The impact of skilled emigration for
developing countries:
brain drain or brain gain?
An empirical analysis
University of Amsterdam
Faculty of Economics and Business
Amsterdam, September 2014
Charlotte Crooijmans
10579303
MSc Economics; Development Economics
Master Thesis
1
Abstract
The focus of this thesis is to examine the impact of skilled emigration on human capital formation in developing countries. Migration prospects could positively affect the decisions of individuals to invest in education as the returns to schooling are higher abroad and skilled workers have a much higher probability to emigrate than unskilled workers. Using a panel data set which contains migration data by educational level and gender for 166 developing countriesduring 1980-2010, the results show that at an aggregate level there is evidence of a positive impact of skilled emigration prospects on the growth rate of the stock of human capital. In addition, the incentive effect appears to be stronger among men than among women. However, when using instrumental variables in order to correct for possible reverse causality and when introducing changes in the specification by using a different human capital indicator, the estimates show opposing results. Although possibly more in line with theory, there is no significant evidence in favor of a positive impact of skilled emigration on enrollments in tertiary education.
2
Table of Contents
1. INTRODUCTION
3
2. LITERATURE REVIEW
5
2.1 Theoretical literature 5 2.2Empirical literature 63. RESEARCH DESIGN
10
3.1 Panel data 103.2 Theory and assumptions 10
3.3 Data sets 11
3.4 Empirical model and methodology 12
4. EMPIRICAL ANALYSIS
13
4.1 Main results 13 4.2 Robustness checks 15 4.3 Alternative approach 20 4.4 Gender differences 215. DISCUSSION
26
5.1 Robustness of results 26 5.2 Limitations 276. CONCLUSION
28
REFERENCES
30
APPENDICES
32
A1. Countries included in sample 32
3
1. Introduction
This thesis examines the impact of skilled emigration on human capital accumulation in developing countries. The large flow of labour is an important aspect of our integrated global economy which has created both opportunities and challenges for many countries and people in the world (United Nations, 2013). Especially the emigration of skilled labour, usually referred to as brain drain, has gained a lot of interest in the migration debate. The last couple of years, especially since the global economic crisis, major immigration countries like the USA, Canada and Australia have become more and more selective in their immigration policies. Their objective is to only allow highly-skilled workers who are most likely to succeed and offer economic benefits to the country, of which the majority comes from the developing world (OECD, 2013; Docquier and Marfouk, 2006). Skilled migration from developing to developed regions is still rising in absolute levels, not only due to quality-selective immigration policies but also because of increasing population sizes and more educated people in developing countries (Gibson and McKenzie, 2011). This increase in international labour mobility and more complex migration policies have all contributed to more research in this particular area. Evidence on the impact of skilled emigration on source countries has been rather mixed. Traditionally it has been thought that the brain drain was detrimental for developing countries as they already suffer from low levels of human capital (Bhagwati and Hamada, 1974; McDulloch and Yellen, 1977; Miyagiwa, 1991). Developed countries would gain at the expense of developing countries as human capital is moving to where it is already abundant. Since human capital is assumed to be an important driving force to economic growth, the movement of skilled labour would be harmful for the developing world and would increase inequality across countries (Mountford and Rapoport, 2011). Recent studies are less pessimistic as the migration of skilled labour might positively affect the source country through various channels (Docquier, 2006; Docquier and Rapoport, 2012). These feedback effects include remittances, return migration with more skills acquired abroad and the creation of trade networks. Moreover, migration prospects are likely to positively affect the decisions of individuals to invest in education as the returns to schooling are higher abroad and skilled workers have a much higher probability to emigrate than unskilled workers1. If there exists an incentive effect, the net effect of the brain drain on human capital formation and indirectly on economic growth may well be positive. If this incentive effect is larger than the actual emigration of skilled labour, the average education level of those who remain in the country may be higher than it would have been without emigration. In other words, the brain drain could actually result in a brain gain. Due to lack of harmonized international data on migration by country of origin and education level, many contributions on this topic have long remained theoretical.
1
The selective immigration policies in the major immigration countries and the tendency for migrants to positively self-select out of the population mainly explain why the emigration rates are higher for the skilled than for the unskilled.
4
Additionally, gender dimensions of international skilled migration have generally been disregarded in the existing literature. However, the share of skilled women migrating to developed countries has increased over the last periods (Docquier et al., 2008). Skilled female migration could have different effects on human capital formation than skilled male migration as women still face unequal access to tertiary education in the developing regions of the world. They might be less able to respond to the higher returns to education abroad and hence be less able to invest in education.
Taking advantage of a new macro-data set constructed by Brücker et al. (2013) which contains data on international migration by education level and gender, this thesis will try to examine the impact of skilled emigration on human capital formation for developing countries. Most empirical studies conducted so far on this topic have relied on cross-sectional analyses and do not go beyond 2000 due to limited data availability (Beine et al., 2001; Beine et al., 2003; Beine et al., 2008; Di Maria and Lazarova, 2012). This thesis will perform an empirical analysis using panel data as it will focus on 166 developing countriesduring 1980-2010. The central research question is: ‘What is the impact of
the brain drain on human capital formation for developing countries?’. Applying the research
question to a new extended panel data set and distinguishing between male and female skilled migration is exactly the contribution this thesis will make to the existing literature. The findings reveal that at an aggregate level, there is evidence of a positive impact of skilled emigration prospects on the growth rate of the stock of human capital, which is in line with the existing literature. In addition, the incentive effect appears to be stronger among men than among women. However, when using instrumental variable estimation and when introducing changes in the specification by using a different human capital indicator, the results are different. Although possibly more in line with theory, there is no significant evidence in favor of a positive impact of skilled emigration on enrollments in tertiary education.
The remainder of this thesis is organized as follows. Section 2 gives an overview of the existing literature on the brain drain and its consequences, in which a distinction will be made between the theoretical and empirical literature. Section 3 will describe the research design which mainly consists of a description of the data used, the methodology and the empirical model. Section 4 displays the results of the empirical analysis, both overall and when distinguishing between gender. Section 5 of this study will comment on the limitations of this study and will recommend possible areas for future research whereas the final section will answer the research question by concluding on the impact of the brain drain on human capital formation in developing countries.
