources generated by a migrating household member (most often a parent) can then be used to enhance children’s access, persistence, and success in formal school settings (Amuedo-Dorantes, Georges, and Pozo 2010, Chen et al. 2009, Edwards and Ureta 2003, Piotrowski and Paat 2012). Yet migration may also create a competing activity that reduces children’s school enrollment, becoming a normative path that reduces incentives for furthering education among young people who intend to become labor migrants themselves (Fox et al. 2012, Kandel and Kao 2000). In this case, migration and education are not complementary but competitive routes to social mobility in migrant-sending communities. The extant literature contains findings consistent with both roles of migration, suggesting both positive and negative forces on children’s schooling (Deb and Seck 2009, Lu and Treiman 2011, McKenzie and Rapoport 2011, Meyerhoefer and Chen 2011, Robles and Oropesa 2011).

Assessing migration’s role on children’s well-being is also complicated because selection into migration and the returns from migration itself can operate differently across contexts. In well-established settings of labor migration, the incentive for migration may be highest among those with lower stores of human capital (i.e. negative selection on education). In other settings, particularly where the cost of migration is higher and migrant networks are less established, migration occurs more frequently among those with greater human capital (i.e. positive selection on education) (Feliciano 2005, Takenaka and Pren 2010, Mora and Taylor, 2005). Thus, there may be great variation in the selection of migrants and the subsequent impact of migration on children from diverse regions.

This paper takes advantage of comparable data across three dynamic settings in Sub-Saharan Africa to address two research questions: (1) Is recent migration from the household similarly associated with school enrollment of left-behind children across three different settings characterized by both internal and international labor migration? and (2) Does the association between recent migration and children’s school enrollment vary depending on whether migrants are positively selected on education? To answer the latter question, we construct a measure of migrant selectivity based on the migrant’s location in the educational distribution of the population of the same age and gender in the sending context. The social context in which migration occurs may alter the relationship between education and the returns to migration for children left behind by migrants. Thus, we compare patterns of school enrollment among children in sending households in three understudied settings of migration: Burkina Faso, Kenya, and Senegal. We further consider the role of destination (international vs. domestic destinations for migrants) and remittances in the relationship between migrant selectivity and children’s school enrollment in these three diverse settings.

Background

Migration is not a random event, and migrants are not randomly selected from across the economic and social realms of sending communities. In the case of labor migration, many migrants come from a middle strata of human capital, where resources are high enough to fund migration trips but low enough to work as an incentive (i.e. a push) to migration (Kaestner and Malamud 2014). The probability of international migration varies depending on the maturity of migrant systems and the degree of educational inequality in sending communities (Takenaka and Pren 2010). Living in a community with a higher prevalence of international out-migration, having family members with migration experience, and coming from households with more human capital (education) are all associated with an increased likelihood of migration (Massey and Aysa 2005). Costs for internal migration may be lower than in the case of international migration but here too, depending on the setting, human capital, household resources and community characteristics can help predict who becomes a migrant (De Vreyer, Gubert, and Roubaud 2010). But internal migrants may need lower stores of human capital and resources than international migrants from the same sending communities (Binci and Giannelli, 2016).

Overall, there is strong evidence of positive selection among many international migrants. Positive selection is evident when individuals who have higher stores of human capital than average for their sending community have greater opportunities through migration than through remaining in their community. Feliciano (2005), for example, finds that most immigrants to the United States are positively selected on education when compared to their country of origin. Positive selection from developing settings is often equated with ‘brain drain,’ particularly in the case of legal international migration to the United States or other OECD countries (Adams 2003). Among health care workers from Sub-Saharan Africa, for example, demand for skills is high abroad and often poorly compensated in the country of origin (Syred 2011). However, positive selection is not restricted to international migration. There is some evidence of positive educational selectivity for internal migration into capital cities in West Africa (De Vreyer, Gubert, and Roubaud 2010). In particular, female education is predictive of migration from rural to urban areas in Sub-Saharan Africa (Brockerhoff and Eu 1993, Reed, Andrzejewski, and White 2010).

Negative selection, on the other hand, is evident when individuals with lower stores of human capital compared to the average in the sending community are drawn into migration. Negative selection is more likely when local labor markets disfavor poorly educated workers, when costs of migration are low, and when destination markets offer employment opportunities for those with low levels of education. Negative selection is more likely where costs and structural barriers to migration are low (Orrenius and Zavodny 2005). In Mexico, for example, international migrants tend to be positively selected from sending communities but internal migrants, particularly those headed to other agricultural areas within Mexico, are less positively selected on education (Deb and Seck, 2009, Mora and Taylor, 2005). Although rural-urban internal migrants tend to have higher levels of education than their rural counterparts in countries we compare here, including Kenya and Senegal (De Brauw, Mueller, and Lee 2014b), there is less positive selection for internal rural-rural migrants who move to work in agriculture or extraction (i.e. mining). In this case, domestic migrants may be more likely to be negatively selected than international migrants who face more structural barriers and distance to travel (Mora and Taylor, 2005).

Other characteristics associated with migrant selectivity may also be associated with children’s schooling. Some migrants, for instance, may be better positioned to remit back to origin households. In many sending communities around the world, remittances have a positive effect on children’s entrance to and persistence in school (Amuedo-Dorantes, Georges, and Pozo 2010, Chen et al. 2009, Edwards and Ureta 2003, Piotrowski and Paat 2012). Remittances have a positive effect on the educational performance of children of internal migrants in Western China (Hu 2013) and offset the negative impact of parental absence of children’s school enrollment in Ghana and South Africa and Vietnam (Adams, Cuecuecha, and Page 2008, Binci and Giannelli, 2016, Lu and Treiman 2011).

One approach researchers take to deal with the role of migrant selection and children’s well-being is to adjust for migrant selection statistically (i.e. with instrumental variables, propensity scores or other techniques) to help articulate the causal relationship between migration and impacts on origin communities. Previous work taking these approaches has yielded very different results for the association between migration and children’s outcomes across a diverse array of sending communities (Antman, 2013, Halpern-Manners, 2011, Hu 2013). Rather than seeking to explain away the selection effect, we focus on understanding whether such differential migrant selection can help explain some of these varied outcomes for children in migrant sending households. The analyses attend to the type of selection (relatively positive or negative), type of destination (internal or international), and receipt of remittances as important mechanisms through which migration may be associated with children’s school enrollment in diverse and understudied settings of migration.

