investment and human capital in
Latin America
By
Jeroen van Hunsel
Rijksuniversiteit Groningen (S3031969)
Newcastle University Business School (B170170806)
Msc. Advanced International Business Management & Marketing
Dissertation
Supervisors: Dr. I. Munro and Dr. R.W. De Vries
Word count: 12.058
ABSTRACT
Much research has been conducted on the relationship between FDI inflow and human capital. In general, a positive relationship is found between FDI inflow and human capital. However, some studies found a negative relationship on human capital in developing countries. Hence, not all studies agreed with each other. Prior studies did also not focus on Latin America. Therefore, this study examines the relationship between FDI inflow and human capital in Latin America. A sample of seventeen Latin American countries between 2000 and 2015 is analysed. Results suggest that FDI inflow is positively related to human capital in Latin America. However, a more specific analysis shows a negative relationship between FDI inflow and tertiary schooling in developed Latin American countries. Nevertheless, it can be assumed that FDI inflow has a positive relationship with tertiary and secondary schooling in developing countries in Latin America. When FDI flows from OECD countries, it can be assumed that the relationship between FDI inflow and tertiary schooling weakens for the more developed countries in Latin America.
Key words: Foreign direct investment (FDI), Human Capital, Latin America, tertiary
ACKNOWLEDGEMENTS
First, I would first like to thank my dissertation supervisors dr. De Vries and dr. Munro of the Economics and Business Faculty and Business School at University of
Groningen and Newcastle University. Whenever I had a question, I received a substantiated answer. Answering these questions helped me to move in the right direction.
TABLE OF CONTENT
1. INTRODUCTION 9
2. THEORY 12
2.1 Foreign direct investment ... 12
2.2 Human capital... 13
2.3 FDI and human capital ... 16
2.4 FDI and human capital in subsamples ... 17
2.5 OECD countries FDI and human capital ... 18
2.6 OECD FDI and human capital in subsample countries ... 19
3. METHODOLOGY 22 3.1 Measurement of human capital ... 22
3.9 Instrumental variables ... 28 3.10 Statistical model ... 29 3.11 Estimated method ... 30 3.12 Method assumptions ... 31 4. EMPIRICAL RESULTS 40 4.1 Descriptive statistics ... 40 4.2 Regression results ... 41
LIST OF TABLES
Table 1. HDI index of the sample ... 24
Table 2. Overview variables ... 28
Table 3. Breusch-Pagan test for heteroscedasticity ... 33
Table 4. White's test for heteroscedasticity ... 33
Table 5. Serial correlation ... 39
Table 6. Descriptive statistics ... 41
Table 7. Tertiary and secondary schooling ... 43
Table 8. FDI and human capital in developed countries ... 45
Table 9. FDI and human capital in developing countries... 47
Table 10. FDI inflow from OECD countries and tertiary and secondary schooling ... 49
LIST OF FIGURES
Figure 1. Conceptual Model 21
Figure 2. HDI index 70
Figure 3. Outliers education spending (% of GDP) 83
LIST OF ABBREVIATIONS
FDI Foreign direct investment
OECD Organisation for economic cooporation and development
HDI Human development index
GDP Gross domestic product
1. INTRODUCTION
Latin America has higher inequality levels than any other continent. The Gini Index ranks countries by income inequality and shows that Latin America is the most unequal continent of the world with a Gini coefficient of 48.3, while the world’s average is approximately 40 (range between 16 and 63) (The World Bank, 2017). The average income in Latin American countries varies between $1.654 and $22.707 per capita, while the world’s average is $15.024 per capita (The World Bank, 2016). Approximately 85 of 632 million Latin Americans have a day expenditure of less than $2,5 per day (Tsounta & Osueke, 2014). More than half of the people in this group are children. Children from bottom income families complete on average eight years of education, while children from top income families complete on average ten years of education (World Fund, n.d.). Besides the lower average years of education, children from bottom income families experience a lower quality of education and have a higher probability to drop out of school after completing a few grades (Reimers, 1999). Chile, the highest scoring country of Latin America on education quality, scored 10% lower than the world average in the Programme for International Student Assessment (PISA) (Chafuen, 2014).
governmental education expenditure in public tertiary schooling, while only 2% of all students attend public tertiary schooling (World Fund, n.d.).
Castelló-Climent and Doménech (2014) claim that an increase in human capital decreases the inequality of education. On its turn, FDI is a key driver of human capital and education as it affects the demand and supply of skilled labour (Slaughter, 2002) and increases knowledge and skill development through technology spillovers (Miningou & Tapsoba, 2017). Several studies already investigated the relationship of FDI and human capital (Noorbakhsh, et al., 2001; Wang, 2011; Zhuang, 2013; Azam, et al., 2015; Zhuang, 2017). Most of the studies agreed with the claim that FDI positively influences human capital. However, there is no existing research available on this topic regarding Latin America. Another reason why it is interesting to examine is because of the huge educational backlog with the rest of the world. This study aims to get a deeper understanding of the relationship between FDI and human capital in Latin America. The outcomes of the study conclude whether FDI has a relationship with human capital and if FDI can contribute to converging the growing inequality on human capital in Latin America. These outcomes will form the answer to the main question of the study: What is the relationship between FDI and human capital in Latin America?
The next sections of this thesis contain a review of the theory (ch. 2), methodology (ch. 3), empirical results (ch. 4), discussion (ch. 5) and conclusion (ch. 6).
2. THEORY
In the theory section, existing literature regarding foreign direct investment and human capital is revised. Based on the revised literature, hypotheses are formulated. Finally, all hypotheses are present in the conceptual model.
2.1 Foreign direct investment
FDI is a well-known definition in today’s global economy. It is explained by the Cambridge dictionary as: “money from one country that is put into businesses in another country” (Cambridge Dictionary, sd). According to the Organisation for Economic Cooperation and Development (OECD) (2008), FDI can be explained as a set of capital, technology, management, and entrepreneurship, which permits a business to operate and provide goods and services in a foreign market (Farrell, 2008). FDI is an important vehicle for local enterprise development, which can help to expand the competitive position of the host and the home economy.
more financial banking crises than other countries. Therefore, countries tried to attract FDI in order to foster economic growth.
There are several determinants for a host country to attract FDI (United Nations, 1998). Host country determinants are: policy framework, economic determinants and business facilitation. Within the first determinant, policy framework, it is important for the host country to attract or refuse certain businesses. The host country can weaken or sharpen laws in order to deal with this issue. Secondly, the economic determinant exists of three types of FDI categorized by firm’s motives: market-seeking (or horizontal FDI), resource/asset-seeking, and efficiency seeking (or vertical FDI). The last determinant for a host country is business facilitation, which means that a country takes into account the background and aims of a company (United Nations, 1998).
In order to attract FDI, clear policies are needed that offer benefits for companies to invest in a certain country (Bárcena, et al., 2017). This is currently an important issue for countries in Latin America and plays a role in their attractiveness for companies. After the host country attracted a company to invest, success will be the next challenge. Success of FDI not only depends on the type of investment. Adequate human capital, economic stability, and liberalized markets are also important in order to foster economic growth in Latin America (Bengoa & Sanchez-Robles, 2003).