5
2. Literature review
2.1 Theoretical literature
Several studies have pointed out feedback effects of the brain drain which could affect source countries in a positive way (Docquier, 2006; Docquier and Rapoport, 2012). Educated people are able to earn more abroad which could lead to high remittances. Those remittances sent by migrated skilled people may help overcome liquidity constraints and stimulate investments in human and physical capital at the country of origin. In addition, the return of previously migrated workers who have acquired additional skills abroad could be beneficial to the source country and skilled migrants could contribute to the facilitation of flows of goods, services and knowledge between home and host countries which stimulates trade and foreign direct investment. On the other hand, research has also shown that high skilled people are less likely to send remittances and less likely to return (Docquier and Rapoport, 2012). Evidence on the impact of these channels is therefore rather mixed.
Another strand of the literature, which is more optimistic, has written about the positive effect of the brain drain on human capital, a major driver of economic growth (Barro, 2001). For developing countries, the return to education is likely to be higher abroad due to exogenous inter-country productivity differentials. Hence, migration opportunities increase the expected return to human capital causing more individuals to invest in education. Mountford (1997) describes the mechanism behind this incentive effect by analyzing the interaction between migration and human capital accumulation in a probabilistic model in which the probability of migration depends on the achievement of a given observable educational requirement. This minimal level of education is a necessary condition for migration to a high-wage destination country which is consistent with the quality-selective immigration policies in most developed countries. He concludes that when migration prospects to developed countries are uncertain, the brain drain can have a positive effect on the productivity of the source country, which is an increasing function of the proportion of educated people. The relation between migration prospects and the level of human capital could be described by an inverted U-shape as the net effect of the brain drain on human capital is positive when the brain drain is not too low neither too high. When the emigration rate exceeds a specific threshold, the human capital loss induced by the brain drain increases exponentially. Both Vidal (1998) and Stark et al. (1998) distinguish between a so called ex-ante brain effect, which induces more people to invest in education at home, and an ex-post drain effect2, due to the actual emigration of skilled labour. They argue that the former effect is likely to be bigger than the latter, and hence there is a net gain or a so called beneficial brain drain to the source country.
2
The brain effect is ex-ante as it takes place before emigration occurs whereas the drain effect takes place after people have made the decision to invest in education and is therefore ex-post.
6
2.2 Empirical literature
Due to data constraints, many contributions on this topic have long remained theoretical. Beine et al. (2001) were the first to conduct an empirical study of the direct effect of the brain drain on human capital formation and its indirect effect on growth using a cross-section analysis of 37 developing countries, not paying attention to possible dynamic effects. At their time of writing good quality emigration data did not exist, certainly not when one wants to make a distinction by educational level. In their analysis, Beine et al. (2001) have used gross migration rates as a proxy for the migration of skilled labour. Based on the data used, the authors reject the hypothesis that there is no incentive (or brain) effect. When including an interaction term of the skilled migration rate with a dummy variable that captures very poor countries3, the authors find that the positive impact of the migration of skilled labour on human capital formation is especially large for those low-income countries. In poor countries, the incentives to invest in education are exceptionally low unless considerable outside options are offered to prospective students. Although poorly estimated4, they also find a positive link between human capital formation and growth of gross domestic product (GDP) per capita.
Beine et al. (2003) conduct a similar cross-section analysis but include 50 countries and base their analysis on a more recent data set by Carrington and Detragiache (1998) who computed emigration rates for three educational levels. This data set also suffers from some limitations as the construction is based on several unrealistic assumptions5 and data is only available for 1990. The authors find a positive and significant effect of the brain drain on human capital formation in the source country. In contrast to Beine et al. (2001), they do not find any evidence of heterogeneous effects and hence no greater impact for poor countries. Furthermore, Beine et al. (2003) distinguish between countries for which the incentive effect is positive and for which the effect is negative and compare the growth performance of each individual country would the skilled migration rate be set to zero. This allows the authors to estimate the net effect of the brain drain and to make a distinction between winning and losing countries. They find that winning countries are those with low levels of human capital and low brain drain rates.
3
Being a very poor country is defined as having a value of GDP per capita that is less than 15% of the average GDP per capita in the G7 countries which are Canada, France, Germany, Italy, Japan, the UK and the US (measured at current prices in international dollars).
4
The estimations are not fully controlled for heterogeneity across countries and should be interpreted with caution.
5
This dataset concerns stocks of migrants rather than flows. The immigration stocks in non-US OECD countries do not report the migrant’s educational attainment nor their age. Carrington and Detragiache (1998) assume that immigrants in this non-US OECD area from a given country are distributed across educational categories in a similar way as US immigrants from that country. This is a strong assumption for migrants from developing countries for which the US is not an important destination country. Also, this can be problematic if the US and other destination countries differ substantially in their immigration policies on quality-selection. In addition, estimates of immigration in non-US OECD countries from small countries may be understated since most of these host countries only record immigrants from top ten (or top five) sending countries.
7
Beine et al. (2008) further expand the cross-section analyses by estimating the net effect of the brain drain for 127 countries using recent good quality data on emigration rates by education level constructed by Docquier and Marfouk (2006). Beine et al. (2008) use a beta-convergence model which implies that they test for potential catching-up effects in the level of human capital. Countries with a low level of human capital are assumed to have a higher growth rate of the human capital stock than countries that already have a high level of human capital. Therefore, the authors regress the growth rate of the ex-ante stock of human capital in the population6 on the initial level of human capital and the skilled migration rate, as a proxy for the migration incentives faced by educated individuals. They indeed find evidence of convergence in human capital levels. In addition, doubling skilled emigration prospects increases human capital formation by 5%. The authors also test for possible heterogeneous effects but do not find any significant results. This could be explained by liquidity constraints outweighing the higher incentive effects in poor countries with as result no different effect for these low-income countries. Docquier et al. (2009) tests for the robustness of the estimates of Beine et al. (2008) but concludes that the use of an alternative indicator of human capital investment (youth literacy) or other functional forms does not change the results significantly.
Di Maria and Lazarova (2012) extend the analysis of Beine et al. (2008) by incorporating the fact that migration of skilled labour might not only change the level but also the composition of human capital as people might focus on disciplines for which the migration opportunities are higher. They analyze the impact of migration for 130 developing countries in 1990 and 2000 and come to the conclusion that there are significant positive effects on both the level and composition of human capital. More and more students choose to enroll in higher education with a scientific or technical major. However, this composition effect only takes place in countries close to the world technological frontier. In contrast to this, the existence of an incentive effect diminishes with the level of technological development of the country.
When using observational data, there are several possible sources of endogeneity. One source of concern is the risk of omitted variable bias. There could be additional factors that affect the decision to invest in education, which are also related to migration decisions. According to the existing empirical literature, these include the level of development of the country, the cost of acquiring education, the supply of education, indicators of political tensions and ethnic diversity, remittances and regional dummies. The variables used to serve as a proxy for these factors are respectively GDP per capita, the population density of the country, the percentage of GDP spent on education and the expenditures on higher education as a percentage of total education expenditure, the fraction of the population not speaking the official or most widely used language and the number of political murders per 1000 inhabitants and finally remittances measured as a share of GDP (Beine et al., 2001; Beine et
6
The ex-ante human capital stock describes the proportion of high-skill natives rather than high-skill residents only which means that the human capital stock of prospective emigrants is also included.