Research Settings

Our study examines the question of whether migrant selectivity is differentially associated with children’s school enrollment in three African countries: Burkina Faso, Kenya, and Senegal (see Figure 1). These countries represent various sub-regions of Sub-Saharan Africa: Burkina Faso and Senegal in West Africa and Kenya in East Africa. They also differ in the prevalence, type, and selectivity of migration, as well as in education levels, opportunities for education, and education systems (Plaza, Navarrete, and Ratha 2011, UNESCO Institute for Statistics 2013, Miller and Elman 2013, République du Sénégal 2003). Though school-age children are expected to attend school in all three countries, many do not because they lack access to schools, have parents who cannot afford school-related expenses, or are needed to engage in paid or unpaid labor. Compulsory education is rarely, if ever enforced, and children and parents do not suffer sanctions for not attending school.

Figure 1.

Map of study countries

In all three countries, internal migration, primarily rural-urban, is relatively common (De Brauw, Mueller, and Lee 2014b). The opportunity to receive higher wages, in both the formal and informal sectors, attracts rural migrants to cities and towns (De Brauw, Mueller, and Lee 2014b). The scarcity of secondary schools in rural areas also increases the likelihood that individuals migrate to urban areas for schooling-related reasons (de Brauw, Mueller, and Lee 2014a). International migration, on the other hand, varies considerably across all three countries in prevalence and in terms of the number and distance of countries to which migrants travel (Plaza, Navarrete, and Ratha 2011). These differences are likely a function of past colonialism, geographic location, and the availability of local economic opportunities.

1. Burkina Faso

Burkina Faso, a landlocked country in West Africa, is one of the poorest and least developed countries in the world. Gross national income per capita (GNI) is $1,591 (2011 PPP $) (UNDP 2015a). According to the Human Development Index (HDI), it ranks 183 out of 188 countries in its level of development. Burkina Faso is one of the least-educated countries in the world: the adult literacy rate is 28.7% among the population 15 years and older (UNESCO Institute for Statistics 2013). Although primary education is free and children are expected to attend school for 6 years, parents are required to pay for books and supplies and contribute to school funds as needed. Parents may also weigh the long-term benefits of schooling against the short-term loss of income if children were to attend school rather than engage in paid labor. These costs may deter parents from sending their children to school. Approximately 39% of 5- to 14-year-old children are involved in economic activities in Burkina Faso (UNICEF 2014). Because school fees are required for secondary school, they can serve as an obstacle preventing children from continuing their education. Very few children, however, are eligible to attend secondary school given that most children do not finish primary school. In 2006, only 31% of children in the relevant age group completed primary school (de Hoop and Rosati 2014).

Approximately 71% of Burkina Faso’s population lives in rural areas (United Nations and Population Division 2014) and relies on subsistence agriculture. Frequent droughts, especially in the northern part of the country, make it difficult for people to make a living (Niemeijer and Mazzucato 2002). Labor migration is, thus, an important strategy that households employ to diversify their risk and protect themselves against crop failure (Wouterse and Taylor 2008). Burkina Faso has a long history of internal and international migration (Konseiga 2007, Cordell, Gregory, and Piché 1996). Like most African countries, internal migration in Burkina Faso is largely rural-urban (de Brauw, Mueller & Lee, 2014) and directed toward large cities, in this case, Ouagadougou and Bobo Dioulasso (Beauchemin and Schoumaker 2005). Most migration is international and likely influenced by its colonial past: During much of the twentieth century while under French colonial rule, a system of sending male migrant workers to Cote d’Ivoire’s cocoa and coffee plantations existed (Cordell, Gregory and Piché 1996). Though Burkina Faso gained independence from France more than fifty years ago, labor migrants continue to travel to Cote d’Ivoire in search of work. Today, more than 80 percent of international migrants from Burkina Faso migrate to Cote d’Ivoire (Plaza, Navarrete, and Ratha, 2011), and most of these migrants have little or no education.

2. Kenya

Of the three countries in this study, Kenya performs the best on economic, development, and education indicators. It is ranked 145 out of 188 countries in level of development and its GNI per capita is $2,762 (2011 PPP $) (UNDP 2015a). Since independence, Kenya’s government has made education a priority (Somerset 2009). Its total public expenditure on education, 7.2% of GNP, is higher than the average for Sub-Saharan Africa (Miller and Elman 2013). The importance placed on education can be observed in Kenya’s relatively high education levels: the adult literacy rate is 72.2% (UNESCO Institute for Statistics 2013). Children start school at age six and are expected to attend eight years of primary school and four years of secondary school. Kenya implemented three separate initiatives to expand access to free primary education, first in 1974, the second in 1979, and more recently in 2003.² The number of children enrolled in school dramatically increased following each initiative (Somerset 2009). The most recent initiative abolished all direct costs, including those for books, school supplies, and school maintenance, and, thus, removed one of the major barriers to attending primary school. Although primary school is free and most children attend primary school for at least a few years, not all children pass the primary school leaving exam to continue on to secondary school. Even if children pass this exam, parents may lack funds to pay secondary school fees.

Similar to Burkina Faso, approximately 75% of Kenya’s population lives in rural areas (United Nations and Population Division 2014), and engages in subsistence agriculture. To increase livelihood security, many rural households engage in labor migration, mainly toward urban centers with more employment opportunities. Circular migration - where a family member migrates to another area to live and work, sends remittances, and eventually returns home – is not uncommon (Bigsten 1996). In such cases, the household head, usually the husband, migrates to an urban area, leaving his wife and children behind (Agesa 2004). There is also a great deal of seasonal migration of men from rural areas to other rural areas. Though less common, some Kenyan migrants move to international destinations such as the United States and United Kingdom (Plaza, Navarette, & Ratha 2011). Relatively few, however, migrate to other African destinations. Compared to migrants from Burkina Faso and Senegal, Kenyan migrants have much higher levels of education, suggesting more positive selection particularly in the case of international migration.

3. Senegal

Senegal, located in West Africa, ranks as one of the poorest and least developed countries in the world, although it performs slightly better than Burkina Faso on a number of economic, development, and education indicators. Senegal’s GNI per capita is $2,188 (2011 PPP $) (UNDP 2015a), and the country ranks 170 out of 188 countries in the level of development (UNDP 2015a). Education levels are relatively low: only half the population 15 years and older is literate (UNESCO Institute for Statistics 2013). In 2001, Senegal attempted to increase education levels by launching a country-wide effort to guarantee education to all children (République du Sénégal 2003). With the financial assistance of the United Nations and donors, the Senegalese government increased the supply of schools and teachers, and, as a consequence, primary school enrollment grew from 62% in 2003 to 73% in 2009 (World Bank 2016). In Senegal, the official starting age is seven years, and school fees are not required to attend primary or secondary school. Despite free schooling, many children do not attend school, often because parents need them to work. Approximately 17% of 5- to 14-year-old children are involved in child labor (UNICEF 2014).

