Secondary education and international labor mobility: evidence from the natural experiment in the Philippines

International labor mobility is a key factor for a well-functioning labor market.

On the one hand, this allows firms in the host country to employ labor at the lowest wage in a geographically broad international market. On the other hand, unemployed workers in the sending country may find employment opportunities in another country.

The direction of the effects of education on labor mobility is theoretically ambiguous and, empirically, not straightforward. The human capital framework argues that an individual moves if the benefit from migration outweighs the cost and, generally, this suggests that education increases international labor mobility. Behind this prediction are three main mechanisms: attaining an extra year of education 1) improves the skills and knowledge that help individuals find alternative employment in a distinct labor market, 2) mitigates credit constraints by increasing borrowing credibility, and 3) reduces the psychological costs associated with international migration through tolerance for new experiences and the acquisition of the language spoken in the host country. On the contrary, education may reduce the propensity of an individual to work abroad. McHenry (2013) argues that additional schooling at a low level enhances local labor market contracts and increases the opportunity cost of moving.

Early studies on economics often used cross-sectional data, reporting a positive association between individual educational attainment and the frequency of regional migration (e.g., Greenwood, 1997). Identifying the causal effects of education on labor mobility is, however, challenging because of the nonrandom assignment of individual educational attainment. Ideally, if educational attainment is randomly assigned to each individual, ordinary least squares (OLS) would provide a consistent estimate of the causal effects of education on the outcomes of interest. In reality, however, individuals endogenously decide how many years to stay in school. Thus, if unobserved factors are correlated with both educational attainment and the decision to migrate overseas, a spurious relationship between these two factors may arise.

For example, as unobserved parental experience in international migration positively affects both educational attainment and the preference for international migration of their offspring, the ordinary least squares (OLS) estimate will be upward biased. Other possible sources of omitted variable bias include individual innate ability, career aspirations, and parents’ preference for investment.

For this purpose, the present study examines the 1988 free secondary education reform in the Philippines to investigate the effects of education on international labor mobility. This reform eliminated the tuition fee of public secondary schools to expand access to secondary education. Since the reform came into effect in the academic year of 1988, there is cross-cohort discontinuity in exposure to the reform. Moreover, since the reform was announced after pivotal cohorts were born, exposure to the reform around the cutoff point is plausibly random. Indeed, Sakellariou (2006) studies the same reform in the Philippines to estimate the rates of returns to education, and this has been shown to disproportionately increase the educational attainment of the cohorts born in 1975 or later. By using this plausibly exogenous variation, we study the effects of education on international migration by employing the two-stage least squares (2SLS) approach. The Labor Force Survey (LFS) of 1995–2014 provides information on whether an individual works abroad as well as on his/her educational attainment and birth year.

The results using fuzzy regression discontinuity design (RDD) suggest that attaining additional years of education reduced the probability of working abroad for men in the Philippines. The OLS estimates, however, show a significant and positive association between two factors, suggesting that existing literature using cross-sectional data is likely to be upward biased. Our results are robust to using different kernel weights. In summary, we, for the first time, show the causal association between individual educational attainment and international migration in the context of developing countries.

The present study contributes to the large literature on labor market returns to education. Quasiexperimental studies often use compulsory education reforms as a natural experiment and mostly report that an extra year of education is associated with higher labor market earnings.

In developed countries, recent studies use compulsory education reforms as a natural experiment and report that increased education raises wages in the United Kingdom (Oreopoulos, 2006; Devereux and Hart, 2010; Dolton and Sandi, 2017), France (Grenet, 2013), and Australia (Leigh and Ryan, 2008). As for evidence from developing countries, Aydemir and Kirdar (2017) find that wage-related returns to education in Turkey are 7%–8% for women and 2%–2.5% for men.

By contrast, evidence from the United States is mixed. On the one hand, Malamud and Wozniak (2012) use the variation in college education due to draft-avoiding behavior among the Vietnam War generation and find that attaining a college degree is positively associated with the likelihood that an individual lives in a state other than his/her birth state. On the other hand, McHenry (2013) exploits the variation in education at a lower level by using the cross-state variation in the CEL reform in the United States and finds that individuals are less likely to migrate across states as they are educated. Overall, existing evidence usually examines the effects of education on labor mobility within a country, and empirical evidence is inconclusive.

The only exception is the study by Fenoll and Kuehn (2017), who examine the association between education and international migration in Europe by using the cross-country variation in the timing of the CEL reform from 1950 to 1990. Consistent with McHenry (2013) but inconsistent with other studies, they find that an additional year of education reduces the number of individuals who migrate across the country by 9%. However, to the best of the authors’ knowledge, evidence on the causal effects of education on international labor mobility from developing countries is lacking. This situation is unfortunate because the developing world sends the largest number of international migrants. The present study, therefore, provides the first evidence on the causal effects of education on the propensity of working abroad in the context of South–North migration from a developing country.

