Master Thesis Human capital inequality and income inequality

(1)

Master Thesis

Human capital inequality and income

inequality

University of Groningen Faculty of Economics and Business MSc Economic Development & Globalization

Student: Andries Slegten Student number: 3841863

Student Mail: A.B.Slegten@student.rug.nl Supervisor: Dr. Anna Minasyan

(2)

2

Abstract

This paper investigates whether a higher level of human capital inequality is associated with higher income inequality. The research period is 1960-2010 and focuses on an unbalanced panel set of 118 countries with 3340 observations. Using Pooled OLS and fixed effects methods for regression analysis, the findings suggest that there exists a positive association between human capital inequality and income inequality. In a subsample analysis using indicator variables for continents interacted with human capital inequality the association remains positive and significant for North-America and Latin-America, while subsequently increasing in magnitude. This suggests that the association only exists in these two continents. The findings are robust when using lagged variables to account for endogeneity concerns stemming from reverse causality.

Keywords: Human Capital Inequality, Income Inequality, Unbalanced Panel data, Pooled OLS, Fixed Effects, Reverse Causality

(3)

3

1. Introduction

From the end of the second world war up until 1980 within-country income inequality stayed more or less at the same level around the world. However, in the 30 years after that, within-country income inequality has increased again (Milanovic, 2016). During this whole period, education expanded on a worldwide scale. Literacy rates increased towards 85 percent in 2010 (World Bank, 2018), and human capital inequality, measured through the dispersion in educational attainment, decreased in all continents of the world. In this paper, human capital inequality is defined as the unequal educational attainment between individuals in a population (de Gregorio & Lee, 2002). Income inequality is defined as the dispersion of income between individuals where each person is assigned his or her household per capita income (Milanovic, 2016). Income is accrued by individuals or households through revenue streams from salaries, wages and other forms of compensation, which do not include wealth statistics such as the value of homes or other possessions. For income inequality to exist, all which is needed are two individuals who earn different incomes (OECD, 2018). In the literature of economic development many theories exist into the causes and consequences of income inequality. Furthermore, much empirical work has been conducted to explain the magnitudes of the causes. Examples of such theories are the influential work of Kuznets (1955) who argues that income inequality exhibits an inverted-U curve due to heterogeneous benefits of economic development over time (Kuznets, 1955). Meltzer & Richard (1981) argue that increasing the voting franchise due to increased democratization, inequality is likely to decline as taxes reach a new equilibrium (Meltzer & Richard, 1981). Technological progress is expected to increase income inequality as it increases the skill premium in favour of the high-skilled individuals (Snower, 1999). Globalization is argued to increase income inequality as it outsources the production facilities in the developed countries in disregard of the low-skilled (Alderson & Nielsen, 2002).

(5)

5 within countries is investigated, with an additional subsample analysis to research continental differences in the association human capital inequality has on income inequality. Furthermore, endogeneity concerns stemming from reverse causality are researched. Reverse causality is researched because one could argue that in countries with higher income inequality, the relative price of schooling might be too high for the low income earners, resulting in lower investments in human capital relative to the high income earners. This suggests that the association between human capital inequality and income inequality might not be as hypothesized, but from income inequality towards human capital inequality (Galor & Zeira, 1993).

In the empirical literature on the relation of human capital inequality and income inequality, Castelló-Climent & Doménech (2020) investigate the period 1950-2010 and find a positive and significant association. Furthermore, de Gregorio & Lee (2002) investigate the period 1960-1990 and find a positive and significant association as well. This paper differentiates by investigating the association within countries, while extending the observations by using the CLIO-INFRA (2013) dataset for human capital inequality rather than the Barro & Lee dataset which is used by Castelló-Climent & Doménech (2020) and de Gregorio & Lee (2002). Furthermore, only de Gregorio & Lee (2002) investigate endogeneity concerns stemming from reverse causality, while Castelló-Climent & Doménech (2020) ignore this. Finally, the research period of this paper is different from the papers provided above, as it analyses an unbalanced panel set of 118 countries over time period 1960 till 2010, with 3340 observations.

This paper uses educational attainment data in calculating the Gini-coefficient of human capital inequality. The Gini-coefficient is a method of measuring inequality in a population, and therefore it is used for measuring income- and human capital inequality. The data used in the analysis are human capital inequality data from CLIO-INFRA (2013) and income inequality data from the Standardized World Income Inequality Database (2019). Both datasets are the most comprehensive datasets available on human capital inequality and income inequality.

(6)

North-6 America, being stronger in magnitude than in the overall regression analysis. To account for endogeneity concerns stemming from reverse causality, a robustness check is conducted with a 5 year lagged value of human capital inequality. The findings suggest that the analysis is robust as all coefficients and the levels of significance remain close to the original estimates.

This paper is organized as follows: In chapter 2 the literature review is presented and hypothesises are constructed. Chapter 3 explains the data collection method complemented by the empirical method used to analyse the hypothesises. In chapter 4 the results are presented and discussed. Finally, in chapter 5, the conclusion is presented with limitations, policy implications and advice for future research.