Senegal has a long history of migration, initially as a destination for migrants and, more recently, as a country of emigration (Adepoju 2004). In 2005, nearly half a million Senegalese lived outside Senegal (D. Ratha and Xu 2007). Among these migrants, 40% were living in other African countries (Gerdes 2007), representing more varied destinations than migrants from Burkina Faso. The destination countries of Senegalese migrants have changed over time. In the 1960s, the top destinations were Mauritania, Mali, Guinea, and Guinea-Bissau. By the late 1960s, Cote d’Ivoire and Gabon began replacing these countries in popularity. Migration to Cote d’Ivoire continued, becoming the top destination for Senegalese migrants moving within Africa, until civil war broke out in Cote d’Ivoire in the early 2000s and disrupted this flow (IOM 2009b). Now Gambia is the most popular destination.

Senegalese migration to non-African destinations commenced while Senegal was a French colony. During World War I and II, many Senegalese men served in the French army and remained in France after their service, finding employment in Marseille harbor (Robin, Lalou, and Ndiaye 2000, Gerdes 2007). After Senegal gained independence from France in 1960, migration toward France intensified with the recruitment of Senegalese workers to France’s growing automobile industry (Pison 1997). In 1985, migration to France became more difficult because Senegalese now needed visas to enter the country. Consequently, Senegalese migration moved toward other European countries, especially Italy and Spain (Toma and Castagnone 2015, Gerdes 2007). Many international migrants come from some of Senegal’s poorest regions and have little or no formal education (Auriol and Demonsant 2012, Diatta and Mbow 1999); however, a substantial minority, approximately 25 percent, represent the other end of the spectrum with tertiary education (IOM 2009b).

The Current Study

In this study, we first ask whether recent labor migration is associated with school enrollment of left-behind children when compared to children in non-migrant households. We examine the extent to which international or internal migration (or both) are positively associated with school enrollment even beyond the educational attainment levels of adults in the sending households when compared to households without migrants. We expect some differences across the three contexts we compare. In Burkina Faso and Senegal, contexts where migrants tend to be less positively selected, we expect a smaller association of migration and children’s school enrollment than we expect in Kenya when analyses compare migrant households to non-migrant households.

We next consider whether educational selection of migrants themselves has an additional role to play in children’s school enrollment. If we observe that children are more likely to be enrolled in school in the presence of a positively selected migrant even when adjusting for the education levels of other adults in the household and household resources, it would suggest that the dichotomous migrant vs. non-migrant distinction misses an important part of the role of migration for those left-behind. Rather, such a result would indicate that differential returns to migration for sending households are associated with the relative ‘quality’ of the migrant they are able to send.

We further examine the role of educational selection beyond the role it plays in type of migration (i.e. beyond the possibility that more positively selected migrants travel internationally vs. internally) or remittances (i.e. beyond the possibility that positively selected migrants are more likely to remit). These analyses are confined to children in migrant sending households only. Here too, we expect to observe some differences across our three contexts. Economic returns to education in Burkina Faso and Senegal are lower than in Kenya so having a positively selected migrant (i.e. living in a household sending migrants with relatively higher levels of education than the context in general) will be positively associated with children’s school enrollment in Burkina Faso and Senegal with a smaller role of migrant selection observed in Kenya.

Data and Methods

There are comparatively few studies of children in sending communities in Africa. We use data from the Migration and Remittances Household Surveys conducted as part of the World Bank’s Africa Migration Project. These cross-sectional surveys, conducted in 2009 and 2010 in six Sub-Saharan African countries (Burkina Faso, Kenya, Nigeria, Senegal, South Africa, and Uganda), collected comparable household-level data on the characteristics of migrants in sending households, remittances sent to their households, and the characteristics of return migration.³ Different methodologies, however, were used to obtain the sampling frames in each country. In Senegal, a nationally representative sampling frame was used while in Burkina Faso and Kenya, sampling frames were representative at the provincial and district level, respectively. Because the Migration and Remittances Household Surveys were designed to gather information about both internal and international migrant-sending households, each country’s sampling frame needed to include a sufficient number of migrant households. Even in countries with high rates of international migration, it is difficult to find households that have migrants currently living abroad (McKenzie and Mistiaen 2007). To capture a sufficient number of migrant-sending households in the primary sampling units, survey teams first conducted household listings with the purpose of classifying households into one of three categories: non-migrant, internal migrant, and international migrant. A household was considered to be a migrant-sending household if at least one former household member lived in another village, urban area, or country for at least six months before the time of the survey. Next, survey teams randomly selected equal numbers of households according to their migration status, purposively oversampling internal migrant and international migrant households. In each household, interviewers surveyed the household head on all modules except the return migrant module, for which the return migrant was surveyed. For the purposes of this survey, the household head was required to be a regular member of the household who is primarily responsible for managing the household’s financial resources, regardless of whether household members usually viewed this member as the household head. Thus, for the purposes of this survey, household members living outside the household, including migrants, were not eligible to be the household head.⁴

Our analytic sample is composed of school-aged children reported by the household head to be living in the surveyed households at the time of the survey. In Burkina Faso and Kenya, eligible children were aged 6-17 years, and in Senegal, they were 7-17 years.⁵ We excluded children from the household who were living elsewhere at the time of the survey because information on their schooling status was not collected in all three countries. If children are living outside the household for schooling-related reasons and migrant-sending households are more likely to have children living outside the household for this reason, then our study could underestimate migration’s effect on school enrollment.⁶ We further restricted our sample to children with non-missing data on all dependent and independent variables: Burkina Faso (N = 6621), Kenya (N = 2182), and Senegal (N = 5220).⁷

Outcome Measure

The dataset is primarily focused on migration and the economic status of households with fewer measures available to compare children’s outcomes across countries. Although several measures of education, including educational attainment and number of grades of schooling completed, were collected in the Migration and Remittances Household Survey, we focused on children’s current school enrollment as our primary outcome of interest because it was the only schooling variable consistently measured across the three countries. Furthermore, current school enrollment measures a recent decision made by parents or the household head to send or keep a child in school and thus is more likely to be affected by the recent labor migration of a household member. Educational attainment and grades of schooling completed capture a child’s cumulative schooling history and may be less impacted by recent migration.

We created the variable current school enrollment, using information collected in the household roster. This variable is measured using the household head’s response to the following question: “What is (NAME)’s current work situation?” If the respondent reported “full-time student”, then we coded the child as currently in school.⁸ All other responses were coded as not enrolled in school. We note that this is a conservative measure of school enrollment because children working and going to school may not be coded as being in school.