2.2 Human capital
(Schultz, 1961, p. 3). The second view elaborated the first view and added knowledge and skills to the concept. Different types of obtaining knowledge and skills were concerned; such as compulsory education, postsecondary education, and vocational education. The third view focuses on the production-oriented perspective of human capital. Based on these different conceptions, human capital can be described as: “a synonym of knowledge embedded in all levels such as an individual, an organization and/or a nation” (Kwon, 2009, p. 13). This last view is used in this study, as this is the most complete and recent view.
There are two types of human capital: ‘the human as labour force’ and ‘the human as creator’. The first type of human capital is related to economic added-value. This is generated by the input of financial capital, land, machinery, and labour hours. The second type is related to the purpose of investment through education and training. The ‘human as creator’ is based on knowledge, skills, competences, and experience developed by the constant connection between humans and the environment. The concept of human capital is mainly about the second type (Kwon, 2009). Human capital can be seen as knowledge in a broad meaning because it covers all levels, such as on individual, organisational and national level. It covers the growth of an individual’s wage, firms’ productivity, national economy (Schultz, 1961; Denison, 1962), a firm’s core competences or competitive advantage (Lepak & Snell, 1999), a worker’s productivity in the workplace (Griliches & Regev, 1995; Lucas, 1988) and national economic growth (Romer, 1986). Additionally, with human capital it is easy to implement job-seeking activities (Vinokur, et al., 2000) and to receive relatively high rewards in the internal and external labour market.
their key strategy in skill development in order to compete with other countries and foster economic growth. Skills are a key factor in order to decreasing inequality and encourage social mobility. Investing in human capital is the most effective way to not only promote economic growth, but also to distribute the benefits of growth more fairly (Taylor, 2012). However, the poorest countries counter major challenges. Policy makers admit the critical role of a strong human resource base in addition to other investments that foster productivity and economic growth. Human skills affect people’s lives and economic and social development in many ways; for example, employment rates and the income of employees (Taylor, 2012). Denison (1985) studied the income growth of the USA during 1929 to 1982 and found that one fourth of the growth resulted from an increase in education. Besides these examples, human capital has several other consequences. For example, highly educated people feel that they have a voice and that they are able to make a difference in social and political life. Although, there are still countries in which human capital is high, but highly educated people are unemployed or working in jobs that are below their skills. A low human capital causes contrary results. Less educated people have a higher likelihood for poor health and participate less in community groups (Taylor, 2012).
and training. Especially for developing countries there are challenges in order to be as effective as possible with the available skills. In order to receive the maximum return on investment people should have: “the ability to assess the quality and quantity of the skills available in the population, determine and anticipate the skills required in the labour market, and develop and use those skills effectively in better jobs that lead to better lives” (Taylor, 2012, p. 1).
2.3 FDI and human capital
Both FDI and human capital have a strong relationship with economic growth (Borensztein, et al., 1995; Lucas, 1988). FDI causes technology spillovers (Miningou & Tapsoba, 2017). Technological spillovers are technological benefits that come from research and development transfers between firms without being shared. These are beneficial for economic growth (Sun & Fan, 2016). Human capital affects economic growth through the development of knowledge and the skills of people. Besides influencing economic growth, FDI also influences human capital. The technology spillover which causes economic growth, that is mentioned by Miningou and Tapsoba (2017), consists not only of machinery, equipment, patent rights, and expatriate managers and technicians, but also of the training or education of local employees (Borensztein, et al., 1998; Blomström & Kokko, 2002). Slaughter (2002) even elaborated on this by arguing that through the impact of demand and supply of skilled labour, FDI can help to boost the stock of human capital in host economies.
USA. The majority of research found a positive relationship between FDI and human capital. Only Zhuang (2017) found a negative impact of FDI on secondary schooling in East Asia. Based on these studies the following hypothesis is formulated.
Hypothesis 1. FDI has a positive relationship with tertiary and secondary schooling in Latin America
2.4 FDI and human capital in subsamples
Most studies argue that FDI inflow has a positive relationship with human capital (Gittens & Pilgrim, 2013; Adenutsi, 2010). However, the majority of these studies focussed on developing countries. Borensztein et al. (1998) for example, claim that FDI flows to developing countries encourage economic development by transferring knowledge through multinational enterprises (MNEs). As already mentioned in the previous paragraph, an increase in economic growth and human capital results from technology transfers of FDI. Technology is not transferred automatically. The effectiveness of technology depends on the absorptive capacity of the host country. However, the absorptive capacity is not the same for every country, but is determined by the level of human capital of the host country (Borensztein, et al., 1998). Hence, a higher level of human capital results in a higher absorptive capacity. In general, developed countries have higher levels of human capital than developing countries. To check if FDI has a different effect on developed countries than on developing countries, a distinction between countries is been made.
predominantly in the service sectors, but also in natural resources and manufacturing. This is similar to developing Latin American countries (Bárcena, et al., 2017). In developed countries in East-Asia only one significant estimate is found for secondary schooling, which shows a positive relationship with FDI inflow. In addition, Mughal and Vechiu (2009) also found a negative relationship between FDI and tertiary schooling in developing countries. No other studies that examined the relationship between FDI and human capital in developed countries have been found. Since the majority of studies found a positive effect of FDI on human capital, it can be assumed that FDI has a positive effect on human capital in developed countries (Wang, 2011; Zhuang, 2013; Azam, et al., 2015; Zhuang, 2017). Based on Borensztein et al. (1998), Zhuang (2017), and Mughal and Vechiu (2009) the following hypotheses are formulated.
Hypothesis 2a. FDI has a positive relationship with tertiary and secondary schooling in developed countries in Latin America
Hypothesis 2b. FDI has a negative relationship with tertiary schooling and a positive relationship with secondary schooling in developing countries in Latin America
2.5 OECD countries FDI and human capital
(Hoffman, 2003), which will lead to an increase in human capital. FDI increases the demand for skilled labour through three channels (Kheng, et al., 2017). First, technology transfers from parent MNEs to foreign subsidiaries. FDI brings new production technologies which require well-educated labour to subsidiaries in host economies. Second, the possibility that domestic firms can buy the right to use new technologies from MNEs or an employee can be hired from an MNE. The third and last channel through which FDI increase demand for skilled labour is investment in physical capital. Often physical capital, such as computers and machines, is needed in order to adopt and apply advanced technologies as good as possible. Physical capital in combination with well-educated employees works often complementary.
This study focuses on the OECD countries, because these are all developed countries and have high-income economies (OECD, sd). Zhuang (2017) also uses OECD countries in her study, because it can be assumed that OECD countries request more advanced technologies in business than developing countries. Therefore, a greater positive relationship might exist between FDI from OECD countries and tertiary schooling than between FDI from OECD countries and secondary schooling. In order to test this expectation, the following hypothesis is formulated.
Hypothesis 3. FDI from OECD countries has a greater positive relationship with tertiary schooling than with secondary schooling in Latin America
2.6 OECD FDI and human capital in subsample countries
in order to attract FDI (Lucas, 1990). This indicates that the least developing countries have difficulties with attracting FDI. In contrast, developed countries face in general no problems in attracting FDI as developed countries invest more often in other developed countries (UNCTAD, 2014). Becker (1994) focussed on developed countries and claims that investment in human capital causes higher gains in less developed countries than in high developed countries. Hence, less developed countries gain more from investment in human capital than high developed countries. However, a country needs a minimum level of skills in order to gain from investment in human capital. In addition, Zhuang (2017) claims that FDI from OECD countries has a positive impact on education in developing countries in East Asia, while this does not hold for developed countries. In her study, Zhuang created subsamples to examine if FDI causes different effects on human capital in developed and developing countries in East Asia. In order to make this distinction, this study includes the Human Development Index (HDI) score as moderator. The HDI score is developed by the United Nations and ranks countries based on their economic development. The higher the HDI score, the more developed a country is. Based on studies of UNCTAD (2014), Zhuang (2017) and Becker (1994), the following hypothesis is formulated.