2. Literature review

2.1 Introduction

In this chapter, a theory is explained on why heterogeneity exists in human capital accumulation between countries and its effect on country-level income. Then, a theory is presented which explains within-country inequality in human capital and its association on within-country income inequality and the hypothesis based on this theory is presented. Finally, a critical review of the empirical literature on the topic is presented.

_𝑖

, 𝑇

_𝑖

− 𝑆

_𝑖

)

(1)

Where alpha is the Mincerian return to schooling which can be explained by the employment earnings as a function of schooling and labor market experience (Heckman & Todd, 2006), W is the wage of an individual and S represents the schooling level. The interest rate is included through r, where g is the growth rate of wages and T is the retirement age, and thus T – S represents the years working. Finally, δ is the discount rate(for a full discussion of equation 1, see Hsieh & Klenow, 2010). In the equation, ceteris paribus, higher Mincerian returns to schooling increase the marginal benefits to schooling. In more developed countries, one could argue that the Mincerian returns to schooling are higher as technological change is theorized to have increased the skill premium as it is biased towards high-skilled people (Snower, 1999). Furthermore, globalization fragments production facilities, increasing the relative wages of high-skilled people in the developed countries as the low-skilled jobs are outsourced (Alderson & Nielsen, 2002). As a result, in more developed countries, an individual might have a higher incentive to obtain a higher educational level than in less developed countries, as the returns are higher. Heterogeneity in Mincerian returns might therefore have an effect on the distribution of human capital in a cross-country setting.

(8)

8 & Restruccia, 2010). Essentially the empirical literature of Benhabib & Spiegel (1994) and Erosa et al. (2010) suggests that the effect human capital has on income inequality between countries is indirect and runs through total factor productivity. This could be empirically researched by including total factor productivity with human capital inequality in the regression analysis, however finding a proxy for total factor productivity remains a hard task prone to many measurement errors. Total factor productivity growth is the result of many factors such as technological innovations, changes in factor shares, organizational and institutional changes and many others. In the literature these various factors are often aggregated as ‘leftovers’ instead of measured directly, which explains why total factor productivity is often referred to as total factor productivity residual (for a full discussion on TFP see Hulten, 2001). If a good proxy were to be included, the effect total factor productivity has on the income distribution should be positive and significant, while the effect human capital inequality has on income inequality should turn insignificant. However, researching this is beyond the scope of this paper, as this paper focuses on the relation of human capital inequality within countries and within-country income inequality.

The theory of Hsieh & Klenow (2010) provides explanations of human capital differences between countries and its effect on income differences. However Benhabib & Spiegel (1994) argue for an indirect effect of human capital inequality on income inequality between countries, as they argue that it runs through total factor productivity. This paper focuses empirically on the association of within-country human capital inequality on within-country income inequality. It is however important to theoretically asses the relation human capital inequality between countries has on income differences between countries to be able to better understand the mechanism within countries, which is presented in the next subchapter.

2.3 Within-country human capital inequality and income inequality

Becker & Chiswick (1966) argue that the total earnings of an individual after completing personal investments in human capital is equal to the earnings from their base level of human capital plus the sum of the returns to human capital investment. If one treats returns as constants for an indefinitely long period, the following expression is given:

𝑌

_𝑖

= 𝐵

_𝑖

+ ∑ 𝑟

_𝑖𝑗

𝐶

_𝑖𝑗

𝑚 𝑗=1

(9)

9 In equation 2,

Y

i is the total earnings of person i,

B

i is his/her baseline level of human capital, rij is his rate of return on his/her human capital investment and Cij is the resources spent

from person i on investment j. Equation 1 explains that total earned income Yi is a function of an individual his/her starting level of human capital plus their individual resource allocation towards human capital investment multiplied with his/her return on this investment. Optimization behaviour is assumed by individuals. This entails that that each person will invest accordingly based on the returns which will maximize his/her economic welfare or total earned income. Based on this assumption different investment levels by individuals exist in human capital, which are in turn associated with different earnings levels. In the model, individuals have heterogeneous marginal returns to investment and there exist heterogeneous supply curves of investments (Becker & Chiswick, 1966). The model is graphically represented in figure 1.

Figure 1 – Supply and demand for investment in human capital (Becker & Chiswick, 1966)

(10)

10 Multiple curves exists because the opportunities are not equal in a population. An individual might have a higher or lower willingness to forego consumption and will invest accordingly, parental wealth and income differences allow for unequal opportunities and the availability of scholarships and loans vary from person to person and from country to country. Furthermore, the supply curves assume that financing becomes more difficult when the amount invested increases due to factors such as higher interest, decreasing amounts of parental wealth available and reduced consumption during this period.