Migration Variables

Our focus is on recent labor migration out of the focal households. We coded migration using information collected from each household in a roster of former household members reported by the household head. Former household members are coded as migrants if they had lived outside the household for more than six months before the time of the survey.⁹ Household heads answered a series of questions about each migrant’s sociodemographic characteristics and migration experience. The first variable, household migration status, is a binary variable indicating whether a household had a former household member working as a labor migrant. A former household member is defined as a labor migrant if the household head reported work-related reasons as the primary reason that a given household member was living outside the household. Because some labor migrants may have left the household decades ago and had little to no contact with the household, we limited our migrant sample to labor migrants who were reported to be living in his/her current location for the last five years. Ideally, we would have restricted the sample to migrants who left the household in the last five years; however, this information was not collected in Kenya. Instead, we captured duration of migration using the following question: “How long has (NAME) lived in his/her current location?”¹⁰ We coded a household as a migrant household if the household head reported at least one former household member who lived outside the household for work-related reasons and resided in his/her current location for five years or less. All other households were coded as non-migrant households.

Our study goes beyond the dichotomy of migrant versus non-migrant household by taking into account the educational selectivity of migrants living outside the households. We consider whether school enrollment is higher for children in migrant-sending households when migrants have higher levels of education than individuals of the same age and gender in their communities. To do so, we followed Ichou’s (Ichou 2014) methodology to construct a migrant relative education score measuring the migrant’s location in the educational distribution of the population of the same age and gender. Specifically, this variable captures “the percentage of people of the same country of origin, gender, and age, who have a lower level of educational attainment, plus half the percentage of the people with the same level of education” (Ichou 2014). Feliciano and Lanuza (2017) rely on a similar measure they refer to as ‘contextual attainment’ and that we refer to as ‘relative education.’ First, we coded migrant’s educational attainment into the following categories: none, primary, secondary, or tertiary. We then located the migrant’s educational attainment in the distribution of educational attainment of individuals who are of the same age and gender in the country of origin, using data from the country’s Demographic and Health Survey (DHS)¹¹ that took place closest in time to the Migration and Remittances Household Survey. We constructed a migrant’s relative education score by summing the percentage of the population of the same age and gender who have lower levels of education and adding half the percentage of the population with similar levels of education as the migrant. For example, in Kenya, the distribution of educational attainment among 20- to 24-year-old men is as follows: 3% have no education, 49% have primary education, 38% have secondary education, and 11% have tertiary education. Thus, the relative education score for a 23-year-old male migrant with secondary education is equal to 71% (3% + 49% + 0.5*38%), which indicates that the migrant’s educational attainment is higher than or the same as 71% of the population of the same age and gender. If a household has more than one labor migrant, we used the highest migrant relative education score.

We also considered the possibility that school enrollment will vary according to the migrant’s destination, specifically whether the migrant is an internal or international migrant, recognizing that migrant education and destination are also related. Finally, we constructed a variable measuring whether the household reported receiving any remittances in the past year. Households that received remittances may have had more financial resources to invest in children’s schooling, which could have increased the likelihood that children were enrolled in school.

Control Variables

We include several control variables associated with children’s school enrollment: age, gender, urban residence, region/province, and number of children in the household. Because family and household resources including education, land, and income are all positively associated with children’s schooling in African countries just as they are in other contexts (Buchmann 2000) and because these finite resources are shared among household members, we control for household wealth and number of children living in the household. To capture household economic resources, we used data collected on household assets to construct a household wealth index using principal components analysis (Filmer and Pritchett 1999). Household wealth is measured in quintiles and coded in the following manner: poorest, second, middle, fourth, and richest. We also include controls for characteristics of the household head because he or she has influence over whether a child attends school. Age of the household head is a proxy for family life cycle stage that is an important predictor of selection into migration (Durand and Massey 1992). The child’s relationship to the household head (child, grandchild, brother/sister, nephew/niece, other) could also be associated with school enrollment. Children with closer biological ties to the household head are more likely to be enrolled (Case, Paxson, and Ableidinger 2004). Finally, we included a measure capturing the educational attainment of adults in the sending household.¹² This variable measures the highest educational attainment (none, primary, secondary, tertiary) of an adult living in the household. In this way we can assess the role of the migrant’s education net of the education levels of other adults in the child’s household.

Methods

We first consider whether having a recent migrant from the household is associated with a higher probability of school enrollment, controlling for the educational attainment of adults in the household with separate multivariate logistic regression models for each country. Our first set of models show how household educational attainment and child and household-level control variables predict current school enrollment. Next, we assess whether there is an additional role of migration by adding a variable measuring household migration status – no migration, internal migration, and international migration – to the model.

We next follow the example of Ichou and others to consider the importance of migrants’ own relative education net of household relative education within the subsample of households with recent migration (Feliciano and Lanuza 2017, Ichou 2014). These analyses do not account for unobserved factors that predict both migration and school enrollment beyond the educational attainment of adults in the household. Rather, with the cross-sectional data available, we can assess the importance of one observable source of that selectivity and the extent to which the differential patterns of migrant education selection across these three contexts are consistently associated with school enrollment net of the underlying education levels of the household. In the first set of models, we show how household educational attainment is related to current school enrollment, controlling for child and household-level control variables. Next, we added migrant’s relative education score and remittances to our models.¹³ Finally, we added type of migration (internal or international) to assess whether this measure mediates the relationship between migrant selectivity and current school enrollment.

In all models, we adjust standard errors to take into account the clustering of children within households because some households contribute more than one child to the analytic sample. We also tested for interaction effects between migration variables and children’s age and gender (Acosta 2011, Antman 2012, Creighton and Park 2010). Because none of the interaction terms were statistically significant across all three countries, we do not include them in the models presented in this paper.

In exploratory models, we took into account the survey’s sampling design by incorporating survey weights into our regression models. We obtained similar results to those presented in the current paper. Because Kenya did not include survey weights in their publicly released dataset and because survey results for Burkina Faso and Senegal do not vary with or without survey weights, we chose to maintain consistency across all three countries by presenting descriptive statistics and regression models without survey weights.¹⁴.

Results

Descriptive Statistics

The characteristics of children in the analytic sample by country are presented in Table 1. In all three countries, the mean age of children is approximately 11 years, and girls make up slightly less than half the analytic sample. Approximately 50 percent of children live in urban areas, except in Burkina Faso, where fewer than 5 percent are urban dwellers. In Burkina Faso, children are evenly distributed by household wealth, which is not the case in the other countries. A greater proportion of children live in poorer households in Kenya while the opposite is true in Senegal. The mean number of children living in a household varies greatly by country, ranging from an average of three children per household in Kenya to an average of seven in Burkina Faso and Senegal.