Hypothesis 4. The effect of FDI from OECD countries on tertiary and secondary schooling in Latin American countries is negatively moderated by the HDI score of these countries
Figure 1. Conceptual model
3. METHODOLOGY
This section contains the measurement of human capital, the sample of the study, significance, the variables that are used to test the hypotheses, the statistical model, the estimation method, and the evaluation of method assumptions such as measurement, linearity, heteroscedasticity, normality, outliers, multicollinearity, endogeneity, robustness test, and serial correlation.
3.1 Measurement of human capital
advantages over the just mentioned proxies. Besides, educational attainment is a valid stock measurement. It quantifies the accumulated educational investment in the current labour force (Jones & Chiripanhura, 2010). Educational attainment can be divided into three different types of education: primary, secondary, and tertiary education.
3.2 Sample
The sample consists of the following seventeen countries in Latin America; Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Mexico, Nicaragua, Panama, Paraguay, Peru, Uruguay, and Venezuela. This study consists of a sample bias, because Cuba, Haiti, Honduras and Puerto Rico cannot be included in the sample due to lacking data. The study tests the relationship between FDI and human capital in the time period from 2000 to 2015. Within this time period five-year intervals are used, because of availability of data. As FDI and human capital are expected to be causally related, time lags are used in this study in order to test for this relationship. The following time lags are tested in Hypothesis 1 and 2: 1999-2000, 2004-2005, 2009-2010, and 2014-2015. Based on available data of FDI inflow from OECD countries for Hypothesis 3 and 4, the following two periods are tested: 2004-2005, and 2009-2010. When data is missing, extrapolation and interpolation is used to gather adequate data. However, not in all cases this was valid because of nonlinear data for FDI, FDI OECD, and education expenditure data.
sd). Thus, this index is more than a ranking based on economic development. Therefore, it is a good measurement tool in order to divide countries based on development. The index has values between 0 and 1, with 1 as the highest possible value. Table 1 presents a list with the HDI values per country in Latin America.
In Hypothesis 2a and 2b the sample is divided in two subsamples: developed and developing countries. The developed countries are Chile, Argentina, Uruguay, Panama, Costa Rica, Venezuela, Mexico, and Brazil. These countries have a minimum HDI value of 0.75. If the countries have a lower value they belong to the group of developing countries. The following countries belong to this group: Peru, Ecuador, Colombia, Dominican Republic, Paraguay, El Salvador, Bolivia, and Guatemala.
Table 1. HDI index of the sample
Country HDI Country HDI
Developed countries Developing countries
Chile 0,847 Peru 0,740
Argentina 0,827 Ecuador 0,739
Uruguay 0,795 Colombia 0,727
Panama 0,788 Dominican Republic 0,722
Costa Rica 0,776 Paraguay 0,693
Venezuela 0,767 El Salvador 0,680
Mexico 0,762 Bolivia 0,674
Brazil 0,754 Guatemala 0,640
Source: (United Nations Development Programme, 2016). Note: the complete HDI ranking is present in Appendix I.
originally signed the Convention on the Organization for Economic Co-operation and Development. By signing this convention, a platform emerged that tries to improve the economic and social well-being of people across the world (OECD, sd). Chile and Mexico are the only two Latin American OECD countries.
3.3 Significance
Hypothesis 3 examines if FDI from OECD countries has a greater positive relationship with tertiary schooling than with secondary schooling in Latin America. A greater positive relationship exists when the coefficient of tertiary schooling is larger than the coefficient of secondary schooling.
3.4 Dependent variable
3.5 Independent variables
The independent variables are FDI and OECD FDI. FDI is characterized by the inflows of FDI in Latin America. It is assumed that a higher FDI causes a bigger increase in human capital because FDI involves advanced technologies to the host country, and it increases the demand for skilled labour and higher education (Zhuang, 2017).The FDI inflows are measured in million dollars. The data is from UNCTAD is available between 1985 and 2010 for all Latin American countries. For hypothesis 3 and 4, OECD-FDI is the independent variable. This variable contains all FDI flows of OECD countries to countries in Latin America. UNCTAD provides bilateral FDI statistics between 2002 and 2015 of all countries.
3.6 Control variables
(Zhuang, 2017). Consequently, this results on its turn in the need for higher human capital. Data of domestic investment is provided by the Penn World Table (PWT) and is annually available between 1950 and 2014. PWT provides data for the last control variable real GDP per employee at constant 2011 PPP dollars. This variable controls the economic prosperity per country.
As this study examines a robustness check (further explained in 3.12.8), new control variables are included in this study. These new control variables are: real GDP per capita (in constant 2011 PPP $) and investment of education (as percentage of GDP). The World Bank provides data for both control variables. For real GDP per capita (in constant 2011 PPP $) data is available between 1990 and 2016, and for investment of education (as percentage of GDP) between 1970 and 2016.
3.7 Fixed-effect variables
Fixed variables are variables with known values (Kreft & de Leeuw, 1998). In this study, time and country fixed variables are used and characterized as: t (time) and I (country). Fixed-effect variables are included in an equation when the sample equals the population, which is the case in this study (Green & Tukey, 1960). Both fixed-effect variables are represented as dummy variables (Blumenstock, sd). The time fixed-effect variable changes each five years because of the five-year interval that is used.
3.8 Moderator
as already reviewed, a ranking based on economic development. The index is valued between 0 and 1, with 1 as the highest possible value. This is the same index that is used for creating the distinction between developed and developing countries in hypothesis 2a and 2b.
3.9 Instrumental variables
Instrumental variables are used when the statistical model has endogenous variables. Those variables are used in order to eliminate endogeneity. Instrumental variables must satisfy two requirements: it must be uncorrelated with the error and correlated with the endogenous explanatory variable (Wooldridge, 2009), FDI inflow. This study contains the following instrumental variables: lagged primary schooling, change in primary schooling, lagged secondary schooling, lagged tertiary schooling, lagged FDI inflows, lagged real GDP per worker, and the change in public education expenditure. The instrumental variables are relevant as the p-value of the F-test shows estimates of 0,03 and 0,09, which is below the significance level of 0,1.