(11)

11 heterogeneous between individuals, human capital inequality is a result of many factors such as parental wealth, competition in college degrees, individual ability, risk tolerance and subsidies to education. Therefore, in the model, a higher level of human capital inequality is associated to higher income inequality. As stated above, policies can be implemented to increase the equality of opportunity, yet the marginal returns to education will be influenced by personal characteristics such as ability and risk tolerance, revealing why human capital inequality is likely to be always positive. Based on the theory provided above hypothesis 1 has been constructed:

H1: A higher level of human capital inequality is associated to higher income inequality

2.4 Empirical literature

In this subchapter the empirical literature on the topic is discussed. First, a thorough look is given on Castelló-Climent & Doménech (2020) which is the paper with the most similar analysis. This is followed by an overview of further empirical literature. After this overview is given, the limitations of each paper are discussed and solutions to account for these limitations are presented which are applied in this paper.

(12)

12 2002). They test the skill-biased technological change hypothesis in the period 2000-2014 for 33 countries, using the earnings gap, the relative supply of skills and the relative technology trend as variables. In this paper the skill-biased technological change hypothesis is not tested directly, however a control variable which proxies for technological change is used to control for the effect it has on the income distribution (Castelló-Climent & Domenech, 2020).

Furthermore, de Gregorio & Lee (2002) conducted empirical research on the relation between human capital inequality and income inequality on a panel set of countries in the period 1960-1990. They constructed the standard deviation of the educational distribution using 5 year intervals from the Barro & Lee (1996) dataset for human capital inequality and the Deininger & Squire (1996) dataset for income inequality. Specifically, they investigated the association of the educational distribution and mean educational attainment on the income distribution and found evidence for their hypothesis. Furthermore, they found evidence for Kuznets inverted-U hypothesis of the income level and income distribution (de Gregorio & Lee, 2002). The authors do investigate heterogeneity between continents, but do not do this for all continents, as they exclude for instance North-America. A more historical study in the 1970’s into determinants of income inequality by Ram (1984) found no evidence for the hypothesis of Becker & Chiswick (1966). Ram (1984) argues that a higher mean year of schooling level does equalize the income distribution, where he measures the income distribution by using the income share of the bottom 80 percent as proxy for income inequality, a measure which is not used in this paper.

(13)

13 expected as well as continental heterogeneity. In the empirical literature this is often overlooked. De Gregorio & Lee (2002) do not investigate within-country differences while they do add indicator variables for Africa, Asia and Latin-America, finding significant positive associations for the former two. Ram (1984) accounts for less developed countries in his empirical specification by using indicator variables, while not focusing on within-country differences. Third, the scope of this study is different from the other studies presented above, only Castelló-Climent & Doménech (2020) analyse a more similar period, their scope being 1950-2010, while de Gregorio & Lee (2002) investigate 1960-1990 and Ram (1984) his research period focuses mostly on the 1970’s. Finally, the dataset from CLIO-INFRA (2013) is used for measuring human capital inequality in this study, allowing for many observations as their data intervals are one year, instead of five years as in the Barro & Lee dataset which are used in de Gregorio & Lee (2002) and Castelló-Climent & Doménech (2020). Combined, these four points indicate the added value of this study in the literature.

Based on the theoretical literature of Chang-Tai & Klewow (2010) and the empirical findings of de Gregorio & Lee (2002) and Ram (1984) stronger or weaker associations are found for developing and developed nations due to variation in returns to education. Based on this hypothesis 2 is constructed which will be analysed through subsample analysis by constructing indicator variables for continents interacted with human capital inequality:

H2: The association between human capital inequality and income inequality is heterogeneous between continents

3. Methodology

3.1 Introduction

In this chapter the methodology and data collection method for the analysis is presented. It will start by explaining the data collection and the variable measurement of income inequality and human capital inequality. After this, control variables are explained. This is followed by the empirical specification and the empirical methods of the analysis. After this, potential endogeneity concerns are discussed. And finally, the descriptive statistics over the sample are presented.

3.2 Income inequality

(14)

14 measure is most suited for this research it is important to evaluate each of these measures against the four inequality principles: The anonymity principle, the population principle, the relative income principle and the Dalton principle (see Appendix B for a discussion of these principles and why they are important) (Ray, 1998).

The Palma ratio is defined as the ratio of the top 10 percent of incomes against the lowest 40 percent of incomes and is developed by José Gabriel Palma. He argues that the middle 50 percent has appropriated a share of 50 percent of national income, and that this is often true for most countries, while the ratio between the top 10 percent and the lowest 40 percent varies significantly and is therefore useful for comparability (Palma & Stiglitz, 2016). This measure is not used because it does not satisfy the Dalton inequality principle: If an individual in the 8th decile makes a progressive transfer to an individual in the 5th decile, inequality does not change in the measure, while it does in reality (Dalton, 1920).

The Atkinson index is a more social welfare approach to measure inequality. Atkinson argues that “Inequality cannot, in general, be measured without introducing social judgements” (Atkinson, 1975:47). Therefore he introduced a sensitivity parameter which ranges from zero to infinity, where zero entails being indifferent about the distribution of income while the latter is being concerned. The outcome of this measure can be used to compute the income needed to arrive at an equal social welfare level (Maio, 2007). However, subjective approaches for inequality measurement to investigate within-country differences are likely to attract a number of reliability concerns. Examples of concerns which arise when using a subjective measure for inequality are cultural differences between social groups and remaining objective. Furthermore, it is a time consuming process to evaluate each country in the analysis through their subjective social judgment framework, which directly results in a dataset with far less observations than the other measures. Finally, having no upper bound makes a measure less useful for comparison. For these reasons the Atkinson index is not used.