8
al., 2003; Beine et al., 2008; Di Maria and Lazarova, 2012). It is important to include these control variables in the regression in order to reduce omitted variable bias. However, even in a multiple regression, the risk of omitted variable bias remains as some control variables are difficult to measure and cannot be included. Also, sometimes one could be ignorant about relevant factors. On the other hand, using more controls in a regression is not necessarily better. Some of the control variables used in the studies described above are actually bad controls. This means that the control variable itself could be a result of the independent variable of interest. Both remittances and the supply of education are likely to be (partly) determined by the skilled emigration rate7. Including them in the regression will probably bias the parameter of interest.
Another source of concern when estimating the impact of the brain drain on human capital formation is the simultaneous causality between emigration of skilled labour and human capital formation. The level of human capital in a country may as well affect the skilled emigration rate. A larger stock of human capital could lead to more emigration due to a reduction in the skill premium on the local labor market compared to foreign ones, for example.All empirical studies described so far try to correct for the possible endogeneity of the skilled emigration rate by using instrumental variables. The validity of an instrument rests on two conditions (Stock and Watson, 2011). First, the instrument should be relevant which means that it should be significantly correlated with the skilled emigration rate so that variation in the instrument is related to variation in the endogenous explanatory variable and second, the instrument should be exogenous which requires that it should be uncorrelated with the error term so that the part of the variation in the endogenous explanatory variable which is captured by the instrument is exogenous.
The two instruments most widely used in the existing empirical literature are the population size of the source country and the stock of migrants in developed countries. The first one is a proxy for immigration quotas which are less binding for smaller countries8 so those countries are likely to be more open to migration. The latter captures the size of the migration network prospective migrants can rely on. Wage differentials, that create the incentive to migrate, and life expectancy, as a proxy for general living conditions, have also been used as instruments but have shown to be correlated with the human capital stock. Although this correlation could work through skilled emigration, the exogeneity of these two instruments is questionable. In addition, lags of the skilled emigration rate have been used as an instrument. However, the lag of the skilled emigration rate is probably not exogenous as it takes time for the incentive effect to occur where the exact time period is uncertain. Hence, the lagged
7
Emigration of labour ensures that there will be remittances and if the incentive effect actually operates (which will be tested empirically in this thesis), skilled emigration will as well affect the expenditures on education. If more people want to invest in education, the supply and hence the expenditures on education are likely to be higher.
8
Immigration policies in many countries are based on both the skill-level of immigrants and a quota system. Destination countries handle a common quota for all countries regardless of the size of the country and hence these quotas are more binding for large countries (Beine et al., 2001).
9
variable could as well affect the human capital stock. Although it is possible to test for the relevance of an instrument, it is not possible to determine if an instrument is truly exogenous which makes it rather difficult to find valid instruments. Using panel data would therefore be a better solution to the possible endogeneity problems associated with observational data and would also lead to more observations. However, due to limited data availability, all empirical studies described above rely on cross-sectional analyses. As said before, they could suffer from the impossibility to capture unobserved heterogeneity between countries and although instrumentation techniques have been used, it is difficult in cross-country settings to detect the exact causality between skilled migration and human capital formation.
Beine et al. (2011) overcome this problem and revisit the impact of skilled emigration on human capital accumulation using panel data for 147 countries during 1972000 with data sampled at a 5-year frequency. They again regress an indicator of the ex-ante human capital formation on the skilled emigration rate and the initial level of human capital and confirm the result that the brain drain exerts a positive impact on human capital, especially in poor countries. In middle- and high-income countries, there is no evidence of an incentive effect and the brain drain unambiguously diminishes the stock of human capital in the source country. In addition, the authors simulate the net effect of the brain drain for low-income countries and find that this effect can be positive if the skilled migration rate does not exceed 20-30%, depending on other country-characteristics.
The existing literature about the impact of the brain drain on human capital formation typically does not distinguish between male or female migration. Women have long been viewed to move as mothers, daughters or partners of male migrants. However, although the absolute number of skilled female migrants is still lower than that of men, women currently exhibit relatively higher skilled emigration rates. This is mainly due to an increase in educational attainment, cultural and social changes in the attitude towards female migration and increased demand for female skilled migrants in healthcare and the domestic sector abroad. The gender gap in skilled migration is strongly and negatively correlated with the gender gap in educational attainment in the developing source country, where women are typically much less educated than men (Docquier et al., 2008). The fact that the number of highly skilled emigrated women is getting much closer to the number of highly skilled emigrated men, together with the fact that in many developing countries there is still unequal access to tertiary education, raises serious concerns about the impact of the female brain drain. Because of their vital role in economic and social development9, the emigration of highly educated women could affect developing countries in a different way (Barro, 2001). Emigration of educated women could increase the risk of negative effects on stocks of female human capital at the country of origin as there are still many barriers to tertiary education for women in developing countries. Although the brain drain might
9
Education of women will, among others, lead to lower fertility rates, lower inequality, more investments in schooling of children and higher economic growth (Barro, 2001; Docquier et al., 2008).
10
cause an incentive to invest in education, women could be less able to do so because of persisting inequalities in access. In addition, Dumont et al. (2007) show a significant and negative impact of the differential emigration rate of educated women as compared to men on key education and health indicators, such as infant mortality, under-5 mortality and the secondary school enrollment rate by gender which is an indicator of human capital. It is therefore important to take the gender dimensions of international migration into consideration.
3. Research design
3.1 Panel data
Most previous studies performed to estimate the impact of the brain drain on human capital formation and growth have relied on cross-section analyses. Though, the analysis in this thesis relies on panel data. Using panel data instead of cross-sectional data enables to control for variables that vary over time but not across countries and eliminates the effect of omitted variables that differ across countries but are constant over time (Stock and Watson, 2011). In other words, it controls for heterogeneity. Time-series and cross-sectional studies, not controlling for this heterogeneity, run the risk of obtaining biased results. In addition, using panel data leads to more observations and with more informative data one can find more reliable estimates. Although using panel data has many clear advantages over cross-sectional data, there will also be coverage problems. In the sample used in the analysis in this thesis, it is inevitable that variables are unavailable for some countries or in some years.