This study also contributes to the growing literature on the effects of free secondary education programs in developing countries. By examining free secondary education reforms in Africa in the 21st century,

The reforms started in South Africa in 2007, the Gambia in 2001–2004, Uganda in 2007, Kenya in 2008, Tanzania in 2016, and Rwanda in 2007.

Finally, our study contributes to the literature on the determinants of international migration. The determinants of migration have been of interest since Borjas (1987) investigated the selection of immigrants from developing countries to developed countries. A series of papers have studied the factors related to international migration, such as the wage distribution and educational attainment (Borjas, 1987; Chiquiar and Hanson, 2005; Docquier and Marfouk, 2006), migration networks in the host country (McKenzie and Rapoport, 2011), and remittances (Gibson and McKenzie, 2011; Rapoport and Docquier, 2006; Yang, 2008, 2011). Nonetheless, evidence on the causal impact of education on the migration decision in developing countries is scarce. Our study thus uncovers the condition surrounding labor mobility and education attainment in a developing country, suggesting that higher education discourages international migration from low-income countries.

The remainder of this paper is organized as follows. Section 2 provides background information on international migration and the 1988 education reform in the Philippines. Section 3 discusses the data. Section 4 reports the empirical strategy, and Section 5 describes the results. Finally, Section 6 concludes the paper.

The data that we used in this study were obtained from the 1995–2014 waves of the LFS (Labor Force, 2014), a quarterly nationally representative repeated cross-sectional survey conducted by the National Statistics Office of the Philippines. We used the across-cohort variation in the exposure to the 1988 reform as a source of the exogenous variation in individual educational attainment to study whether education improves labor mobility. Ideally, we would measure the birth year of each individual directly from the LFS data set. However, such information is lacking in our primary data set. Thus, we imputed the year of birth information from the age of the respondent at the time of the interview.

Although the LFS is conducted in January, April, July, and October every year, to avoid measurement errors in our instrument, we only used data collected in January. By doing so, we are most likely to identify the correct year of birth.

Our main outcome was a dummy variable indicating whether the individual worked abroad or in the domestic labor market at the time of the survey. As the present study is interested in the effects of education on workers’ locations rather than on whether they work or not, the sample was restricted to individuals in the labor force. The main advantage of using the LFS is that it provides information on employment status in both local and international labor markets. This information allows us to identify the year of birth and educational attainment of both overseas and domestic workers in the same data set.

Figure 3 illustrates the reduced-form relationship between our treatment and the propensity of working abroad. The vertical axis shows the estimated coefficients (symbols) and confidence intervals (bars) of birth-year dummies from a regression of indicator for working overseas on birth-year dummies, as well as age fixed effects. Figure 3 also shows the estimate of global second-order polynomials fitted to these data simply for exploratory visual aid; the statistical inference in the main analysis uses local linear regressions with respondents’ age controls, which are discussed in the Section 4. The upper panel represents a reduction in the share of respondents who work abroad around the cutoff point for males. The discontinuity in educational attainment around the cutoff point in Figure 2 suggests that education may have a causal impact on international labor mobility, particularly among men in the Philippines. In contrast, the impact of education for females in terms of international labor mobility is somewhat unclear, and we will formally test this point in the main analysis.

Figure 3

The data were collected by using a questionnaire-based interview; the interviewees, mostly household heads, answered the questions for all household members. An individual was defined as an overseas worker if he/she worked abroad at the time of the interview. As a result, this study defines only temporary overseas workers who have left their families behind in the original country as international migrant workers. There may be some concern that examining the effects of education on permanent migration is empirically more relevant than its effects on temporal migration because the status of permanent migration may help identify the impacts of educational attainment on lifetime income. However, we argue that being a temporary migrant is relevant in the context of the North–South migration from developing countries. Dustmann and Gorlach (2016) clarify that the behaviors and economic phenomena of temporary migrants differ from those of permanent migrants. For instance, migrants expecting a shorter stay in the host country send more remittances to their home country. Remittance is an important income source for supporting the family left behind, particularly in developing countries; therefore, examining the determinants of being a temporary overseas worker is at least as important as is the case of permanent migrants in developing countries.