The coefficient of variation is calculated through dividing the standard deviation of the income distribution by the mean. In more equal societies the standard deviation will be smaller. Similarly as the Atkinson index this measure has no upper bound, making it less useful for comparison and extremely low or high values can influence the mean and the standard deviation as such that the measure is no longer appropriate (Maio, 2007).

(15)

15 be seen in figure 3.1. In this framework, one finds on the horizontal axis the cumulative percentage of the population, and on the vertical axis the cumulative percentage of income earned. In this framework, the 45 degree perfect distribution line represents an income inequality Gini-coefficient of 0, which means that all individuals earn the same income. The further the Lorenz curve line is bent from this 45 degree line, the higher the income inequality level in the population. In the extreme case one person would earn all the income in a population, the maximum Gini value of 1 would be attributed.The Gini-coefficient satisfies all four inequality principles (Ray, 1998:179). Moreover, it is by far the most widely used measure of inequality, therefore many rich datasets exits. For these reasons the Gini-coefficient is chosen as dependent variable to measure income inequality.

(16)

16

Figure 3.1 Lorenz Curve (Davis & Cobb, 2010)

3.3 Human capital inequality

(17)

17 based on the quality of education is not feasible, and the Gini-coefficient will be calculated based on educational attainment levels.

Fewer datasets exits for human capital inequality measured through educational attainment than for income inequality. Castelló-Climent & Domenech (2020) use the Barro & Lee (2013) dataset for measuring human capital inequality. However, the dataset is not the most comprehensive one available, as it measures 5 years intervals. The dataset from CLIO-INFRA (2013) is constructed by van Leeuwen, van Leeuwen-Li and Foldvari (2013) and combines several datasets on human capital inequality into one and offers data ranging from 1850-2010 on the population aged 15 years or older. The methodology is similar to the Barro & Lee (2013) dataset and it uses central statistical agencies data as inputs after 1960. Data prior to 1960 is based on estimates, yet these do not fall within the scope of this article. Using the CLIO-INFRA (2013) dataset over the Barro & Lee (2013) dataset allows for a considerable increase in observations, increasing the precision of the estimates. The calculation of the Gini-coefficient is done using the method by Thomas et al. (2000), Castelló & Domenech (2002) and Checchi (2004) . They use the following equation (Leeuwen, Leeuwen-Li, & Foldvari, 2013):

𝐺

_𝑒

=

1 2𝜇

∑ ∑|𝑥

𝑗

− 𝑥

𝑘

|𝑛

𝑗

𝑛

𝑘

(4)

3 𝑘=0 3 𝑗=0

In equation 4,

𝜇

represent the average years of schooling of the population aged 15 years or older, j and k represent different educational levels. xj

and x

k

are the cumulative mean average

(𝑛

1

+ 𝑛

2

3

)

(5)

(18)

18 maximum value is 100. A minimum value of 0 would indicate that there exists no inequality in education attainted in a population, while a value near the maximum of 100 would entail that nearly the entire population would have no educational attainment at all, while a select few would have the highest level of education (Leeuwen, Leeuwen-Li, & Foldvari, 2013) (Castelló & Doménech, 2002).

3.4 Control variables

The following control variables are added: Gross Domestic Product (GDP) per capita and the squared term of GDP per capita, trade openness, financial openness, democracy and research and developments (R&D) expenditure as percentage of GDP.

Kuznets (1955) hypothesized that in initial stages of economic development, inequality will rise since investment opportunities will only be fruitful for those who have anything to spend. Once developed, upward mobility is possible due to human capital accrual for individuals and their employment opportunities increase, decreasing income inequality (Kuznets, 1955). Due to this non-linear relationship, GDP per capita and the squared term of GDP per capita are added as control variables. Data for GDP per capita is obtained from the Penn World Tables 9.1 (Feenstra, Inklaar, & Timmer, 2015).

Trade openness is added as a control variable as proxy for measuring globalization. A higher level of trade openness decreases demand for low-skilled workers in developed countries, while increasing the demand for capital and the relative demand for high-skilled people, increasing income inequality. In skilled economies demand will increase for low-skilled workers in turn decreasing income inequality (Kanbur, 2000). Trade openness per country is measured by adding exports and imports and dividing this by its GDP and the data is obtained from the Penn World Tables 9.1. The effect trade openness has on income inequality is ambiguous, Kanbur (2000) argues that it increases income inequality due to the reasons described above, while Jaumotte, Lall & Papageorgiou (2013) argue that it decreases income inequality (Jaumotte, Lall, & Papageorgiou, 2013).