3.2 Theory and assumptions
The analysis in this thesis relies on several assumptions, as do the analyses in the existing literature about the brain drain. First of all, it is assumed that labor is heterogeneous and that education is a discrete rather than continuous variable. This means that individuals are either skilled (obtained at least tertiary education) or non-skilled. In addition, all skilled migrants are assumed to have a higher probability to emigrate, mainly due to the selective immigration policies in the major immigration countries. This makes the incentive effect likely to occur, as migration prospects could positively affect the decisions of individuals to invest in education since the returns to schooling are higher abroad. Hence, it is assumed that the probability of migration depends on the achievement of higher education rather than a person’s productivity or ability. Though, this achievement of higher education is a necessary but not sufficient condition to be allowed to emigrate. Both internal and external factors of uncertainty, like migration policies or time lags between education and migration decisions, could make it impossible for an educated person to be able to emigrate. This is necessary in order for a beneficial brain drain to take place: if every educated person would be able to emigrate in order to obtain higher returns abroad, a brain gain would hardly occur. In other words, if individuals would
11
have rational expectations, they know that there is a chance that they will not be allowed to emigrate and therefore they will not invest in education which means that the incentive effect that will be estimated in this study does not exist. The assumption that individuals do have irrational expectations is therefore essential in this analysis.
3.3 Data sets
To identify the impact of the brain drain on human capital formation in developing countries for a time-period ranging from 1980 to 2010, data from various sources will be used. A new macro-data set constructed by Brücker et al. (2013) contains data on international migration to 20 OECD countries10 by gender and educational level for 194 developing and developed countries11 from 1980 to 2010 in 5-year intervals. Reliable data is obtained by harmonizing national censuses and population registers statistics from the destination countries. Nevertheless, some imputation of missing data is inevitable. This data set focuses on South-North and North-North migration as over 90% of the skilled migrants live in one of the OECD countries. A distinction is made between three levels of education namely low-skilled, medium-skilled and high-skilled (obtained at least tertiary education). The skilled emigration rate of a country is defined as the number of high-skilled emigrants over the originally high-skilled population.
Several human capital indicators will be used as dependent variables in order to investigate if the estimates are robust across different specifications. Barro and Lee (2010) have constructed a data set that captures the percentage of the population with tertiary education. This variable is needed to calculate the ex-ante stock of human capital in the population (including both residents and migrants), used as the main dependent variable in all existing empirical studies about the brain drain12. In addition, the data set of Barro and Lee (2010) contains the average years of tertiary schooling attained in the population. Both variables are disaggregated by gender for 146 countries in the world from 1950 to 2010 in 5-year intervals. Another dependent variable that would be helpful in order to examine the incentive effect, is the number of students that enroll in higher education as a proxy for investments in higher education. The World Bank Database does contain such a variable, namely total enrollment in tertiary education, expressed as a percentage of the total population of the five-year age group following on from secondary school leaving (regardless of age). Other country-specific
10
The 20 OECD destination countries are Australia, Austria, Canada, Chile, Denmark, Finland, France, Germany, Greece, Ireland, Luxembourg, Netherlands, New Zealand, Norway, Portugal, Spain, Sweden, Switzerland, the United Kingdom, and the United States.
11
Description of all countries in the sample can be found in Appendix A1.
12The ex-ante stock of human capital is clearly not observable but is calculated by the following formula:
where is the skilled emigration rate and is the ex-post stock of human capital, once migration is netted
12
variables which will be included as controls in the regression will also come from the World Bank Database.
3.4 Empirical model and methodology
The analysis in this thesis covers 166 developing countries13 during 1980-2010 in 5-year intervals, so for 7 periods in time. As this study will primarily test the presence of a beneficial brain drain in the long run and not specifically human capital dynamics in the short run, there is not necessarily a need to use a dynamic panel data model. However, in order to examine if the estimates are robust across different specifications and to compare the results with the dynamic panel data analysis of Beine et al. (2011), the analysis in this thesis will rely on both a static model (1) and a dynamic panel data model (2). The regressions that will be tested empirically in this study, both considering all persons and for males and females separately, are based on the empirical studies conducted so far and will have the following functional forms (Beine et al. 2001; Beine et al. 2003; Beine et al. 2008; Beine et al. 2011; Di Maria and Lazarova 2012):
The static fixed-effects regression model (1) includes both time-fixed ( ) and country-fixed effects ( ). These country-fixed effects account for time-invariant country-specific factors that could be related to migration decisions, such as the region in which a country is situated, education policies, governance or ethnic diversity, which are all assumed to remain rather stable over time. On the other hand, the time-fixed effects capture the impact of common shocks across countries. The dependent variable in this model is a measure of the amount of human capital in the population. It will be
represented by various indicators, in order to be able to compare the results found in this thesis with results obtained in previous studies about the brain drain and to investigate if the estimates are robust across different specifications. The dependent variable will be regressed on the log of the skilled emigration rate . This brain drain rate is a proxy of skilled workers’ probability to migrate. Control variables are captured by the vector and vary both over time and across countries. These primarily include the level of development of the country and the cost of acquiring education, measured by population density, as they may affect the decision to invest in education and could be related to migration decisions. It is necessary to take into account that the skilled migration rate is not completely exogenous as there could be risk of reverse causality. In order to correct for this possible endogeneity, not only ordinary least squares (OLS) but also instrumental variable (IV) estimation will be carried out. The instruments used should both be relevant and exogenous. The population size of
13
13
the source country and emigration rates for the population as a whole (no matter education level) are two such instruments used in the existing empirical analyses about the brain drain14. The first one is a proxy for immigration quotas which are less binding for smaller countries which makes those countries more likely to be open to migration. The latter captures the size of the migration network prospective skilled migrants can rely on. Both instruments are expected not to directly impact the amount of human capital in the country of origin. The dynamic model (2) regresses the human capital variable on similar explanatory variables. In addition, the initial level of this dependent variable is included in the regression to test for potential catching-up effects. Since this thesis tries to estimate the impact of skilled migration on human capital formation, the main parameters to be estimated are
and .
There are several regression techniques possible when estimating panel data models, depending on the specification and data used. However, none of them will be optimal when estimating dynamic equations (Islam, 1995). The use of fixed effects regression or autoregressive models leads to inconsistency of the estimates, especially when the number of periods is large. Although the ratio of the number of countries to the number of time periods may suggest that this bias will be rather small in the analysis in this thesis, generalized method of moments (GMM) regression would be a good alternative. This method overcomes some of the problems associated with fixed-effects regressions. Though, GMM methods could also suffer from significant small sample bias. Hence, it would be good to compare the results across different estimation methods, including a fixed-effects model, a random-effects model and GMM estimation, in order to determine if the estimates are robust across different estimation techniques, in line with what Beine et al. (2011) did. However, as the use of GMM is beyond the scope of this thesis and random-effects estimates might be invalid15, the estimation method that will mainly be used in this analysis is fixed-effects regression.