Table 1.

Characteristics of Children, Migration and Remittances Household Surveys, 2009-10

	Burkina Faso	Kenya	Senegal
Mean age (years)	10.9	11.4	11.2
Female (%)	47.9	49.9	48.7
Lives in urban area (%)	3.5	43.2	53.9
Household wealth index (%)
Poorest	23.5	21.8	11.0
Second	17.4	26.5	19.1
Third	23.1	19.1	22.6
Fourth	18.3	18.1	24.3
Wealthiest	17.6	14.6	22.9
Mean number of children living in household	7.2	3.5	7.4
Mean age of household head (years)	51.6	46.9	54.3
Relationship to household head (%)
Child	76.4	75.0	55.9
Grandchild	7.6	15.7	17.1
Brother/sister	3.8	1.5	2.1
Nephew/niece	7.5	3.0	13.1
Other	4.7	4.8	11.9
Household migration status (%)
Non-migrant	68.0	66.7	61.9
Internal migration	12.8	21.9	15.3
International migration	19.2	11.4	22.8
Household educational attainment (%)
None	53.7	7.1	36.5
Primary	24.2	29.1	23.9
Secondary	21.9	39.8	28.9
Tertiary	0.3	24.1	10.8
Total	6621	2182	5220

We also present characteristics of the household head due to the important role he or she typically plays in the allocation of household resources and decisions regarding schooling in Sub-Saharan Africa. The household head’s mean age ranges from 47 years in Kenya to 54 years in Senegal. In all three countries, the sample is primarily composed of biological children, followed by grandchildren, of the household head.

Across all three countries, approximately one-third of the sample lives in migrant-sending households.¹⁵ In Burkina Faso and Senegal, the majority of these children live in international migrant-sending households. Household educational attainment, which captures the highest educational attainment of an adult member, varies across the three countries; it is highest in Kenya, where 64% have secondary or tertiary education and lowest in Burkina Faso, where only 22% have this level of education.

Our dependent variable, children’s school enrollment, also varies considerably across all three countries (not shown). School enrollment is lowest in Burkina Faso, where 42% of school-age children are in school full-time, and highest in Kenya, where 83% are in school. In Senegal, 61% of children are enrolled in school at the time of the survey. In all three countries, current school enrollment varies significantly by household migration status (Figure 2). In Burkina Faso and Senegal, children in international migrant-sending households have lower current school enrollment than those in non-migrant households. No significant difference is observed in these two countries between children in internal migrant-sending households and non-migrant households. In Kenya, we observe a different pattern: current school enrollment is higher in internal and international-migrant sending households than in non-migrant households.

Figure 2.

Children’s current school enrollment by household migration status

Note: Significance levels are relative to non-migrant category.

*** p<30.001; +p<0.10

As expected, school enrollment also varies considerably according to the education levels of adults and migrants in these households. Current school enrollment is higher in households with greater levels of household educational attainment (not shown). A similar pattern is observed in migrant-sending households: current school enrollment is higher in households where migrants have higher relative education scores (Figure 3). In Kenya, however, the proportion currently enrolled plateaus at 85% once the migrant’s relative education score reaches 40 or higher. School enrollment patterns in Kenya are quite different from those observed in Burkina Faso and Senegal. First, the overall levels of full-time school enrollment are higher. Second, school enrollment levels are highest for children in migrant-sending households. This suggests that selection into migration operates differently across the three countries and the possibility of a differential association between migration and children’s schooling, depending on the type of selection into migration that occurs.

Figure 3.

Bivariate association between migrant’s relative education score and children’s current school enrollment

The relationships depicted in Figures 2 and 3 suggest the importance of looking at type and selectivity of migration when examining the association between migration and children’s schooling. In Table 2, we present characteristics of current labor migrants from migrant-sending households in our sample. In all three countries, migrants are, on average, 30 years of age and overwhelmingly male. Close to half of all migrants in Kenya and Senegal come from urban areas while only 4 percent of migrants from Burkina Faso do so. The majority of migrants are children of the household head, with another sizable minority reported to be the household head’s sibling. Whereas approximately two-thirds of migrants in Kenya and Senegal sent remittances in the past year, less than half did so in Burkina Faso. Migrants’ destination also varies across the three countries. Most migrants in Kenya and Senegal are internal migrants, likely engaged in rural-urban migration; in Burkina Faso, 60 percent are international migrants, almost exclusively working in another African country. Approximately half of all Senegalese migrants are working internationally; they are evenly distributed between other African countries and countries outside Africa (not shown). In Kenya, 31% of migrants work internationally.

Table 2.

Characteristics of current labor migrants by country, Migration and Remittances Household Surveys, 2009-10

	Burkina	Kenya	Senegal
Mean age (yrs)	27.9	30.7	32.0
Male (%)	95.3	68.4	93.5
From urban area (%)	4.3	38.1	52.1
Relationship to household head (%)
Spouse/partner	3.1	15.9	7.2
Son/daughter	52.0	66.3	58.9
Father/mother	6.6	0.9	0.3
Brother/sister	30.7	10.6	17.7
Grandchild	5.6	2.3	NA
Other	2.1	3.9	15.9
Sent remittances in past year (%)	43.2	63.4	69.6
Destination (%)
Internal	40.2	69.1	50.6
International	59.8	30.9	49.4
Educational Attainment (%)
None	75.6	0.7	55.4
Primary	15.4	24.6	22.5
Secondary	8.4	47.1	16.6
Tertiary	0.6	27.6	5.6
Mean relative education score	41.1	68.5	47.6
N	682	433	735

NA = Not applicable.

In addition to differences in the destinations of migrants by country, migrants’ educational selectivity varies across the three countries. In Burkina Faso and Senegal, 76% and 56% of migrants, respectively, had never attended school. Among migrants who ever attended school, most attended only primary school. We further examined migrant selectivity by comparing migrants’ educational attainment to that of individuals of the same age and gender (Figure 4). In Burkina Faso and Senegal, we observed that most migrants are negatively selected: 76% and 57% of migrants, respectively, had relative education scores that were below 50. The opposite pattern existed in Kenya, where 75% of migrants were positively selected.

Figure 4.