Income inequality is measured using the Gini-coefficient for variation in income. β0 is the constant term. β1 is human capital inequality, which is measured using the Gini-coefficient for human capital inequality. Human capital inequality is expected to have a positive association with income inequality. β2 and β3 are variables for GDP per capita and the squared term for GDP per capita, respectively. It is expected that β2 will increase income inequality, while β3 will decrease income inequality. β4 is a measure of trade openness and its effect on income inequality is ambiguous. β5 is a measure of financial openness and it is expected to increase income inequality. β6 is a measure of democracy and it is hypothesized that a higher level of democracy decreases income inequality. Finally, β7 measures R&D expenditure as percentage of GDP andis expected to increase income inequality.

(20)

20 model is that unobserved fixed effects which are correlated with the independent variables are accounted for. This could otherwise bias the estimates (Wooldridge, 2013:484-492). In a random effects model differences between countries are expected to be random, and are therefore not correlated with the independent variables as in fixed effects models. As the same countries are tracked over time it is more likely that a fixed effects model is appropriate. One can test which model is most appropriate by conducting a Hausman test. In the results section, for each model this test is conducted. The null hypothesis for the Hausman test indicates that there is no correlation between the error term and the independent variables. If enough evidence is found to reject the null hypothesis, a fixed effects model is consistent(Wooldridge, 2012:495).

3.6 Endogeneity

(21)

21 income inequality is, to the best of my knowledge, not feasible, and therefore the instrumental variable method is not used in this paper while lagged variables are implemented in the robustness check.

Another endogeneity issue which frequently appears in OLS regression analysis is omitted variable bias. With omitted variable bias one of the independent variables is omitted which is correlated with the dependent variable and with one of the independent variables, and this can either overestimate or underestimate the effect of the independent variable which is correlated with the omitted variable (Wooldridge, 2012:89). In the literature, much of the variation in income inequality is often left unexplained, therefore it is likely that in this paper omitted variable bias will cause endogeneity concerns. A method which is constructed and used in this paper to test for the presence of omitted variables in Pooled OLS models is the Ramsey Regression-Error Specification Test (RESET) (Wooldridge, 2012:306). The null hypothesis in the Ramsey RESET test states that the model has no omitted variables.A method to reduce the potential omitted variable bias is fixed effects regression analysis, because with country fixed effects, one potentially controls for country characteristics which are correlated with the independent variables, even if these are not observable. One further method to decrease the likelihood of omitted variable bias is including year fixed effects, to control for time trends (Wooldridge, 2012:484-492). In the results section Pooled OLS regression results are presented first, followed by either fixed or random effects regression models, and finally, to account for endogeneity concerns stemming from reverse causality, a lagged independent variable model is included.

Furthermore, multicollinearity might bias the results. With multicollinearity, independent variables are highly correlated with one another. In the descriptive statistics section attention is given to endogeneity concerns which might result from multicollinearity (Wooldridge, 2012:97).

(22)

22

3.7 Descriptive statistics

The dataset consists of an unbalanced panel set of 118 countries. The period of analysis is 1960-2010. Data for more countries on human capital inequality and income inequality is not available, nor is data available post 2010 for human capital inequality, or data on income inequality before 1960. A broad set of 118 countries is chosen from all continents to investigate whether there exist heterogeneity between continents and within countries. A list of the countries included in the analysis is presented in Appendix A. In table 3.1 descriptive statistics can be found. The dataset consists of 3340 observations over the period 1960-2010. In Appendix C the observations per year are presented. What becomes clear when looking upon Appendix A and C is that income inequality data availability improves as the years progress. At the beginning of the period of analysis only 3 countries have data available and this remains below 30 until 1972.

Table 3.1 – Descriptive statistics

Income inequality shows a mean value of 38. The United Nations see a Gini-coefficient of less than 40 as a goal for each country. The mean value of 38.15 indicates that on average income inequality has met that target. However, it remains very close to the target of 40 and the standard deviation and the maximum value tell many countries have not met that goal over the years (UNRISD, 2012). The minimum and maximum values are neither high nor low when one compares them with the human capital inequality values. The mean value of 31 for human capital inequality entails that on average inequality in educational attainment is not severe, yet the standard deviation explains that there is much variation between countries and between time periods. The minimum value is 4.033 and this is the human capital inequality level of Austria in 2007. This value entails that in Austria, in 2007, nearly the entire population attainted the same level of higher education. The maximum value of 93.223 is Burkina Faso in 1994, and this entails that a big proportion of the population had no education, while a select few had the highest level of education. Furthermore, the logarithms of GDP and the squared term of GDP

Variable Obs Mean Std.Dev. Min Max

(23)

23 reveal that there exits much variation between countries over time. For instance, the average logarithmic value of 8.8 of indicates huge dispersion between GDP per capita levels between countries over time. A mean trade openness (imports plus exports divided by GDP) value of 0.679 entails that on average a countries’ trade openness is more than two thirds of their GDP. The maximum value in the sample of 4.4 entails that a country is highly dependent on their exports and imports, while a minimum value of 0.085 entails that a country is not far from complete autarky from an international trade perspective. The mean value for financial openness of 0.679 entails that on average most countries are reasonably open when considering capital accounts. A maximum value of 1 entails the maximum degree of capital accounts openness, while the minimum of 0 entails no capital accounts openness at all. The minimum and maximum value for democracy are 0 and 14. Where 0 indicates no democracy and 14 indicates the highest score for democracy. An average value of 9.5 indicates that on average, a country in the analysis will be reasonably democratic, while the standard deviation shows that much dispersion exists between countries. The average value of 1.007 for R&D expenditure as percentage of GDP indicates that on average a country invests approximately 1 percent of their GDP in R&D. Here a maximum of 4.427 indicates that one country in a certain year invested 4.427 percent of their GDP in R&D. The standard deviation of 0.959 entails that even more dispersion between countries over time exists than for the levels of democracy.