4. Empirical analysis
4.1 Main results
It is interesting to compare the results found in the panel data analysis of Beine et al. (2011) with the results in this analysis, when including a different number of countries and expanding the time horizon. The authors analyze both the dynamics of human capital accumulation and the role of migration of skilled workers on human capital formation. The empirical approach taken by Beine et al. (2011) is similar to the dynamic model (2) from Section 3.4 and is defined by the following equation:
14
It is important to note that the use of IV estimation requires the choice of instruments that vary both across countries and across time as the impact of time-invariant variables cannot be jointly estimated with fixed individual effects (Islam, 1995).
15
As explained by Islam (1995), random-effect estimates might be invalid because the random country-specific effects are probably correlated with the explanatory variables and/or instruments. Fixed effects automatically control for this.
14
(3)
The average annual growth rate of the ex-ante human capital stock (as data is also measured in 5-year intervals) is regressed on the skilled migration rate and the initial level of human capital, using various estimation methods. However, the analysis in this thesis will only replicate the fixed-effects, random-effects and IV estimation with data from a different sample: covering only 121 developing countries but for the period 1980-2010 whereas the sample used by Beine et al. (2011) included 147 developing and developed countries during the period 1975-2000. The results, when using the empirical model (3) from Beine et al. (2011), are replicated in Table 1 (Columns 1-4). Column 1 reports the OLS estimates when including country-fixed effects whereas both country- and time-fixed effects are included in Column 2. Column 3 gives the results for random effects estimation and Column 4 reports the estimates when using instrumental variables, including both country- and year-fixed effects, to account for the possible endogeneity of the skilled migration rate. The instrument used is the lagged value of the skilled migration rate, according to the panel data analysis of Beine et al. (2011). First stage regressions confirm that a lag of the skilled emigration rate is a strong predictor of the current skilled emigration rate.
As can be seen from Table 1, the results suggest that the impact of skilled emigration on human capital formation is positive and highly significant when using fixed-effects OLS estimation. This means that there is evidence in favor of the incentive effect. However, this effect is not robust across different estimation methods. Although the random-effects estimates, which are reported in Column 3, are probably invalid because of correlation between the country-specific effects and the skilled emigration rate, the estimates of the migration rate found in Column 4 when using IV estimation are not significant. This in contrast to the results found by Beine et al. (2011) who report qualitatively similar results across all regression techniques.
Furthermore, the empirical analysis predicts convergence of the human capital indicator, as the coefficient on the initial level of human capital is negative and highly significant. The implied speed of convergence ranges from about 2% to 7% per year across the estimation methods which is a fairly wide range. Evidence of convergence of the human capital indicator is in line with the results of Beine et al. (2011), although they find a speed of convergence towards the country-specific steady state which is slightly higher and slightly more homogeneous across the estimation techniques. It would be interesting to test the findings above with several robustness checks such as assessing the validity of the instruments, changing the specification by adding control variables and allowing for non-linearities in the regression as this has hardly been done by Beine et al. (2011).
15
TABLE 1: Ex-ante human capital formation and migration prospects
Average annual growth rate of the ex-ante human capital stock
OLS Random Effects IV
(1) (2) (3) (4) (5)
Log (initial level of human capital) -0.037*** (0.009) -0.031*** (0.010) -0.018*** (0.003) -0.067*** (0.008) -0.042*** (0.011) Skilled emigration rate 0.002*** 0.002*** -0.000* 0.000 0.000
(0.000) (0.000) (0.000) (0.001) (0.001) Constant 0.028** 0.023** 0.040*** 0.086*** 0.057** (0.012) (0.011) (0.005) (0.020) (0.027) Observations 605 605 605 484 598 R2 0.184 0.225 Number of countries 121 121 121 121 120
Country FE YES YES YES YES
Year FE YES YES YES
F- test (first stage) 33.93 18.50
Note: Robust standard deviations are in parentheses. *** Significant at the 1% level. **Significant at the 5% level. * Significant at the 10% level. Instrument used in Column 4 is a one-period lag of the skilled emigration rate and instruments used in Column 5 are population size and total emigration rate. The first stage F-statistic gives information about the strength of the instruments for the skilled emigration rate. As a rule of thumb, if the F-statistic is less than 10, then the set of instruments is weak. If so, the TSLS estimator will be biased, and statistical inferences (standard errors, hypothesis tests, confidence intervals) can be misleading.
4.2 Robustness checks
The skilled emigration rate is instrumented by its lagged value in Column 4 of Table 1, according to Beine et al. (2011). First stage regressions confirm that this instrument is a strong predictor of the current skilled emigration rate as the F-statistic is above 10. However, lags of the skilled emigration rate are probably not exogenous as the exact time period for the incentive effect to occur is uncertain. Hence, the lagged skilled emigration rate could as well affect human capital formation. For this reason, the authors have used a dynamic panel data model. In such models, one actually can use lagged values of the endogenous explanatory variables as instruments. In order for such instruments to be reliable, all serial correlation in the specification has to be modeled. In this situation, GMM is a better estimation method than fixed-effects regression, as it takes the endogeneity of the lagged dependent variable into account. Therefore, the results in Column 4, when using fixed-effects estimation, are probably biased. However, when using other variables rather than lags of the endogenous explanatory variables as instruments, it is possible to have serial correlation in the error term, as long as the instruments are relevant and exogenous and do not correlate with observations from the past. The population size of the source country and emigration rates for the population as a whole (no matter
16
education level) are two such variables used in the existing cross-sectional analyses about the brain drain. First stage regressions confirm that both variables are strong predictors of the skilled emigration rate as the F-statistic is above 10. Column 5 from Table 1 shows that, when using these two different instruments, although there is again evidence of convergence in a magnitude similar to all other regressions, the results do not lend significant support in favor of the incentive effect.
Table 2 shows alternative specifications of the model when using OLS fixed-effects estimation. The main purpose of this analysis is to test the presence of a beneficial brain drain in the long run and not specifically human capital dynamics in the short run. The introduction of the dynamic component, the initial human capital indicator, is therefore not necessary and this initial level of ex-ante human capital might be endogenous and could bias the results. However, Column 2 in Table 2 shows that once we drop this variable from the model, the coefficient of the brain drain rate is still significant. The inclusion of other exogenous explanatory variables in Column 3 does not alter the results on the brain drain coefficient either. Although small in magnitude, population density shows to be significant. This serves as a proxy for the cost of acquiring education, as it is likely to reduce distances to school and hence to decrease the opportunity cost of education. The coefficient on GDP per capita is not significant. Although one could expect that this factor is likely to influence the level of human capital in a country, it apparently has no significant effect on the average growth rate of the ex-ante human capital stock.