Distribution of migrant’s relative education score by country

Regression Analyses

We begin our multivariate analyses by examining whether household labor migration is associated with current school enrollment once we control for the other characteristics of sending households in each country (Table 3). Model 1 includes the household educational attainment of adults along with controls for child and household characteristics. Overall, higher household educational attainment is positively associated with children’s school enrollment in all three countries. Other characteristics matter as well. Child’s age is associated with school enrollment in a non-linear fashion, as is consistent with the normative increase with entrance to school and attrition from school in adolescence. Girls in Burkina Faso were less likely to be enrolled than boys, but we do not observe this gender difference in Kenya or Senegal. More household wealth and urban residence were associated with school enrollment in Burkina Faso and Senegal. Children in urban areas in Kenya appeared less likely to be ‘full-time students’ than those in non-urban areas. This likely reflects the sample design and the inclusion of children from very poor communities surrounding large urban areas. Finally, we note that in all three countries, not being a close relative of the household head (i.e. ‘other’ relationship) is negatively associated with school enrollment.

Table 3.

Logistic regression models predicting children’s current school enrollment by household migration status, Migration and Remittances Household Surveys, 2009-10

	Burkina Faso		Kenya		Senegal
VARIABLES	Model 1	Model 2	Model 1	Model 2	Model 1	Model 2
Household educational attainment (ref. None)
Primary	2.10*** (0.17)	2.10*** (0.17)	1.46 (0.50)	1.45 (0.50)	1.95*** (0.26)	1.94*** (0.25)
Secondary	4.38*** (0.44)	4.41*** (0.44)	1.40 (0.49)	1.43 (0.50)	4.02*** (0.56)	3.97*** (0.56)
Tertiary	3.07 (2.09)	3.06+ (2.07)	1.96+ (0.75)	2.03+ (0.78)	4.40*** (1.02)	4.40*** (1.02)
Household migration status (ref. Non-migrant)
Internal		1.10 (0.12)		1.51* (0.30)		1.26+ (0.17)
International		0.88 (0.08)		1.65+ (0.49)		1.09 (0.15)
Child’s age	2.94*** (0.20)	2.95*** (0.20)	2.81*** (0.39)	2.77*** (0.38)	2.58*** (0.21)	2.59*** (0.21)
Child’s age squared	0.95*** (0.00)	0.95*** (0.00)	0.96*** (0.01)	0.96*** (0.01)	0.96*** (0.00)	0.96*** (0.00)
Female child	0.83** (0.05)	0.83** (0.05)	0.97 (0.12)	0.96 (0.12)	0.98 (0.07)	0.98 (0.07)
Urban residence	1.71* (0.38)	1.71* (0.38)	0.44*** (0.08)	0.44*** (0.08)	1.58*** (0.19)	1.60*** (0.20)
Wealth index (ref. Poorest)
Second	1.01 (0.10)	1.02 (0.11)	1.30 (0.30)	1.29 (0.29)	1.61** (0.29)	1.58* (0.29)
Middle	1.19+ (0.12)	1.20+ (0.12)	1.35 (0.36)	1.30 (0.35)	2.34*** (0.45)	2.33*** (0.45)
Fourth	1.26* (0.14)	1.27* (0.14)	1.68+ (0.51)	1.64 (0.50)	2.57*** (0.56)	2.60*** (0.57)
Richest	1.61*** (0.19)	1.61*** (0.19)	1.27 (0.42)	1.22 (0.41)	3.39*** (0.81)	3.43*** (0.82)
Number of children in household	0.96*** (0.01)	0.96*** (0.01)	1.07 (0.05)	1.06 (0.05)	0.98 (0.01)	0.98 (0.01)
Household head’s age	1.00 (0.00)	1.00 (0.00)	1.00 (0.01)	1.00 (0.01)	0.99** (0.00)	0.99** (0.00)
Child’s relationship to household head (ref. Child)
Grandchild	1.01 (0.13)	1.02 (0.13)	0.97 (0.25)	0.92 (0.24)	1.14 (0.16)	1.13 (0.16)
Brother/sister	0.65* (0.11)	0.65* (0.11)	1.17 (0.82)	1.18 (0.85)	0.76 (0.20)	0.77 (0.20)
Nephew/niece	0.89 (0.11)	0.89 (0.11)	0.68 (0.30)	0.70 (0.30)	1.00 (0.14)	0.98 (0.14)
Other	0.38*** (0.07)	0.39*** (0.08)	0.29*** (0.08)	0.29*** (0.09)	0.80 (0.12)	0.79+ (0.11)
Constant	0.00*** (0.00)	0.00*** (0.00)	0.01*** (0.01)	0.01*** (0.01)	0.01*** (0.00)	0.01*** (0.00)
N(children)	6,621	6,621	2,182	2,182	5,220	5,220
R2	0.127	0.127	0.102	0.107	0.205	0.206

Note: Robust standard errors, clustered at household level to account for correlation between household members, are shown in parentheses.

^***p<0.001,

^**p<0.01,

^*p<0.05,

⁺p<0.10

Model 2 adds household migration status, which captures whether the household had engaged in recent migration, and if the household did, the type of migration (internal or international). In Burkina Faso and Senegal, little to no association is observed between household migration status and current school enrollment. In Kenya, in contrast, internal and international migration are both positively associated with schooling.

The results in Table 3 suggest that household educational attainment is indeed predictive of children’s school enrollment. Furthermore, in Burkina Faso and Senegal, there is very little difference in school enrollment by household migration status. For children in Kenya, though, being in a household with recent labor migration, both internal and international, is positively associated with school enrollment, even when controlling for household educational attainment. The next step is to consider whether a differential probability of school enrollment exists depending on whether the household sent a positively selected migrant.

To examine this question, the analyses in Table 4 are restricted to migrant-sending households. The first model establishes that the same factors that were predictive of school enrollment for children in all households (i.e. results in Table 3) are also predictive of school enrollment among children in migrant-sending households only. The results are consistent with the models for children in all households. Household educational attainment is positively associated with children’s school enrollment in Burkina Faso and Senegal, where overall levels of education are low. The age pattern of school enrollment is consistent across all three countries while being more distantly related to the household head (i.e. ‘other’ relationship) is negatively associated with school enrollment in Burkina Faso and Kenya.

Table 4.