Table 3.2 – Correlation Matrix

Furthermore, in table 3.2 a correlation matrix is presented to investigate whether endogeneity concerns stemming from multicollinearity might arise when doing econometric predictions. A first look at this table in column 1 tells us that trade openness might not have much explanatory power for the income distribution as its correlation coefficient is very low. Unsurprisingly, the logarithm of GDP per capita and the squared term of GDP per capita reveal

(24)

24 almost perfect collinearity, however this is not an issue as it is in line with the theory and a non-linear effect is expected.

Figure 3.2 – Distribution of Income Inequality Figure 3.3 – Distribution of Human Capital Inequality

Histograms of the income inequality variable and human capital inequality variable are presented in figure 3.2 and 3.3. Neither of these variables show a perfectly normal distribution, yet the skewness is acceptable to not transform either of them to logarithms. To observe whether logarithms or normal values are needed for the control variables histograms are constructed and presented in Appendix D. GDP per capita, the squared term of GDP per capita, trade openness and financial openness are transformed into logarithms, while the democracy index and R&D expenditure are not.

(25)

25 The main variables of interest are the Gini-coefficient for income and for human capital. In figure 3.4 and figure 3.5 both variables are regressed over the years 1960-2010 and presented with 95% confidence intervals. What becomes clear from both figures is that human capital inequality has decreased in the period 1960-2010, as shown in figure 3.4, while on the other hand, income inequality has increased in the period 1960-2010, represented in figure 3.5.

In the period 1960-2010, rapid economic development occurred all over the world. However, this was not an equal process where each country developed as much as other countries. The size of the confidence intervals of especially the human capital inequality evolution over time reveal this in figure 3.4. Additionally, figure 3.6 is presented where the trend in the human capital inequality is shown differentiated per continent. What becomes clear in the figure is a heterogeneous trend when one compares continents. In Europe and North-America a small decline occurred over the years in human capital inequality, which might be attributable to their initially high level of development. Furthermore, in Asia, Latin-America and Africa a stronger decline in human capital inequality occurred, which again, could be attributed to their initially low levels of development. Finally, Australia shows an unexpected trend, namely an increase in human capital inequality. A deeper look into the data reveals that only 3 countries from the continent are included in the analysis, explaining why the confidence interval is of this size. Guinea enters the analysis in 1991, with a human capital inequality level of 78, while on the other hand, Australia and New Zealand have a human capital inequality level of 13 and 19. This explains the unexpected trend for the continent of Australia.

(26)

26 inequality has increased the most. A deeper look upon the trend in Australia again reveals the Guinea is an outlier which causes this steep increase in income inequality.

Figure 3.6 – Evolution of human capital inequality per continent

Figure 3.7 – Evolution of income inequality per continent

(27)

1960-27 2010, differences between continents are likely to be observed. Therefore in the econometric analysis indicator variables are included to control for continent-specific time trends to see if there are any statistically significant differences between them.

As human capital inequality decreased worldwide and income inequality increased worldwide, one could perhaps speculate that the hypothesized relationship of this study does not exist. Yet the theory remains adamant, and the correct control variables should reveal this. Furthermore, figure 3.8 shows the correlation between the average value of human capital inequality in the period 1960-2010 per country on the y-axis and the average income inequality value in the period 1960-2010 per country on the x-axis, and yields a positive correlation coefficient of 0.33, represented by the red line.

Figure 3.8 – correlation between human capital inequality and income inequality

4. Results

4.1 Introduction

(28)

28 results, and finally, the robustness check to account for reverse causality is presented. An overview of the Stata commands used in each analysis is presented in Appendix F.

4.2 Pooled OLS regression results

Model 1 – 6 of table 4.1 represent the Pooled OLS regression results. Not all control variables have data availability going back to 1960. Therefore, in the third row from below of table 4.1 one can see the time period of each analysis, which states for which years the data was available for the analysis. Robust standard errors are used in all models because the Wooldridge test for autocorrelation is rejected, indicating that the models suffer from autocorrelation.