Table 2. Overview variables
Type of variable Name of variable Description
Dependent Human capital Measured in average years of tertiary and secondary schooling
Independent FDI (in millions $) Foreign direct investment from all countries over the world into one specific Latin American country Independent FDI of OECD countries (in millions
$)
Foreign direct investment from OECD countries into one specific Latin American country
Control Domestic investment Ratio of gross domestic investment at current purchasing power parity. Control Real GDP per worker (in constant
2011 PPP $)
Real GDP per worker Control Investment of education (% of
government expenditure)
Ratio of government educational spending in total government expenditure
Fixed-effect Country Dummy variable per country. Indicates one of the following countries: Argentina, Bolivia, Brazil, Chile, Colombia, Costa Rica, Dominican Republic, Ecuador, El Salvador, Guatemala, Mexico, Panama, Paraguay, Peru, Uruguay, Venezuela
Moderator Human development index score A country’s development score based on the HDI index
Control variable (robustness) Real GDP per capita (in constant 2011 PPP $)
Real GDP per capita
Control variable (robustness) Investment of education (% of GDP) Ratio of government educational spending in GDP
Instrumental variable Lagged primary schooling Average years of primary schooling (one period lag)
Instrumental variable Change in primary schooling Percent change in years of primary schooling
Instrumental variable Lagged secondary schooling Average years of secondary schooling (one period lag)
Instrumental variable Lagged FDI inflow Foreign direct investment from all countries over the world into one specific Latin American country (one period lag)
Instrumental variable Lagged GDP per capita/employee (in constant 2011 PPP $)
Real GDP per worker/capita (one period lag)
Instrumental variable Change in education expenditure as percentage of total expenditure/ GDP
Change in education expenditure as percentage of total expenditure/GDP of previous year
3.10 Statistical model
The statistical model in this study is developed by Zhuang (2017). The model is suitable in this study because it tests the effect of FDI on human capital in countries.
Hit = 0 + 1 FDIit + 2Xit + t + I + it, i = country
t = year
H = human capital
FDI = foreign direct investment X = control variables
DI = domestic investment
RGDEMP = real GDP per worker
EDUGOV = ratio of government educational spending in total government expenditure = fixed effect variable for time
= fixed effect variable for country = error term
3.11 Estimated method
and the independent variable are quantitative and endogenous. The estimated method is determined per model that is examined.
3.12 Method assumptions
OLS can be executed when data meets the following assumptions: measurement of variables, linearity between the dependent and independent variable, homoscedasticity, normality, independent residuals, measurement mistakes, outliers, and multicollinearity (De Vries & Huisman, 2007). Independent residuals and measurement mistakes are not included in this section because they are not applicable to this study. This study meets all remaining assumptions that are mentioned by De Vries and Huisman (2007). For 2SLS these assumptions are important as well. In addition, serial correlation is tested as this may occur when instrumental variables are used (Wooldridge, 2009).
3.12.1 Measurement
First, FDI inflow, FDI inflow from OECD countries, investment of education (as % of government expenditure), HDI score, investment of education (as % of GDP), change primary schooling, and change in education expenditure as percentage of total expenditure and GDP are measured on interval level. Because the variables are measured on interval and ratio level this study meets the first assumption.
3.12.2 Linearity
countries, the scatter plots and residuals plots showed linear relationships for all periods separately. However, when all periods are combined, tertiary schooling and FDI do not show a linear relationship. Although all periods apart show linear relationships, the assumption is accepted.
3.12.3 Heteroscedasticity
OLS assumes that variances of the residuals are constant, also known as homoscedasticity or ‘the constant variance assumption’ (Wooldridge, 2009). On the other hand, when variances of the residuals are not constant it is called heteroscedasticity. When heteroscedasticity is present, the relative reliability of every observation is unequal. If the variance becomes larger, the weight of the observations becomes lower. Heteroscedasticity violates the F-value, the standard error of estimate and, the T-ratios (Gupta, 2000). This will lead to an invalid conclusion. So, when the regression is violated variables have to be transformed. Another possibility is to use the weighted least square regression analysis. To test heteroscedasticity, the Breusch-Pagan test and White’s test for heteroscedasticity are used. The Breusch-Pagan test is one of the most commonly used tests for heteroscedasticity. The OLS estimator is the best unbiased estimator when a form of homoscedasticity is known in the test (Wooldridge, 2009). Even though the test checks on errors, not all errors are excluded when there is no heteroscedasticity. There can still be errors, but those are not related to the independent variable (Bickel, 1978; Breusch & Pagan, 1979).
Table 3. Breusch-Pagan test for heteroscedasticity
Tertiary schooling Secondary schooling
Sig. Sig.
Hypothesis 1. Total 0,335 0,569
Hypothesis 2a. Developed countries 0,081 0,104
Hypothesis 2b. Developing countries 0,022 0,186
Hypothesis 3. Total - OECD 0,463 0,851
Hypothesis 4. Total – OECD – Moderator 0,151 0,859
Table 4. White's test for heteroscedasticity
Tertiary schooling Secondary schooling
Sig. Sig.
Hypothesis 1. Total 0,437 0,611
Hypothesis 2a. Developed countries 0,053 0,099
Hypothesis 2b. Developing countries 0,595 0,182
Hypothesis 3. Total - OECD 0,409 0,736
Hypothesis 4. Total – OECD – Moderator 0,115 0,974
In order to confirm the Breuch-Pagan test, White’s test will be used. The White test for heteroscedasticity has six more regressors than the Breusch-Pagan test, which means that nine restrictions are tested (Wooldridge, 2009). White’s test for heteroscedasticity compares the consistent estimator to the usual covariance matrix estimator (White, 1980). A significance level higher than 0.05 indicates that heteroscedasticity is not present. In table 4, the results of the White’s test for heteroscedasticity are shown.
3.12.4 Normality
Residuals in the population have to be normally distributed with a mean of zero and a constant standard deviation. When variables are not normally distributed it affects OLS by enlarging or reducing the significance level. Normality can be measured with the Kolmogorov-Smirnov and Shapiro-Wilk test. The first one is used for samples that are larger than 2000. If a sample is smaller, the Shapiro-Wilk test has to be used (Wolverton, sd). This study consists of seventeen countries with a maximum of 68 observations and therefore, all variables are tested with the Shapiro-Wilk test. With normal quantile plots (Q-Q plots) the first impression of normality is checked. To confirm this, the Shapiro-Wilk test is conducted. Results of the Shapiro-Wilk test show for secondary schooling, expenditure on education as % of total government expenditure, and domestic investment significance levels higher than 0.05. This indicates that these variables are normally distributed. However, a significance level of less than 0.05 is shown for tertiary schooling, FDI inflow, FDI inflow from OECD countries, and the real GDP per worker. This means that these variables are not normally distributed. As normality is one of the main assumptions of OLS, it has to be satisfied. Transformation is a possible solution to convert non-normal distributed variables into normal distrusted variables (De Vries & Huisman, 2007).
3.12.5 Outliers
Studies with small data sets are sensitive for outliers. Unusual observations can influence OLS estimates very easy. Outliers exist when some observations, countries in this study, are very different compared to the others (Wooldridge, 2009). It is hard to decide to keep or drop an outlier, because it is possible that a country is performing extraordinary good or bad for instance. Even though the results are realistic and correct, outliers can affect the results of the study, due to the fact that outliers for example affect normality. Another reason why outliers can occur is because of data that is generated from a different model than the rest of the data (Wooldridge, 2009).