Logistic regression models predicting children’s current school enrollment in migrant households, Migration and Remittances Household Surveys 2009-10

	Burkina Faso			Kenya			Senegal
VARIABLES	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3	Model 1	Model 2	Model 3
Household educational attainment (ref. None)
Primary	2.07*** (0.32)	1.92*** (0.30)	1.92*** (0.30)	2.01 (1.16)	1.89 (1.02)	1.86 (0.99)	2.66*** (0.49)	2.11*** (0.40)	2.11*** (0.40)
Secondary	3.71*** (0.66)	3.41*** (0.62)	3.45*** (0.63)	1.78 (1.00)	1.42 (0.74)	1.40 (0.72)	3.62*** (0.71)	2.51*** (0.52)	2.51*** (0.52)
Tertiary	0.88 (1.56)	0.71 (1.13)	0.70 (1.09)	2.90+ (1.82)	2.02 (1.28)	1.98 (1.23)	3.47** (1.49)	2.50* (1.00)	2.50* (1.01)
Migrant’s relative education score		1.01** (0.00)	1.01** (0.00)		1.02+ (0.01)	1.02+ (0.01)		1.02*** (0.00)	1.02*** (0.00)
International migration (ref. Internal migration)			0.86 (0.12)			0.90 (0.35)			0.99 (0.16)
Received remittances in past year (ref. No remittances in past year)		1.04 (0.13)	1.05 (0.13)		1.57 (0.57)	1.59 (0.58)		1.04 (0.18)	1.04 (0.18)
Child’s age	3.23*** (0.41)	3.26*** (0.41)	3.28*** (0.41)	3.96*** (1.23)	3.77*** (1.21)	3.76*** (1.21)	2.99*** (0.39)	3.09*** (0.40)	3.09*** (0.40)
Child’s age squared	0.95*** (0.01)	0.95*** (0.01)	0.95*** (0.01)	0.94*** (0.01)	0.94*** (0.01)	0.94*** (0.01)	0.95*** (0.01)	0.95*** (0.01)	0.95*** (0.01)
Female child	0.80* (0.08)	0.80* (0.08)	0.80* (0.08)	1.49 (0.38)	1.51 (0.39)	1.52 (0.40)	1.00 (0.12)	1.03 (0.12)	1.03 (0.12)
Urban residence	1.14 (0.55)	1.22 (0.61)	1.22 (0.61)	0.57 (0.28)	0.62 (0.29)	0.62 (0.30)	1.32 (0.23)	1.19 (0.21)	1.19 (0.21)
Wealth index (ref. Poorest)
Second	0.78 (0.14)	0.82 (0.15)	0.83 (0.15)	1.30 (0.53)	1.32 (0.57)	1.33 (0.57)	1.47 (0.43)	1.34 (0.38)	1.34 (0.38)
Middle	1.07 (0.20)	1.07 (0.20)	1.08 (0.20)	1.14 (0.51)	1.03 (0.48)	1.05 (0.49)	2.10* (0.67)	1.91* (0.60)	1.91* (0.60)
Fourth	1.00 (0.20)	1.01 (0.20)	1.02 (0.20)	2.41+ (1.29)	1.98 (1.09)	2.02 (1.12)	2.27* (0.74)	2.02* (0.66)	2.03* (0.66)
Richest	1.74** (0.36)	1.65* (0.34)	1.67* (0.35)	0.86 (0.56)	0.63 (0.44)	0.65 (0.45)	2.15* (0.78)	1.71 (0.62)	1.71 (0.63)
Number of children in household	0.96* (0.02)	0.96* (0.02)	0.96* (0.02)	1.16+ (0.09)	1.22* (0.10)	1.22* (0.10)	0.96* (0.02)	0.97+ (0.02)	0.97+ (0.02)
Household head’s age	1.01 (0.01)	1.01 (0.01)	1.01 (0.01)	1.01 (0.01)	1.01 (0.02)	1.01 (0.02)	0.99 (0.01)	0.99 (0.01)	0.99 (0.01)
Child’s relationship to household head (ref. Child)
Grandchild	0.86 (0.16)	0.90 (0.17)	0.91 (0.17)	0.66 (0.31)	0.59 (0.28)	0.59 (0.28)	1.27 (0.29)	1.15 (0.26)	1.15 (0.26)
Brother/sister	1.06 (0.33)	1.09 (0.35)	1.12 (0.36)	0.23 (0.24)	0.16+ (0.17)	0.17 (0.18)	0.91 (0.43)	0.84 (0.41)	0.84 (0.41)
Nephew/niece	1.06 (0.23)	1.05 (0.22)	1.06 (0.22)				0.99 (0.18)	0.96 (0.18)	0.96 (0.18)
Other	0.40** (0.14)	0.40** (0.14)	0.40** (0.14)	0.17*** (0.08)	0.16*** (0.08)	0.16*** (0.08)	1.15 (0.22)	1.03 (0.20)	1.03 (0.20)
Migrant’s relationship to household head (ref. Spouse/partner)
Child	0.58 (0.21)	0.60 (0.21)	0.59 (0.21)	1.15 (0.55)	1.46 (0.70)	1.42 (0.69)	1.02 (0.32)	1.00 (0.31)	1.00 (0.31)
Parent	0.80 (0.36)	0.74 (0.33)	0.74 (0.34)				0.63 (0.57)	0.23 (0.21)	0.23 (0.21)
Brother/sister	0.61 (0.22)	0.61 (0.22)	0.59 (0.22)	1.86 (1.06)	2.38 (1.63)	2.34 (1.62)	0.91 (0.29)	0.92 (0.29)	0.92 (0.29)
Grandchild	0.63 (0.30)	0.65 (0.30)	0.61 (0.29)	1.29 (1.09)	2.18 (1.96)	2.19 (1.95)
Other	0.55 (0.25)	0.57 (0.27)	0.56 (0.26)	0.95 (0.72)	1.16 (0.78)	1.18 (0.79)	1.38 (0.45)	1.40 (0.46)	1.39 (0.46)
Constant	0.00*** (0.00)	0.00*** (0.00)	0.00*** (0.00)	0.00*** (0.00)	0.00*** (0.00)	0.00*** (0.00)	0.01*** (0.01)	0.00*** (0.00)	0.00*** (0.00)
N(children)	2,092	2,092	2,092	680	680	680	1,934	1,934	1,934
R2	0.132	0.137	0.138	0.147	0.164	0.164	0.204	0.225	0.225

Note: Robust standard errors, clustered at household level to account for correlation between household members, are shown in parentheses.

^***p<0.001,

^**p<0.01,

^*p<0.05,

⁺p<0.10

Model 2 in Table 4 adds characteristics of migration, including the migrant’s relative education score and whether the household received remittances in the past year, to the baseline model. Here, the analyses suggest that having a migrant with a higher relative education score is indeed positively associated with children’s school enrollment even beyond the role of household education overall. Receipt of remittances is not statistically significant for predicting school enrollment in any of the three countries.

Model 3 tests whether type of migration (internal or international) helps explain the relationship between migrant selectivity and current school enrollment. Results indicate that type of migration is not significantly associated with school enrollment for children in any of the three countries. Further, including type of migration in the models does not reduce the size of the coefficients for migrant’s relative education. In sum, migrant’s relative education, net of the educational attainment of adults in the household, is an important predictor of children’s school enrollment that is not explained by the likelihood that positively selected migrants go to different destinations or are more likely to remit than those with lower relative education scores.