Moreover, it is important to note that the relation between human capital formation and migration prospects might differ across countries. The size of the incentive effect might depend on the level of development of a country. Low-income countries have low incentives to invest in education unless considerable outside options are offered to prospective students, which might be the case when skilled emigration rates are high. Therefore Column 4 in Table 2 includes an interaction term of the skilled migration rate and a dummy variable which captures the level of development of the country16. The coefficient on the interaction term takes the expected sign as the incentive effect is higher in poor countries, but the estimate is insignificant17. Although incentives to invest in education might be higher when the level of development of a country is low, liquidity constraints are likely to be a problem and could outweigh this incentive effect. The coefficient on the development dummy is negative and significant at a 5% level which means that in low-income countries, investments in human capital are lower, in line with what one would expect.
16
This dummy variable takes the value 1 if GDP per capita is below 750 dollars (measured at current prices in international dollars).
17
Even when using alternative threshold values (400 or 1000 dollars) or when using a different development dummy that takes the value 1 if GDP per capita is below 15%, 30% or 40% of the average GDP per capita in the G7 countries, the coefficient on the interaction term remains positive though insignificant.
17
TABLE 2: Different specifications fixed-effects OLS estimation
Note: Robust standard deviations are in parentheses. *** Significant at the 1% level. **Significant at the 5%
level. * Significant at the 10% level. The development dummy takes the value 1 if GDP per capita is below 750 dollars (measured at current prices in international dollars). Results in Column 4 do not change substantially when using alternative threshold values for the dummy variable.
In addition, the theoretical literature suggests the relation between migration prospects and human capital to be an inverted U-shape. This suggests that, at both low levels and high levels of skilled emigration, human capital formation is low whereas at medium levels of skilled emigration, the growth rate of the ex-ante stock of human capital is at its maximum value. This is tested by introducing a quadratic brain drain term in Column 5 of Table 2. Results provide that the coefficient on this term, although very small, is indeed negative and significant at a 5% level and hence in line with theoretical expectations. Regardless of the retained specifications in Table 2, the coefficients on the brain drain rate and the initial level of human capital are highly significant, take the expected sign and are quantitatively rather similar. A one-unit increase in the skilled emigration rate increases the growth rate of the ex-ante human capital stock with (on average) 0.20%.
Average annual growth rate of the ex-ante human capital stock
(1) (2) (3) (4) (5)
Log (initial level of human capital) -0.031*** (0.010) -0.039*** (0.011) -0.035*** (0.010) -0.028*** (0.010) Skilled emigration rate 0.002*** 0.002*** 0.002*** 0.001*** 0.003***
(0.000) (0.000) (0.000) (0.000) (0.001) Population density 0.000*** (0.000) GDP per capita -0.000 (0.000) Development dummy -0.025** (0.010) Skilled emigration rate
* Development dummy
0.001 (0.001) Skilled emigration rate
squared -0.000** (0.000) Constant 0.023** -0.013* 0.024** 0.036*** 0.014 (0.011) (0.007) (0.011) (0.013) (0.012) Observations 605 605 545 605 605 R2 0.225 0.188 0.253 0.238 0.232 Number of countries 121 121 119 121 121
Country FE YES YES YES YES YES
18
Table 3 presents the IV estimates for various alternative specifications, similar to Table 2. A lag of the skilled emigration rate is not used as an instrument as it is expected to have a direct impact on the dependent variable and therefore not to be exogenous. First stage regressions confirm that the population size and the total emigration rate, used as instruments in the regressions below, do explain enough variation in the skilled emigration rate as the F-statistic is above 10 in every regression. Although one might question the validity of the instruments as it is hard to find instruments that are both exogenous and relevant, it can be concluded that there is no significant impact of the brain drain on human capital formation. Though, there is significant evidence of human capital convergence in every alternative specification of the model but in a slightly larger magnitude than when using OLS estimation.
TABLE 3: Different specifications fixed-effects IV estimation
Average annual growth rate of the ex-ante human capital stock
(1) (2) (3) (4) (5)
Log (initial level of human capital) -0.042*** (0.011) -0.057*** (0.012) -0.045*** (0.010) -0.041*** (0.010) Skilled emigration rate 0.000 (0.001) 0.001 (0.001) -0.001 (0.001) 0.000 (0.001) 0.001 (0.004) Population density 0.000** (0.000) GDP per capita -0.000* (0.000) Development dummy -0.030** (0.014) Skilled emigration rate * Development dummy 0.001 (0.001) Skilled emigration rate squared -0.000 (0.000) Constant 0.057** 0.002 0.080** 0.067** 0.053* (0.027) (0.015) (0.031) (0.027) (0.031) Observations 598 598 545 598 598 Number of countries 120 120 119 120 120
Country FE YES YES YES YES YES
Year FE YES YES YES YES YES
F- test (first stage) 18.50 10.11 12.61 14.39/63.65 23.29/9.49
Note: Robust standard deviations are in parentheses. *** Significant at the 1% level. **Significant at the 5% level. * Significant at the 10% level. The development dummy takes the value 1 if GDP per capita is below 750 dollars (measured at current prices in international dollars). Instruments used are population size and total emigration rate. In Column 4 and Column 5, two F-statistics are reported where the first one refers to the strength of the instruments for the skilled emigration rate and the second one to the strength of the instruments for its interaction term or squared term.