This study makes a distinction by performing multiple techniques with outliers. ‘Winsorizing’ is the first technique that is performed. This technique modifies the value of an
3.12.6 Multicollinearity
Multicollinearity refers to the correlation between the independent variables in a multiple regression model (Wooldridge, 2009). To detect if multicollinearity occurs, variance inflation factors (VIF) and collinearity tolerance are used. VIF indicates how much the estimated variance of the estimated regression coefficient is increased due to collinearity. Prior research set different thresholds for the maximum VIF and collinearity tolerance. Several authors use a VIF of 10 or a collinearity level of 0,1 as guideline for multicollinearity (Mason, et al., 1989; Neter, et al., 1989; Kennedy, 1992). However, Menard (1995) states that multicollinearity is cause for concern when the VIF is higher than 5 and the collinearity tolerance is lower than 0,2. He is critical to other studies and states that with, a VIF between 5 and 10, and a collinearity tolerance between 0,1 and 0,2 is already cause for concern. This study uses a maximum VIF threshold of 5 and a collinearity tolerance of 0.2, in order to avoid multicollinearity. Statistical tests that calculate VIF indicate that multicollinearity may be problematic in this study. The highest calculated VIF is 2,688 (collinearity tolerance of 0,372) and belongs to FDI inflow from OECD countries in Hypothesis 4. As the highest VIF is below the threshold of 5 or a collinearity tolerance of 0,2 it can be concluded that this assumption does hold. All multicollinearity tests are presented in appendix III.
3.12.7 Endogeneity
variable is present. In order to test endogeneity, this study uses the Durbin-Hausman-Wu test. It is important to know if endogeneity exists in a study, because it indicates what test has a higher validity and reliability. When the explanatory variable is exogenous, OLS is a better method. Once the explanatory variable is exogenous, 2SLS can have very large standard errors. This indicates that 2SLS has a lower validity and reliability than OLS. If the explanatory variable is endogenous, 2SLS is a better method (Wooldridge, 2009).
The Durbin-Hausman-Wu test compares instrumental variable estimates with OLS estimates. FDI and OECD FDI are tested on endogeneity, both for tertiary and secondary schooling. The Durbin-Hausman-Wu test uses the same criteria as the chi-square test. A significance level smaller than 5% indicates that the null hypothesis of exogeneity can be rejected. The estimates show a significance level of less than 5% only for secondary schooling in hypothesis 2b. The structural equation has a p-value of 0,004 while the p-value of the robustness test is 0,016. This indicates that endogeneity is present. Therefore, the 2SLS test is used for secondary schooling in hypothesis 2b, while OLS is used for the other tests. The estimates of the Durbin-Hausman-Wu test for each model are presented in table 7, 8, 9, 10, and 11.
expenditure as percentage of total expenditures and as percentage of GDP. Lagged explanatory variables are often used to move the dataset through which endogenuous estimates replace from a “selection on observables” assumption to an equally untestable “dynamics among unobservables” assumption. Prior studies used one period lagged explanatory variables to address the endogeneity problem in the 2SLS method (Alfaro & Chauvin, 2016; Salim, et al., 2014; Hinkman & Olney, 2011). Therefore, this study uses one period (one year) lags as well. The relevance of these variables is measured with the F-test. Estimates of the F-test are shown in table 9.
3.12.8 Robustness test
In order to check if the concept can operate exclusive of failure under a diversity of circumstances, the quality of the data will be assessed with a robustness test. With the robustness check several situations are created by adding and removing regressors. The robustness check examines if the core regression coefficient is certain and examines the validity of the test when coefficients are robust and plausible (Lu & White, 2014). In this study, two different examinations will show if the concept is robust. Different measures of output and educational expenditure are used. First, real GDP per worker and education expenditure (% of government spending) are examined. In the second situation, real GDP per capita and education spending (% of GDP) are examined. The robustness check is presented in table 7, 8, 9, 10, and 11.
3.12.9 Serial correlation
affected. Those are smaller in such a situation, which will lead to a situation in which the parameters estimates are more precise than they really are (Williams, 2015; O'Halloran, n.d.). Serial correlation in this study can be seen as; if FDI and human capital are positively related in one year, the probability is greater that next year’s result is positively related than
negatively related. To test for serial correlation, the Durbin-Watson test is used. Field (2009) suggests that values below 1 or higher than 3 are cause for concern. The values in table 5 do not show any cause for concern as no value is out of the range.
Table 5. Serial correlation
Hypothesis Durbin-Watson estimate
Tertiary schooling Secondary schooling
Hypothesis 1 2,714 2,714
Hypothesis 2a 1,598 1,203
Hypothesis 2b 1,430 1,409
Hypothesis 3 1,855 1,423
4. EMPIRICAL RESULTS
In this section, the OLS results are presented and explained. First the descriptive statistics are discussed, followed by statistics of the regression.
4.1 Descriptive statistics
The descriptive statistics of all variables are analysed. By analysing these statistics, the first thoughts are developed. Therefore, descriptive statistics are important. This study contains a maximum of 68 observations. The observations include four time periods of seventeen Latin American countries. The descriptive statistics of the main variables show plausible means and standard deviations, except for FDI inflow and FDI inflow from OECD countries. The coefficients of FDI inflow and FDI inflow from OECD countries is negative. With a negative sign UNCTAD indicates reversed investment or disinvestment (UNCTAD, sd). This minimum does not occur later in the study because both variables are transformed with Log10. Variables with negative values cannot be transformed with logarithm and therefore they are presented as missing values (Cox, 2005). After the descriptive statistics are shown, it is important to deal with outliers before continuing the study. To detect outliers the variables are analysed in more detail. The histogram and outlier table of the variable show what observations are potential outliers (appendix IV). As mentioned in previous section, outliers are changed with the ‘winsorize’ method if possible.
some data is still missing. Missing data affects the study because for some time-lags the whole sample could not be tested.
Table 6. Descriptive statistics
Variable Obs. Min. Max. Mean Std. Deviation
Tertiary schooling 68 0,12 1,17 0,47 0,23
Secondary schooling 68 0,73 4,01 2,29 0,70
FDI inflow (Million $) 68 -983 73086 5886,02 11257,24 Log 10 FDI inflow (Million $) 65 1,55 4,92 3,0970 0,77235 FDI inflow from OECD countries
(million $)
34 -306 24669 3114,76 6197,29 Log 10 FDI inflow from OECD countries
(million $)
29 1,28 4,6 3,0446 0,83507 Expenditure on education (% GDP) 65 5,96 24,53 16,15 4,33 GDP per worker (Constant 2011 PPP $) 68 10201 47383 25168,73 9981,67
Domestic investment 68 0,10 0,36 0,19 0,05
4.2 Regression results
4.2.1 FDI and human capital – hypothesis 1
Model 1.2 shows the estimates of the robustness check. Real GDP per worker and education expenditure as percentage of government spending are left out in this model and real GDP per capita and education expenditure as percentage of GDP are included. The estimates in model 1.2, confirm that an increase of FDI inflow might lead to an increase of tertiary schooling. They further show insignificant estimates for real GDP per capita related to tertiary schooling, which means that making a conclusion is invalid. This does not count for expenditure on education as percentage of GDP. With a significance level of 10% it can be concluded that if expenditure on education as percentage of GDP increases, tertiary schooling increases as well. The last significant estimates of model 1.2 belong to the year 2015. It can be assumed that in 2014 an increase of FDI inflow have led to an increase of the average years of tertiary schooling in 2015. It can be concluded that relationship between FDI inflow and tertiary schooling is robust.