Model 1 shows the hypothesized association that a higher level of human capital inequality is associated with higher income inequality. The coefficient of 0.145 with a constant of 33.79 reveals that a one Gini point increase in human capital inequality is associated with an increase of income inequality from 33.79 Gini points to 33.935. The constant of 33.79 indicates that when human capital inequality would be zero, the average the level of income inequality in the sample is 33.79. The R-squared of 0.113 entails that 11,3% of the variation in income inequality is explained by human capital inequality. When controlling for the logarithm of GDP per capita and the squared version of GDP per capita the human capital inequality coefficient loses its significance represented in model 2, while even turning negative and significant on the 10 percent level in model 3, indicating that a higher level of human capital inequality is associated with lower income inequality. In these models, as well as in model 4 and 5, the inclusion of the logarithm of GDP per capita and the squared version of GDP per capita allow for the constant term to take a size which is unexpected. It remains unclear as to why these values are of this size. Both the logarithm of GDP per capita and the squared version of GDP per capita have the expected signs, while being highly significant in all models. However, the logarithms of trade openness and financial openness switch signs once included in the models and the levels of significance differ as well.

(29)

29 increasing human capital inequality with one point decreases income inequality with 0.253 points. In all other continents a significant and positive association is found. The association being the strongest in North-America and Latin-America, with a positive coefficient of 0.435 and 0.510, respectively.

In model 5 (where indicator variables for continents are excluded), and 6, a subsample analysis is conducted including R&D expenditure as percentage of GDP as control variable. R&D expenditure as percentage of GDP is highly significant and negative, in contrast to the hypothesis which states that it should increase income inequality. In model 5 human capital inequality is significant and positive with a value of 0.0958. Model 6 is combining the empirical specification of model 4 and 5. Again, Europe is the reference group and has a negative and significant association between human capital inequality and income inequality. In all other continents a significant and positive association is found.

(30)

30 Table 4.1 – Pooled OLS regression results

Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1

4.2 Fixed effects regression results

Pooled OLS might not be the appropriate regression analysis method, as stated in the previous subchapter. In table 4.2 the fixed effects regression results are presented. As stated in chapter 3.5 for each model a Hausman test is conducted to indicate whether fixed or random effects are appropriated. In all models the null hypothesis for the Hausman test is rejected, indicating that there is correlation between the error term and the independent variables and that a fixed effect model is appropriate (Hill, Griffiths, & Lim, 2012:559). Furthermore, the Wooldridge test for autocorrelation is conducted and rejected, indicating that the models suffer from autocorrelation. To account for this, robust standard errors are used in the fixed effect models. Year fixed effects have been included as well in models 1, 2, 3 and 5, as the F-test indicated there is enough evidence to reject the null hypothesis that all year coefficients are jointly zero.

Pooled OLS model (1) (2) (3) (4) (5) (6)

Income Inequality Educgini 0.145*** -0.00298 -0.0163* -0.253*** 0.0958*** -0.139*** (0.0066) (0.00849) (0.00981) (0.0251) (0.0230) (0.0461) lngdp 37.03*** 36.92*** 24.42*** 24.52*** 11.96*** (1.768) (1.831) (2.097) (3.415) (3.968) lngdpsq -2.356*** -2.387*** -1.582*** -1.515*** -0.748*** (0.101) (0.104) (0.121) (0.196) (0.221) lto 0.0912 0.898*** -2.620*** -0.690* (0.257) (0.236) (0.496) (0.388) lfo 0.784*** -0.709*** 2.154*** 0.0985 (0.249) (0.219) (0.533) (0.437) dmc 0.109* -0.115 0.141 (0.0602) (0.132) (0.128) rdexp -1.806*** -1.760*** (0.305) (0.281) Asia * educgini 0.186*** 0.249*** (0.0225) (0.0382) North-America * educgini 0.435*** 0.474*** (0.0197) (0.0239) Latin-America *educgini 0.510*** 0.570*** (0.0236) (0.0301) Australia * educgini 0.217*** 0.0862*** (0.0242) (0.0213) Africa * educgini 0.266*** 0.241*** (0.0241) (0.0432) Constant 33.79*** -103.6*** -99.58*** -53.38*** -59.08*** -12.01 (0.277) (8.737) (8.279) (9.280) (15.48) (18.42)

Year fixed effects Yes Yes Yes Yes Yes Yes

Time period 1960-2010 1960-2010 1970-2010 1980-2010 1996-2010 1996-2010

Observations 3,340 3,340 2,652 2,652 866 866

(31)

31 In models 4 and 6 there was not enough evidence to reject the null hypothesis and it is assumed that all year coefficients are jointly zero. The F-test for testing the significance of the year and interaction coefficients have been included in the models in table 4.2. In table 4.2, year coefficients have been dropped for a clear overview.