19 TABLE 4: Tertiary enrollment and migration prospects
Log (Enrollment in tertiary education)
OLS IV
(1) (2) (3) (4) (5) (6) (7) (8)
Log (Skilled emigration rate) -0.185*** -0.157** -0.037 -0.029 -0.111 -0.079 0.081 0.095
(0.064) (0.079) (0.064) (0.071) (0.070) (0.074) (0.093) (0.094)
Development dummy -0.114 0.137 -0.077 0.233
(0.190) (0.209) (0.223) (0.242)
Log (Skilled emigration rate) * Development dummy -0.061 (0.097) -0.093 (0.071) -0.084 (0.095) -0.136 (0.110) One-period lag of log
(Enrollment in secondary education) 0.584*** (0.198) 0.582*** (0.211) 0.614*** (0.101) 0.622*** (0.104)
Log (GDP per capita) 0.205 0.229**
(0.196) (0.110) Constant 1.772*** 1.789*** -1.814 -0.214 1.612*** 1.625*** -2.361** -0.620 (0.118) (0.136) (1.584) (0.804) (0.159) (0.165) (1.016) (0.476) Observations 663 663 423 435 663 663 423 435 R2 0.613 0.620 0.690 0.684 Number of countries 153 153 133 135 153 153 133 135
Country FE YES YES YES YES YES YES YES YES
Year FE YES YES YES YES YES YES YES YES
F-test (first stage) 69.28 51.64/ 295.66 31.16 26.58/ 177.24
Note: Robust standard deviations are in parentheses. *** Significant at the 1% level. **Significant at the 5% level. * Significant at the 10% level. The development dummy takes the value 1 if GDP per capita is below 750 dollars (measured at current prices in international dollars). OLS estimation in Column 1-4 and IV estimation in Column 5-8. Instruments used are population size and total emigration rate. In Column 6 and Column 8, two F-statistics are reported where the first one refers to the strength of the instruments for the skilled emigration rate and the second one to the strength of the instruments for its interaction term.
20
4.3 Alternative approach
A different approach to test the incentive effect, which would also be more in line with theory, is to choose enrollments in tertiary education as dependent variable rather than human capital formation measured by the growth of the fraction of the population with higher education. The higher the probability of migration for skilled individuals, the higher the enrollment rate for tertiary education should be. Contrary to human capital formation which happens over time, the effect of migration prospects on enrollment is expected to occur in the same time-period so the static Model (1) from Section 3.4 would likely be the right empirical model to use as it measures the long term relation between migration prospects and enrollment in tertiary education. Table 4 shows the results. Surprisingly, one can see that there is significant evidence against the incentive effect when estimating the empirical model by ordinary least squares in Column 1. Skilled migration prospects decrease rather than increase enrollments in tertiary education, contradicting both theory and the results found before. When using instrumental variables in Column 5, no significant impact of the skilled emigration rate is found.
It is interesting to examine these findings, for instance by taking possible heterogeneous effects into account and by adding control variables to the regression (Columns 2-5 and Column 6-8), similar to what has been done with the results from the empirical model used by Beine et al. (2011). As expected, tertiary enrollment is strongly determined by the lagged enrollment rate at the lower level. However, most importantly, across all estimation methods and specifications, there is no significant evidence in favor of the incentive effect.
These results are rather confusing as they are not in line with the results found before. Whereas the growth of the ex-ante stock of human capital is positively affected by skilled migration prospects, using tertiary school enrollment rates as dependent variable does not lead to support for the incentive effect. Migration prospects even tend to decrease rather than increase enrollment in tertiary education as the only significant estimates found, describe a negative impact of the skilled emigration rate. One explanation could lie in the data on migration used in this analysis. Although Brücker et al. (2013) only include foreign-born immigrants aged over 25 in their dataset, as this group is less likely to contain students who temporarily emigrate for educational purposes, they do not control for age of entry. Hence, students who have acquired their tertiary education abroad could be included in the skilled emigration figures and do not count as a loss of human capital for the country of origin. This could explain the absence of a positive effect of skilled migration prospects on tertiary enrollment at the country of origin. Though, an increase in skilled migration probabilities could then generate incentives for individuals to enroll in secondary schooling rather than tertiary schooling, in order to be able to migrate to a foreign county and obtain tertiary education abroad. However, when using
21
enrollment in secondary education as dependent variable18, the impact of skilled migration prospects remains insignificant. Hence, this explanation is unlikely to hold. On the other hand, human capital formation as measured by the change in the proportion of highly educated among both residents and potential migrants, accounts for return migration of skilled emigrants whereas school enrollment indicators do not take this into account. This could be another explanation for the contradicting results found above.
4.4 Gender differences
The results in favor of the incentive effect found by all authors who previously have researched the impact of the brain drain on human capital information in developing countries, have shown not to be very robust across different estimation methods or specifications, especially not when using another, more appropriate dependent variable. Nonetheless, it would still be interesting to see if the results would be different when looking at the effect of the emigration of skilled individuals for females and males separately. Women currently exhibit relatively higher skilled emigration rates than men mainly due to cultural and social changes in the attitude towards female migration and increased demand for female skilled migrants in some specific sectors abroad. The data in the sample, used for analysis in this thesis, confirms this observation as can be seen from Table 5. For the period 1980-2010, average brain drain rates are higher for women than for men as 24.3% of the skilled women emigrates compared to 20.1% of the skilled men. In addition, the gender gap in skilled migration is not representative of the gender gap in educational attainment in the developing source country as the values for the percentage of women with tertiary education or completed tertiary education and the average years of tertiary education attained by females are all lower than the values for similar male variables.
TABLE 5: Descriptive statistics by gender
(1)
(2)
(3)
(4)
GENDER Skilled emigration rate Percentage of population with tertiary schooling Percentage of population who completed tertiary schooling Average years of tertiary schooling attained in population Males 20.07 8.12 4.37 0.25 Females 24.26 7.09 3.71 0.22Note: This table describes the gender differences in the mean of the skilled emigration rate and several human
capital indicators across all countries and time periods in the data set.