Table 7. Tertiary and secondary schooling Tertiary schooling Model 1.1 Tertiary schooling Model 1.2 Secondary schooling Model 2.1 Secondary schooling Model 2.2
Log FDI inflow 0,107**
(0,046) 0,112** (0,048) 0,347*** (0,115) 0,347*** (0,126) Domestic investment -0,961* (0,552) -0,926 (0,634) -2,728* (1,384) -2,870* (1,666) Log Real GDP per worker 0,091
(0,203)
1,604*** (0,509) Education expenditure (% of government spending) 0,005
(0,007)
0,001 (0,016)
Log Real GDP per capita -0,031
(0,120) 0,915*** (0,320) Education spending (% of GDP) 0,040* (0,021) -0,022 (0,056) Year 2000 0,064 (0,080) 0,014 (0,081) 0,312 (0,219) 0,183 (0,253) Year 2005 0,144 (0,110) 0,107 (0,126) 0,244 (0,244) 0199 (0,278) Year 2010 0,116 (0,114) 0,084 (0,126) 0,389 (0,270) 0,415 (0,330) Year 2015 0,126 (0,126) 0,306* (0,133) 0,489 (0,284) 0,797** (0,324) Constant -1,020 (0,828) -0,660* (0,349) -5,303*** (2,077) -1,497 (0,915) Observations 63 58 63 58 R-squared 0,17 0,217 0,452 0,429
Exogeneity of FDI (p-value of Durbin–Hausman– Wu F-test)
0,545 0,496 0,097 0,215 Note: Standard errors in parentheses, ***p<0,01, **p<0,05, *p<0,1. Year 2000, 2005, 2010 and 2015 are estimated by the relationship between tertiary or secondary schooling (based on the model) and FDI inflow in the particular year.
4.2.2 FDI and human capital in subsamples - hypothesis 2a and 2b
Hypothesis 2a tests if inward FDI has a positive relationship with tertiary and secondary schooling in developed countries in Latin America. The Durbin-Hausman-Wu test indicates that all models have to be tested with OLS, as the p-value is higher than 5%. The estimates in model 1.1 of table 8 show a negative and significant relationship between FDI inflow and tertiary schooling (B = -0,099 and p-value<0,01). This suggests that when FDI inflow increases with one percent, the average years of tertiary schooling in developed countries decreases with 0,099%. Other significant results in model 1.1 are estimated for real GDP per worker (B = 0,581 and p-value<0,1) and education expenditure as percentage of government spending (B = 0,024 and p-value>0,01). Table 8 includes different years as well. Model 1.1 shows a significant estimation for the year 2000. It can be assumed that an increase of one percent in FDI inflow in 1999 have led to a decrease of 0,112% in average years of tertiary schooling in developed countries. The estimates of model 1.2 are used for the robustness check. The estimates do not show any significant results and therefore the relationship between FDI inflow and tertiary schooling does not hold under different circumstances. The estimation of R-squared of model 1.1 (0,595) is thereby much higher than R-squared of model 1.2 (0,172). Meaning that model 1.1 explains 59,5% of the outcomes of tertiary schooling in developing countries and leaves only 40,5% out, while model 1.2 explains just 17,2% of the outcomes.
Table 8. FDI and human capital in developed countries Tertiary schooling Model 1.1 Tertiary schooling Model 1.2 Secondary schooling Model 2.1 Secondary schooling Model 2.2
Log FDI inflow -0,099***
(0,033) -0,064 (0,054) 0,022 (0,139) 0,164 (0,168) Log Domestic investment 0,171
(0,244) 0,642 (0,509) 1,370 (2,078) 0,772 (1,576) Log Real GDP per worker 0,581***
(0,128)
2,451*** (0,534) Education expenditure (% of government spending) 0,024***
(0,005)
0,025 (0,019)
Log Real GDP per capita -0,038
(0,146) 1,577*** (0,452) Education spending (% of GDP) 0,080* (0,039) -0,178 (0,120) Year 2000 -0,112* (0,037) -0,155 (0,068) -0,152 (0,217) -0,291 (0,296) Year 2005 -0,111 (0,079) -0,158 (0,099) 0,018 (0,365) -0,002 (0,291) Year 2010 -0,102 (0,112) 0,113 (0,205) 0,089 (0,522) 0,644 (0,629) Year 2015 -0,043 (0,156) - 0,175 (0,533) - Constant -2,830*** (0,548) 0,144 (0,595) -8,440*** (2,283) -2,701 (1,845) Observations 31 27 31 27 R-squared 0,595 0,172 0,529 0,461
Exogeneity of FDI (p-value of Durbin–Hausman–Wu F-test)
0,661 0,145 0,594 0,983
Note: Standard errors in parentheses, ***p<0,01, **p<0,05, *p<0,1. Year 2000, 2005, 2010 and 2015 are estimated by the relationship between tertiary or secondary schooling (based on the model) and FDI inflow in the particular year.
which indicates that the endogeneity problem exists between FDI inflow and secondary schooling in developing countries. Therefore, both models are tested with 2SLS.
Estimates in model 1.1 show a significant and positive relationship between FDI inflow and tertiary schooling in developing countries. The results suggest that an increase of one percent FDI inflow leads to an increase of 0,329% in average years of tertiary schooling in developing countries. Other estimates are negative and significant and are shown for domestic investment (B= -0,619 and p-value<0,05), real GDP per worker (B = -1,359) and education expenditure as percentage of government spending (B = -0,027 and p-value<0,01). It can be assumed that in 2004 an increase of one percent FDI inflow have led to an increase of 0,461% in average years of tertiary schooling in developing countries. In 2015 this might be increased with 0,438%. Model 2.1, for secondary schooling, shows a positive and significant relationship between FDI inflow and secondary schooling. It can be assumed that an increase of one million dollar in FDI inflow leads to an increase of 0,01188% in average years of secondary schooling. R-squared in table 9 shows that in model 1.1, FDI inflow is able to explain 67,2% of the outcomes of tertiary schooling. Where FDI inflow in model 2.1, is able to explain 56,9% of the outcomes of secondary schooling in developing countries.
Table 9. FDI and human capital in developing countries Tertiary schooling Model 1.1 Tertiary schooling Model 1.2 Secondary schooling Model 2.1 Secondary schooling Model 2.2 Log FDI inflow 0,329***
(0,058) 0,294*** (0,068) 1,188*** (0,398) 0,438 (0,423) Log Domestic investment -0,619**
(0,257) -0,729** (0,276) 8,406 (6,130) 7,314 (5,972) Log Real GDP per worker -1,359***
(0,398) -6,657 (3,997) Education expenditure (% of government spending) -0,027*** (0,009) -0,45 (0,114) Log Real GDP per capita -0,106
(0,190) 0,025 (1,184) Education spending (% of GDP) 0,057** (0,023) 0,387* (0,184) Year 2000 0,266 (0,130) 0,283 (0,125) - - Year 2005 0,461** (0,078) 0,395 (0,208) - - Year 2010 0,343 (0,163) 0,123 (0,161) 1,109 (1,026) -0,662 (0,274) Year 2015 0,438** (0,117) 0,288 (0,210) X 0,246 (0,440) Constant 4,381*** (1,795) -1,688** (0,672) 26,133 (17,744) -2,054 (3,202) Observations 31 30 21 21 R-squared 0,672 0,565 0,449 0,530
Exogeneity of FDI (p-value of Durbin–Wu-Hausman F-test) 0,320 0,535 0,004 0,016 Instrument relevance (p-value of F-test - - 0,030 0,09
Note: Standard errors in parentheses, ***p<0,01, **p<0,05, *p<0,1. Year 2000, 2005, 2010 and 2015 are estimated by the relationship between tertiary or secondary schooling (based on the model) and FDI inflow in the particular year. Model 2.2 includes the following instrumental variables: lagged primary schooling, change of primary schooling, lagged tertiary schooling, lagged FDI, lagged real GDP per capita and change in expenditure on education as percentage of GDP.