Discussion

Migration can be an important strategy for households to maximize economic opportunities and enhance children’s well-being. There are many circumstances of migration, however, and they may not all be associated with the same outcomes. Previous research across diverse contexts yields very mixed conclusions about the importance of migration for enhancing children’s schooling in the origin household (Antman, 2012, Binci and Giannelli, 2016, Deb and Seck 2009, Lu and Treiman 2011, McKenzie and Rapoport 2011, Meyerhoefer and Chen 2011, Robles and Oropesa 2011). The analyses presented here build upon this previous work by not only considering whether children’s schooling is associated with recent migration from the household but also examining the potential for migration to have a differential association depending on the educational selectivity of migrants within the household and across diverse contexts. The comparable household-level data collected in three Sub-Saharan African countries (Burkina Faso, Kenya, and Senegal) represent different geographical regions, migration patterns, and migrant selectivity. Burkina Faso represents a context where migrants have lower human capital and fewer resources flowing back to sending households, when compared to Kenya and Senegal. Migrants from Senegal, in contrast, tend to be more positively selected on education and send more remittances. Kenyan migrants are the most positively selected with similar proportions as their Senegalese counterparts sending remittances back to their households.

Overall, there is a weak association between living in a migrant-sending household and current school enrollment across these three different contexts in Sub-Saharan Africa. Rather, children living in households with greater levels of adult educational attainment have higher probabilities of school enrollment in Burkina Faso and Senegal, regardless of whether the household sends migrants or not. Only in Kenya does a strong and positive association exist between household migration and school enrollment after controlling for both child and household characteristics. And, here, both internal and international migration are similarly associated with school enrollment suggesting labor migration in general is helpful for keeping children in school in this context regardless of destination.

However, looking only at a dichotomous indicator of migration without considering migrant ‘quality’ or relative education misses the more complex relationship between migration and children’s schooling. Our second research question, thus, addressed the role of migrant selectivity in the relationship between school enrollment and migration net of the educational attainment of adults in the household. In these analyses, restricted to children in migrant-sending households, migrants with higher levels of education relative to the country of origin were positively associated with children’s school enrollment in the origin household. The results hold whether households send internal or international migrants.

These results also provide some compelling insight for studies that have found very mixed support for the role of migration on children’s well-being. It may be that living in households with positively selected migrants invoke higher aspirations among children who perceive that education and migration are a successful combination (Feliciano & Lanuza, 2017). Such expectations of improved lives may explain why the perception that migrants are successful enhances children’s outcomes regardless of the actual amount of economic remittances provided by the migrant (Yabiku, Agadjanian, and Cau 2012). At the other end of the relative education distribution, the success of labor migrants with low levels of education may provide an additional incentive for children to orient themselves toward becoming labor migrants rather than remaining in school (Fox et al., 2012; Kandel & Kao, 2004). This seems particularly likely in Burkina Faso and Senegal, where so many migrants have little or no formal education. Children in these poorly resourced settings may also face higher demands for their own labor in the household, thus reducing children’s school enrollment where migrants’ relative education levels are also low (Antman 2012, Amuedo-Dorantes, Georges, and Pozo 2010).

But in Kenya, where overall education levels are higher and more high-skilled migration is the norm, we find that migration overall is associated with children’s schooling. Children in migrant-sending households are more likely to be enrolled in school than their peers in non-migrant households. In this case, it may be that observing the returns to education in the form of migration also encourages attachment to schooling on the part of children left-behind. In other words, our results suggest that migration can be positively associated with school enrollment but that this is more likely in contexts where more positively selected migration is the norm or where households in poorly resourced settings can take advantage of sending a more educated migrant out. The results also have implications for concerns about ‘brain drain’ where positively selected international migrants reduce the human capital by removing high skilled migrants from the labor force. Our results suggest that when these migrants leave family members behind in the origin country, growth in human capital – through children’s schooling – continues.

Although our study takes advantage of comparable data in three under-studied contexts of migration, the data also have features that limit our ability to draw firmer conclusions. First, the data are cross-sectional, preventing us from making stronger causal arguments about the relationship between migration and children’s schooling. Clearly, longitudinal data from such under-studied settings would help further investigate these compelling questions (Binci and Giannelli, 2016). Second, we could not identify the relationship of the child to the migrant. This is important because the type of relationship between migrant and child may be associated with the extent to which children benefit from migration. For example, a migrant’s biological child may benefit more from migration than his/her niece or nephew.

Third, we limited our migrant sample to labor migrants who were reported to be living in their current location for five years or less rather than those who left the household in the last five years. We did not use the latter definition because the survey did not collect this information in Kenya. We did, however, tabulate the number of labor migrants whose duration since migration was less than five years with labor migrants who had lived in their current location for the same duration. We found that most labor migrants whose duration since migration was less than five years were also coded as living in the current location for the same amount of time. We chose to focus on recent migrants to maintain proximity between the experience of sending labor migrants and children’s school enrollment.

Fourth, our study focused solely on children reported to be regular members of the household at the time of the survey. Due to the lack of availability of schools, some children, especially those from rural areas, may have been sent to live with other family members, typically in towns, cities, or larger rural communities, to attend school. Our results cannot reflect the extent to which migrant households were more likely than non-migrant households to send children to live with other family members to attend school. However, in Kenya, the survey includes data on children living outside the household so we could examine the relationship between migration variables and current school enrollment for children in and outside the household. The results were similar for both sets of children (not shown). We also cannot observe migration of all household members or cases where children migrate. So our results are applicable to the case of children left behind in sending households.

And, finally, our study found a positive association between migrant selectivity and children’s current school enrollment in migrant-sending households but our regression models could not simultaneously control for migrant’s educational attainment. Fortunately, our analyses were able to control for overall household educational attainment which captures some of the human capital available in children’s homes and is correlated with migrants’ education levels as well. We further attempted to disentangle the relationship by including migrant’s educational attainment and migrant’s relative education score in the same model; however, high levels of collinearity affected the fit of the model and made the coefficient estimates unstable.¹⁶

Conclusion

Our study reveals the importance of taking into account the context and type of migration and of going beyond the use of a dichotomous indicator of household migration status. We examined internal and international migration and children’s current school enrollment across three diverse sending contexts representing various regions of Sub-Saharan Africa and different streams of migration. Similar to previous studies, we obtained an inconsistent relationship between household migration status and children’s schooling. However, once we considered educational selectivity of migrants, controlling for the educational attainment of adults in the household, we observed that migration was more likely to be positively associated with children’s schooling when migrants were more positively selected on education. These findings point to the need to consider the relative resources of sending households (including education of migrants) to understand the association between migration and children’s schooling.