18
22
TABLE 6: Gender differences in ex-ante human capital formation and migration prospects (OLS)
Average annual growth rate of the ex-ante human capital stock
WOMEN MEN
(1) (2) (3) (4) (5) (6) (7) (8)
Log (initial level of human capital) 0.068*** (0.009) 0.071*** (0.010) 0.071*** (0.009) 0.075*** (0.010) 0.089*** (0.011) 0.092*** (0.011) 0.089*** (0.011) 0.097*** (0.012) Skilled emigration rate 0.003*** 0.003*** 0.003*** 0.006*** 0.004*** 0.004*** 0.004*** 0.007***
(0.000) (0.000) (0.000) (0.001) (0.001) (0.001) (0.001) (0.001)
Development dummy 0.036** 0.007
(0.014) (0.010)
Skilled emigration rate * Development dummy -0.001** (0.000) 0.000 (0.000) Population density 0.000** 0.000** (0.000) (0.000) GDP per capita 0.000 0.000 (0.000) (0.000)
Skilled emigration rate squared -0.000*** (0.000) -0.000*** (0.000) Constant -0.075*** -0.078*** -0.089*** -0.110*** -0.151*** -0.155*** -0.153*** -0.178*** (0.015) (0.017) (0.016) (0.020) (0.026) (0.028) (0.028) (0.030) Observations 605 566 605 605 605 566 605 605 R2 0.360 0.360 0.366 0.419 0.470 0.485 0.472 0.515 Number of countries 121 119 121 121 121 119 121 121
Country FE YES YES YES YES YES YES YES YES
Year FE YES YES YES YES YES YES YES YES
Note: Robust standard deviations are in parentheses. *** Significant at the 1% level. **Significant at the 5% level. * Significant at the 10% level. The development dummy
23
Educated women are vital for economic and social development and their emigration could therefore more likely affect developing countries in a negative way. One could expect that the incentive effect is less likely to occur for women due to barriers to higher education still present for them in many developing countries (Barro, 2001). This means that the impact of skilled emigration on human capital formation might be different across men and women. This will be tested by replicating some of the regressions conducted before for females and males separately. Table 6 and Table 7 describe the results when regressing the average annual growth rate of the ex-ante human capital stock on the skilled migration rate, using respectively OLS and IV. Columns 1-4 describe the impact of skilled female emigration on human capital formation of women over time whereas Columns 5-8 do the same for men, using various specifications. One can see from Table 6 that when using the same empirical model as Beine et al. (2011), there is again significant evidence in favor of the incentive effect across all different specifications. Additionally, as expected, the estimates for the skilled emigration rate are higher for men than for women irrespective of the specification used. Moreover, for women, the incentive effect seems to depend on the level of development of the country. In poor countries, where barriers to female education are possibly higher, the incentive effect is less likely to occur among women. Surprisingly, there is no evidence of human capital convergence in contrast to what was found for the population as a whole.
The results are different when using IV estimation in Table 7. First stage regressions show that the instruments used for the female skilled emigration rate in Column 1-4 are weak and explain little of the variation in this endogenous variable. The estimators are therefore biased and t-statistics and confidence intervals are unreliable. On the contrary, the results found for men do suggest that there is evidence of a positive impact of migration prospects on human capital formation. As the estimates for women are unreliable, it is hard to compare gender differences on the incentive effect when trying to correct for the risk of reverse causality by using IV estimation.
Is it interesting to test if the results change when using enrollments in tertiary education as dependent variable. Table 8 shows the results of gender differences in tertiary enrollment and migration prospects using both OLS and IV. In line with the results found for the population as a whole, there is again no significant evidence in favor of the incentive effect when using enrollment in tertiary education as dependent variable of interest. The results do not differ substantially for men and women, so the hypothesis that women will have lower tertiary enrollment rates as a result of skilled emigration prospects cannot be rejected.
24
TABLE 7: Gender differences in ex-ante human capital formation and migration prospects (IV)
Average annual growth rate of the ex-ante human capital stock
WOMEN MEN
(1) (2) (3) (4) (5) (6) (7) (8)
Log (initial level of human capital) 0.085*** (0.014) 0.095*** (0.017) 0.085*** (0.012) 0.082*** (0.025) 0.077*** (0.009) 0.082*** (0.009) 0.078*** (0.009) 0.080*** (0.024)
Skilled emigration rate 0.006*** 0.007*** 0.006*** 0.024 0.003*** 0.003*** 0.003*** 0.005
(0.002) (0.003) (0.002) (0.020) (0.001) (0.001) (0.001) (0.010)
Development dummy 0.055*** 0.023*
(0.019) (0.013)
Skilled emigration rate * Development dummy -0.002** (0.001) -0.001 (0.001) Population density 0.000* 0.000* (0.000) (0.000) GDP per capita 0.000 -0.000 (0.000) (0.000)
Skilled emigration rate squared -0.000 (0.000) -0.000 (0.000) Constant -0.154*** -0.200*** -0.155*** -0.157 -0.115*** -0.122*** -0.119*** -0.125 (0.054) (0.075) (0.049) (0.095) (0.024) (0.025) (0.024) (0.079) Observations 599 566 599 599 599 566 599 599 Number of countries 120 119 120 120 120 119 120 120
Country FE YES YES YES YES YES YES YES YES
Year FE YES YES YES YES YES YES YES YES
F-test (first stage) 7.26 6.39 5.32/58.36 9.48/3.39 28.28 20.29 19.87/74.79 33.67/11.40
Note: Robust standard deviations are in parentheses. *** Significant at the 1% level. **Significant at the 5% level. * Significant at the 10% level. The development dummy
takes the value 1 if GDP per capita is below 750 dollars (measured at current prices in international dollars). Instruments used in Column 1-4 are female population size and total emigration rate among females whereas the instruments used in Column 5-8 are male population size and total emigration rate among males. In Column 3, 4, 7 and 8, two F-statistics are reported where the first one refers to the strength of the instruments for the skilled emigration rate and the second one to the strength of the instruments for its interaction term or squared term.
25 TABLE 8: Gender differences in tertiary enrollment and migration prospects
Log (Enrollment in tertiary education)
OLS IV
WOMEN MEN WOMEN MEN
(1) (2) (3) (4) (5) (6) (7) (8)
Log (Skilled emigration rate) -0.116* 0.004 -0.128 -0.006 0.050 0.077 -0.020 0.133
(0.062) (0.080) (0.088) (0.101) (0.081) (0.097) (0.072) (0.095)
Development dummy 0.314 0.272 0.369 0.541*
(0.329) (0.253) (0.406) (0.312)
Log (Skilled emigration rate) * Development dummy -0.125 (0.100) -0.129 (0.085) -0.147 (0.168) -0.264* (0.154) One-period lag of log
(Enrollment in secondary education) 0.601** (0.236) 0.687*** (0.274) 0.617*** (0.128) 0.751*** (0.141) Constant 1.107*** -0.650 1.790*** -0.662 0.701*** -0.874 1.556*** -1.197* (0.174) (0.926) (0.163) (1.075) (0.213) (0.537) (0.173) (0.614) Observations 528 313 528 313 524 313 524 313 R2 0.652 0.714 0.561 0.624 Number of countries 145 111 145 111 142 111 142 111
Country FE YES YES YES YES YES YES YES YES
Year FE YES YES YES YES YES YES YES YES
F-test (first stage) 93.28 58.73/200.96 93.64 55.16/ 114.94
Note: Robust standard deviations are in parentheses. *** Significant at the 1% level. **Significant at the 5% level. * Significant at the 10% level. The development dummy takes the value 1 if GDP per capita is below 750 dollars (measured at current prices in international dollars). Instruments used in Column 5 and 6 are female population size and total emigration rate among females whereas the instruments used in Column 7 and 8 are male population size and total emigration rate among males. In Column 6 and 8, two F-statistics are reported where the first one refers to the strength of the instruments for the skilled emigration rate and the second one to the strength of the instruments for its interaction term.