4.2.3 FDI of OECD countries and human capital - hypothesis 3
model 1.1 and 1.2 are directed to tertiary schooling and model 2.1 and 2.2 to secondary schooling. Estimates in model 1.2 show a positive but insignificant relationship between tertiary schooling and FDI inflow of OECD countries. A positive, but insignificant, relationship is also found for real GDP per worker, education expenditure as percentage of government spending, FDI inflow of OECD countries in 2005 and 2010 as well. On the other side, a negative but insignificant relationship is found for domestic investment and tertiary schooling. The combination of FDI inflow of OECD countries, domestic investment, real GDP per worker and education expenditure of government spending together can account for 14,4% in variation in tertiary schooling. So, from this model it is known that FDI inflow of OECD countries can explain 14,4% the outcomes in tertiary schooling. This leaves 85,6% of the model unexplained.
The robustness check of model 1.1 is presented in model 1.2. Real GDP per worker and education expenditure as percentage of government spending are changed with real GDP per capita and education spending as percentage of GDP. Estimates of model 1.2 show a positive but insignificant relationship between FDI inflow of OECD countries and tertiary schooling. A positive and insignificant relationship is also found for real GDP per capita, education spending as percentage of GDP and year 2010. Finally, a negative but insignificant relationship is found for domestic investment and year 2005 with tertiary schooling.
35,5% outcomes in secondary schooling. This leaves 64,5% of the model unexplained. Model 2.2 is used for the robustness check and the estimates show a positive but insignificant relationship between FDI inflow of OECD countries and secondary schooling as well. Domestic investment and education spending as percentage of GDP are both negatively related, but insignificant. In contrast, real GDP per capita is positively related and has a significance level of less than 5%. R-squared of model 2.2 is 0,36 and explains 36% outcome of secondary schooling.
Table 10. FDI inflow from OECD countries and tertiary and secondary schooling
Tertiary schooling Model 1.1 Tertiary schooling Model 1.2 Secondary schooling Model 2.1 Secondary schooling Model 2.2 Log FDI inflow of OECD
countries 0,047 (0,072) 0,008 (0,071) 0,120 (0,165) 0,138 (0,177) Domestic investment -0,354 (1,265) -0,271 (1,268) -2,414 (2,917) -2,519 (3,144) Log Real GDP per worker 0,248
(0,361) 1,732* (0,833) Education expenditure (% of government spending) 0,012 (0,011) -0,004 (0,025)
Log Real GDP per capita 0,261
(0,205) 1,149** (0,509) Education spending (% of GDP) 0,070 (0,035) -0,051 (0,086) Year 2005 0,034 (0,105) -0,118 (0,075) -0,006 (0,195) -0,011 (0,238) Year 2010 0,035 (0,130) 0,027 (0,132) 0,232 (0,303) 0,319 (0,345) Constant -1,735 (1,452) -1,591** (0,640) -5,052 (3,347) -1,554 (1,586) Observations 26 25 26 25 R-squared 0,144 0,291 0,355 0,360
Exogeneity of FDI (p-value of Durbin–Hausman–Wu F-test)
0,748 0,112 0,853 0,901
4.2.4 FDI of OECD countries and human capital with HDI score - hypothesis 4
The fourth hypothesis tests if the effect of FDI from OECD countries on tertiary and secondary schooling in Latin American countries is negatively moderated by the HDI score of these countries. The Durbin-Hausman-Wu test indicates that all models have to be tested with OLS, because the p-value of the test is higher than 5%. The estimates in table 11 show in model 1.1 several significant estimates. The moderator, HDI score, is significant and negative (B = -0,181 and p-value<0,001). This suggests that that if a country’s HDI score increases, the relationship between FDI inflow of OECD countries and tertiary schooling weakens. Besides the moderator, the model shows several other significant estimates. An increase of the HDI score is related to an increase in tertiary schooling. Domestic investment and education expenditure are also positively related to tertiary schooling in Latin American countries. Real GDP per worker is the only estimate that shows a negative and significant relationship with tertiary schooling in Latin American countries. Model 2.1 shows the estimates of FDI inflow of OECD countries and secondary schooling. In this model, only one estimate is significant. A higher HDI score of a Latin American country might lead to an increase in secondary schooling in these countries. R-squared in table 11 shows that model 1.1 explains 54,5% of the outcomes of tertiary schooling, where model 2.1 explains 57,3% of the outcomes of secondary schooling in Latin America.
Table 11. FDI inflow from OECD countries and tertiary and secondary schooling with the moderator (HDI) Tertiary schooling Model 1.1 Tertiary schooling Model 1.2 Secondary schooling Model 2.1 Secondary schooling Model 2.2 Log FDI inflow of OECD
countries x HDI score
-0,181*** (0,063) -0,085 (0,051) 0,010 (0,167) 0,081 (0,161) Log FDI inflow of OECD
countries 0,013 (0,043) -0,030 (0,044) -0,003 (0,113) -0,011 (0,137) HDI score 0,114** (0,053) 0,048 (0,053) 0,378** (0,139) 0,312* (0,167) Domestic investment 2,348* (1,201) 1,034 (1,038) 4,482 (3,181) 4,192 (3,251) Log Real GDP per worker -0,634*
(0,356) -0,022 (0,943) Education expenditure (% of government spending) 0,024** (0,009) 0,030 (0,024) Log Real GDP per capita -0,019
(0,244) 0,024 (0,073) Education spending (% of GDP) 0,107*** (0,023) 0,101 (0,073) Constant 1,608 (1,518) -0,960 (0,884) 1,043 (4,022) 0,942 (2,770) Observations 26 25 26 25 R-squared 0,545 0,690 0,573 0,579
Exogeneity of FDI (p-value of Durbin–Hausman–Wu F-test)
0,748 0,112 0,853 0,901
5. DISCUSSION AND LIMITATIONS
5.1 Discussion
In the discussion, results of this study are explained based on literature. The aim of this study is to assess the relationship between FDI and human capital in Latin America. This study demonstrates that FDI inflow has a positive relationship with tertiary and secondary schooling in Latin America. The robustness check confirms the positive relationship between FDI inflow and tertiary and secondary schooling in Latin America. This is also in line with the theory (Wang, 2011; Zhuang, 2013; Azam, et al., 2015). Therefore hypothesis 1 can be supported. This adds value to the literature, as prior research did not focus on Latin America in this context.