Thus, to simplify the interpretation of the results, our primary explanatory variable is “continuous years of education”. For this purpose, ideally, LFS has a direct measure of completed years of education. One limitation of the LFS, however, is that, among the 1995–2014 waves, a continuous indicator (i.e., years of schooling) is available only in the 2012–2014 waves. By contrast, those in the 1995–2011 waves coded the highest grade completed as seven categorical levels: no grade completed, elementary undergraduate, elementary graduate, high school undergraduate, high school graduate, college undergraduate, and college graduate. To have a sufficient number of observations and high statistical power, the main analysis in this study uses data from 1995 to 2014. Therefore, we recategorize educational attainment at each education level

These are “0” for no grade completed, “3” for under elementary education, “6” for elementary education completed, “8” for under high school, “10” for high school education completed, “12” for under college, and “14” for college education completed.

Another advantage of using the LFS is that the data cover >130,000 respondents per wave over a decade. This setting allows us to identify the impact of a change in educational attainment on international migration after controlling for the flexible time trend (i.e., survey year fixed effects). Including the survey year fixed effects in the analysis allows us to control for the unobserved year-specific factors that affect both demand for and the supply of international labor. For instance, employers in the host country may be reluctant to hire additional workers from other countries when they face financial shocks. Such cross-year confounding factors that affect all the countries at the same time are controlled for by using the survey year fixed effects. In summary, we examine whether individuals are more likely to work abroad as they are educated within the same survey year by including the survey year fixed effects.

Migration studies often lack information on illegal migrants since the official data in host countries do not recognize them. However, we are likely to incorporate illegal migrants as well as legal migrants because the family member of the overseas worker living in the sending countries provides the information in the LFS. This situation is another advantage of this study, compared with existing studies.

Our study is interested in the extent to which educational attainment changes the propensity of working abroad in a developing country. Therefore, we consider the following causal model:

where Y_ijc indicates the outcome variable, which takes the value of “1” if the individual i, born in year j, observed in survey c works in a foreign country; and the value “0” if he/she participates in the local labor market (including employed and unemployed people). The term S_ijc indicates the continuous years of schooling, and α₁ represents the impacts of schooling on the decision to work in another country. The term ε_ijc is the idiosyncratic error term. λ_c is the survey year fixed effects, which allows us to control for the individual’s age at the time of the interview as well.

The OLS estimates from Equation (1), however, are likely to suffer from bias because of the endogeneity of the observed educational attainment, since this tends to be correlated with unobservable factors that affect an individual’s decision to work overseas. For example, people who have high career aspirations tend to go to school and migrate to other places (Bauernschuster et al., 2014), and an unobserved factor such as this may produce the spurious positive association between educational attainment and the propensity of working abroad. Thus, to address the endogeneity of educational attainment, we use the fuzzy RDD approach and exploit the across-cohort variation in the exposure to the free secondary education reform in 1988 to identify the effect of education on the probability of working overseas.

Our first-stage equation of the fuzzy RD approach using the reform exposure indicator T_j is as follows:

The treatment group consists of those born in 1976 or after, and therefore, T_j is a dummy indicating whether an individual was exposed to the program at age 12 or younger; this dummy takes the value “1” if the individual was born in 1976 or after and takes the value “0” otherwise. Here, f(B_j) and λ_c are polynomial functions of the birth year and survey year fixed effects, respectively. We estimate Equation (2) using nonparametric local linear regression proposed by Calonico et al. (2014a, 2014b) to avoid making strong assumptions about the functional form that describes the relationship between the running variable and potential outcomes. Here, η_ijc is the error term, and the standard errors are clustered by the year of birth to allow individuals in the same cohort to correlate with others in the cohort. To investigate the heterogeneous impacts of the reform on educational attainment for men and women, we perform regression on the equation for each sex separately. Therefore, we obtain different β₁ values for men and women; these are our interests in Equation (2).

The reduced-form equation of the fuzzy RD approach is as follows:

where the coefficient of interest is λ₁, which we expect to be positive if an individual is more likely to work abroad as he/she is educated.

In a fuzzy RD approach, the estimator of the interest is a Wald estimator in which the estimated discontinuity in outcome at the discontinuity point, γ₁ is divided by the corresponding discontinuity in the educational attainment β₁. We choose an optimal bandwidth to minimize the mean squared errors with a triangular kernel weight. To investigate the heterogeneous impacts of educational attainment on international labor mobility for men and women, we also perform regression on Equation (3) for these groups separately.

To consistently estimate the coefficient of interest by using a fuzzy RDD approach, an instrument (i.e., reform exposure indicator) should satisfy two conditions. First, the instrument, the dichotomous indicator for being born in 1976 or later, should explain a sufficient degree of the variation in the endogenous variable (i.e., educational attainment). Since the model statistics for nonparametric RDD have not been developed in the literature, we will report the first-stage estimates to examine whether the empirical evidence supports the assumption.