Another important finding of this study is that it is assumed that FDI inflow have a negative effect on tertiary schooling in developed countries. This is in contrast to existing literature (Checchi, et al., 2007; Zhuang, 2017). This is the first study that found a negative effect on tertiary schooling. The most likely explanation of this negative effect is that the most important sectors of economies in developing countries in Latin America have lower-skilled-labour, resulting in less demand for high-skilled-labour. Bárcena et al. (2015) show that the distribution of FDI in the sector ‘services’ in Brazil and Mexico is approximately 35%, while manufactures and natural resources belong to the other 65%. For secondary schooling, no significant estimates have been found. Therefore, no conclusions can be drawn. As tertiary schooling is not in line with the hypothesis and secondary schooling does not have valid data, hypothesis 2a has to be rejected.
circumstances. The positive relationship of FDI and tertiary schooling is not in line with the study of Zhuang (2017). This can be explained because Zhuang focused on Asian countries where in standardized and unskilled labour is one of their main export products (Fukao, et al., 2003). This discourages tertiary schooling and encourages secondary schooling. Moreover, it is not the first time a study finds a positive result for tertiary schooling in developing countries. Blömstrom and Kokko (2002) argue that MNEs increase tertiary schooling in developing countries, due to the demand for skilled labour. MNEs are almost always established in developed countries (Fortune, 2017). The positive relationship between FDI with tertiary schooling in developing countries in this study can be explained by the same theory. From the total foreign investment in Latin America, approximately 73% is coming from high developed countries such as the United States and countries in the European Union. The other 27% is coming from other countries that may be less developed (Bárcena, et al., 2017). The last main finding of this study is that FDI inflow is positively related with secondary schooling in developing countries in Latin America. This is in line with the theory and can be confirmed by the fact that the main sectors of developing countries in Latin America are natural resources and manufacturing followed by services (Bárcena, et al., 2015). Hypothesis 2b is partially supported as only the positive relationship between FDI inflow and secondary schooling is in line with the theory.
OLS except for tertiary schooling’s normality. This can explain the insignificant result for the relationship between FDI inflow from OECD countries and tertiary schooling (De Vries & Huisman, 2007). Tertiary schooling is not normally distributed (p-value<0,05). In summary, estimates in this study only show insignificant results in the relationship between FDI inflow from OECD countries and human capital and therefore hypothesis 3 has to be rejected.
The last hypothesis, Hypothesis 4, also focused on FDI inflow from OECD countries. In addition, a moderator is included in this hypothesis. A significant and negative relationship is found for the HDI score as moderator and tertiary schooling in Latin American countries. This is in line with the theory, as Becker (1994) claims that investment in human capital causes higher gains in less developed countries than in high developed countries. For secondary schooling and the robustness check, no significant estimates are found. Therefore, no conclusion can be drawn. It can be assumed that this is also due to the low number of observations. Hypothesis 4 also contains 26 observations. Based on these results, hypotheses 4 is partially supported as only evidence is found for tertiary schooling.
In a concluding note, this means that hypothesis 1 is fully supported and hypothesis 2b and 4 are only partial supported. No evidence has been found to accept hypothesis 2a and 3.
5.2 Limitations
Although this study provides a detailed understanding of the relationship between FDI and human capital in Latin America, some limitations are identified. The measurement of human capital causes the first limitation. This study uses educational attainment as proxy for human capital. However, this proxy has some limitations. First, a person’s years of school attainment does not necessarily mean that human capital increases during the attained years. Secondly, this proxy does not make the distinction between the quality of education across countries and the time the education takes. Finally, educational attainment does not make the distinction between education categories and assume that workers are perfect substitutes for each other even though they are from another education category (Judson, 2002). Therefore, further research has to focus on a better measurement of human capital in order to take into account these limitations of educational attainment.
Secondly, FDI inflows are sensitive to fluctuations due to disinvestments and large mergers for instance (Noorbakhsh, et al., 2001). This encourages random events and a less reliable analysis. Further research can use three or five year averages in order to reduce the possibility of random events.
of France and the United States.
The fourth limitation of this study is about using winsorizing. The variable education expenditure as percentage of GDP is winsorized. If the data was not winsorized, the variable was non-normally distributed. For that purpose the data of Bolivia is adapted. This results in an unfair view of Bolivia when looking at the results of the robustness check. Therefore, it is not fully appropriate to include Bolivia when using results of this study in further research. Further research on Bolivia or on Latin America should create new appropriate estimates on FDI inflow and human capital. Furthermore, extensive research in this area is necessary for what the effect of FDI inflow on human capital in Latin America is. This could create new insights that are useful for the development of new (trade) policies in order to encourage human capital and economic growth.
6. CONCLUSION
REFERENCES
Adenutsi, D., (2010). Long-run macroeconomic impact of international migrant remittances on human development in low-income countries: a panel analysis of sub-Saharan Africa. Journal of International Economic Studies, (24), pp. 113 - 132.
Alfaro, L., Chanda, A., Kalemli-Ozcan, S. and Sayek, S., (2006). How Does Foreign Direct Investment Promote Economic Growth? Exploring the Effects of Financial Markets on Linkages. Harvard Working Papers, 07(013), pp. 1 - 54.
Alfaro, L. and Chauvin, J., (2016). Foreign Direct Investment, Finance, and Economic
Development. In: M. Spatareanu. 3rd edition. Encyclopedia of International Economics
and Global Trade. Singapore: World Scientific, forthcoming., pp. 1 - 32.
Azam, M. et al., (2015). Foreign direct investment and human capital: evidence from developing countries. Investment Management and Financial Innovations, 12(3), pp. 155 - 162.
Azariadis, C. and Drazen, A., (1990). Threshold externalities in economic development.
Quarterly Journal of Economics, 105(2), pp. 501 - 526.
Bárcena, A., Prado, A., Cimoli, M. and Pérez, R., (2015). Foreign Direct Investment in Latin
America and the Caribbean, Santiago: United Nations. Available at
http://repositorio.cepal.org/bitstream/handle/11362/38215/S1500534_en.pdf. [Accessed on 25 10 17]
Bárcena, A., Prado, A., Cimoli, M. and Pérez, R., (2017). Foreign Direct Investment in Latin
America and the Caribbean. Santiago: United Nations. Available at
Baron, R. and Kenny, D., (1986). The moderator-mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of
Personality and Social Psychology, (61), pp. 1173 - 1182.
Barro, R., 1991. Economic growth in a cross section of countries. Quarterly Journal of
Economics, 106(2), pp. 407 - 443.
Barro, R. & Lee, J., (1996). International measures of schooling years and schooling quality.
American Economic Review, 86(2), pp. 218 - 223.
Barro, R. & Lee, J.-W., 2013. A New Data Set of Educational Attainment in the World, 1950-2010. Journal of Development Economics, (104), pp. 184 - 198.
Basu, B. and Yao, J., (2009). Foreign Direct Investment and Skill Formation in China.
International Economic Journal, 23(2), pp. 163 - 179.
Becker, G., (1994). Human Capital Revisited. In: G. Becker. Human Capital: A Theoretical
and Empirical Analysis with Special Reference to Education.3rd ed. Chicago: The
University of Chicago Press., pp. 15 - 28.
Bengoa, M. and Sanchez-Robles, B., (2003). Foreign direct investment, economic freedom and grwoth: new evidence from Latin America. European Journal of Political
Economy, 19, pp. 529 - 545.
Bickel, P., (1978). Using residuals robustly I: Test for heteroscedasticity and nonlinearity.
Annals of Statistics 6, pp. 266 - 291.
Blomström, M. and Kokko, A., (2002). FDI and Human Capital: A Research Agenda. OECD
Development Centre Working Papers. (195), pp. 1 - 32.
Blumenstock, J., (n.d.). Fixed Effect Models.