Second, our instrument needs to affect the decision to work overseas only through its effects on individual years of schooling. We argue that this exclusion restriction is likely to hold in the present study because the reform in 1988 was an unanticipated exogenous shock for the parents of the sample children. There may be some concern, for example, that treated mothers may manipulate the timing of conception and birth so that their children are young enough to benefit from the free secondary education reform. If such nonrandom assignment of the year of birth is empirically observed, the characteristics of the children, parents, and households may be systematically different between children born in 1976 and those born in 1975. This situation is, however, unlikely in our setting because we exploit the as-if-random across-cohort variation regardless of whether children were aged ≥12 years in 1988; thus, parents must have accurately anticipated in which year the reform would be implemented 12 years before it occurred. Therefore, our instrument, namely, the year of birth around the cutoff point, is unlikely to be correlated with ability, motivation, or other unobserved characteristics, and thus the estimated coefficient of educational attainment is likely to represent the effects of education on labor mobility alone. In summary, the identification assumptions in the current study are likely to be satisfied. We, therefore, use the fuzzy RD method to estimate the impacts of schooling on the decision to work overseas.

Under these identification assumptions, the estimated treatment effects of the individual’s education on his/her probability of working abroad are interpreted as Local Average Treatment Effects (LATE). The parameters are average treatment effects among the compliers of the reform (i.e., those who stop their education in the absence of the reform but continue to attend an extra year of schooling if the tuition fee of the secondary education is eliminated). This suggests that the parameter estimated by fuzzy RD is more likely to apply to the subpopulation from the lower socioeconomic households, whose decision on their child’s education is linked to their liquidity constraints.

Relatedly, to consistently estimate the effects of the reform on educational attainment and satisfy the exclusion restriction in the second stage, another education reform around 1988 should not disproportionately affect the treated or control cohorts. Although Sakellariou (2006), who uses the same reform as a natural experiment, does not explicitly mention this point in his study, government secondary schools nationally increased dramatically from 1988 to 1990 in response to the 1987 Constitution (United Nations Educational, Scientific and Cultural Organization [UNESCO], 2009).

Similarly, hundreds of public schools had been constructed since the start of the Aquino administration in 1986.

Some may be concerned, therefore, that the coefficient of interest in the first stage may be overestimated because of the effects of the school construction program. Unlike the free secondary education reform, however, the school construction program is likely to have affected both the cohort born in 1976 or after and that born in 1975 or earlier. Such effects, which differ across cohorts, should, therefore, be controlled for by including the polynomial function of the birth year as a control variable. Therefore, we argue that the variation in educational attainment used to estimate its effects on international labor mobility is derived from the 1988 free secondary education reform. In summary, by using the 1988 reform as a source of exogenous variation in educational attainment, we use the 2SLS approach to identify the effects of educational attainment on international labor mobility and present the results of the first- and second-stage estimations in the following section.

Previous studies have identified mixed effects of individual educational attainment on labor mobility in developed countries. However, we know little about whether individuals in the developing countries become more likely to work abroad as they are educated. This study uses the 1988 free secondary education program in the Philippines to examine whether individuals in developing countries become more likely to work abroad as they are educated. We exploited the across-cohort variation in exposure to the program in the Philippines as a source of exogenous variation in educational attainment. By using the LFS data from 1995 to 2014, we found that additional schooling reduced the propensity of working abroad for males in the Philippines. The fuzzy RD estimates suggest that an extra year of education reduces the likelihood of working abroad by 3.2% points on average for men, whereas it had little impact on women, suggesting that the policy improving access to secondary education does not encourage international labor mobility in the Philippines. The fuzzy RD estimates of the effects of education on labor mobility and the OLS estimates have opposite signs. This finding implies that the positive association between education and migration rates reported in the empirical literature may have been largely overestimated in the context of developing countries.

The findings of this study also contribute to the literature on returns to free secondary education in developing countries in general by showing the long-term effects of the program on labor mobility. Considering that an increasing number of developing countries in sub-Saharan Africa have eliminated secondary school tuition fees since the late 2000s, this region may observe a smaller rate of international labor mobility in the future. In summary, the evidence of this study underscores the importance of considering the consequences of long-term labor mobility when designing secondary education reforms in developing countries.

Finally, we emphasize that the estimated effect of education on international labor mobility in the current study is LATE, and the effects of education at different levels of education may be different. Specifically, what we estimated was the treatment effects for the subpopulation that does not continue its education in the absence of the free secondary education reform but would continue if the tuition fee of secondary education is eliminated. Therefore, the reform that promotes access to college education may, for instance, have different effects on international labor mobility even in developing countries. The current study cannot assess this possibility, and future research should explore this point.