In model 1, the variable for human capital inequality is regressed against income inequality. The results reveal the correct sign, yet the coefficient is insignificant. The R-squared of 0.11 indicates that 11 percent of the variation within each country is explained by the model. In model 2 the logarithm of GDP per capita and the squared version are included. Neither control is significant, however the independent variable of interest, human capital inequality, is significant on the 10 per cent level. The coefficient of 0.0697 indicates that a one point increase in human capital inequality (e.g. from 30 to 31) is associated with an increase in income inequality of 0.0697 points, providing evidence for hypothesis 1. On average per country in the period 1960-2010, human capital inequality decreased with 5.5 Gini points, indicating that this change is related to a decrease of 0.383 income inequality Gini points. In the third model the analysis includes trade openness, financial openness and democracy as control variables. Again, as in model 2, neither of the control variables appear significant. However, the association of human capital inequality and income inequality is now larger in magnitude than in model 2, with a value of 0.0813 and is significant on the 10 percent level. This association is economically significant, as in the period 1970-2010 one standard deviation change in human capital inequality of 19.287 would be associated with a change in income inequality of 1.57 points, which is approximately 17 percent of the standard deviation of the Gini-coefficient of 9.104. In model 4, R&D expenditure as percentage of GDP is included. In this model human capital inequality loses its significance while R&D expenditure is significant and positive on the 10 percent level. Indicating that a one percentage point increase (e.g. from 1 percent of GDP to 2 percent of GDP) of R&D expenditure as percentage of GDP increases income inequality by more than 1 Gini point. A one standard deviation change of R&D expenditure as percentage of GDP (0.959) is related to a change of income inequality of 0.979 points.

(32)

(33)

33 Table 4.2 Random- and fixed effects regression results

4.3 Reverse causality robustness checks

As stated in chapter 3.4 there is a potential endogeneity issue stemming from reverse causality. In table 4.3 the results are presented when using 5 year lags of the human capital inequality regressed on income inequality. The results suggest that the findings of table 4.2 are robust. As all the coefficients for human capital inequality have the same sign and the magnitude of the association is more or less the same. The only difference being that in the period 1996-2010, a negative and significant association is found.

Fixed Effects (1) (2) (3) (4) (5) (6) Income inequality educgini 0.0441 0.0697* 0.0813* -0.147 -0.0102 -0.0924 (0.0433) (0.0417) (0.0427) (0.104) (0.134) (0.184) lngdp 6.199 3.245 1.392 2.931 -0.615 (5.128) (4.842) (6.123) (4.428) (5.476) lngdpsq -0.362 -0.161 -0.143 -0.158 -0.0303 (0.279) (0.259) (0.325) (0.245) (0.288) lto 0.0133 0.970 0.379 1.011 (0.696) (0.793) (0.619) (0.685) lfo -0.0331 0.542 -0.0420 0.661* (0.245) (0.436) (0.250) (0.389) dmc -0.0672 -0.0384 -0.113 (0.109) (0.0474) (0.104) rdexp 1.021* 0.749 (0.534) (0.501) Asia * educgini 0.0244 -0.250 (0.144) (0.219) North-America * educgini 0.168 0.684* (0.149) (0.374) Latin-America * educgini 0.217* 0.696* (0.129) (0.357) Australia * educgini 0.00190 -0.557** (0.256) (0.248) Africa * educgini 0.152 0.235 (0.131) (0.229) Constant 39.78*** 12.19 19.35 10.35 21.47 45.41* (4.001) (23.29) (22.25) (66.32) (20.30) (26.87) Hausman

Year fixed effects

(34)

34 (1) (2) (3) (4) Income Inequality l5.educgini 0.0574 0.080* 0.0923** -0.186** (0.0469) (0.0422) (0.0442) (0.0897) lgdp 8.467 6.801 -0.356 (5.380) (5.501) (5.452) lgdpsq -0.433 -0.333 -0.0685 (0.290) (0.293) (0.294) lto -0.0410 0.802 (0.776) (0.787) lfo 0.0374 0.540 (0.265) (0.441) Dmc -0.0800 (0.110) rdexp 0.985* (0.499) Constant 38.81*** -2.29 0.508 50.73** (3.782) (24.87) (25.46) (25.08) Hausman 0.000 0.000 0.000 0.000

Year fixed effects yes Yes yes no

Observations 2,760 2,760 2,335 856

Time period 1965-2010 1965-2010 1970-2010 1996-2010

R-squared 0.112 0.131 107 0.144

Countries 112 112 0.125 87

Table 4.3 - 5 year lags

While the results do suggest that no endogeneity concerns stemming from reverse causality are present, they have to be cautiously interpret. Reed (2015) argues that using lagged values of independent variables does not necessarily eliminate bias in the estimates and that one should use instrumental variable estimation (Reed, 2015). Furthermore, Bellemare, Masaki & Pepinsky (2017) argue in line with Reed (2015), that using lagged independent variables is still likely to provide biased estimates in the case that the model suffers from autocorrelation (Bellemare, Masaki, & Pepinsky, 2017).

5. Conclusion

(35)

35 with 3340 observations over the period 1960-2010. Using Pooled OLS and fixed effects regression methods with year fixed effects this paper finds evidence in favour of this hypothesis. For the period 1960-2010, a significant, positive and economically relevant association is found of human capital inequality on income inequality. For the period 1970-2010 the same conclusion arises, as a one point increase in human capital inequality is associated with a 0.0813 Gini point increase in income inequality. The findings are robust for endogeneity concerns stemming from reverse causality. In a subsample analysis researching continental differences from the period 1980-2010, a significant and positive association is found in Latin-America. While in the period 1996-2010, the association is significant and positive for North-America and Latin-America, being stronger in magnitude than in the baseline regression analysis. Furthermore, this paper provides evidence that in the period 1996-2010, technological change, proxied by R&D expenditure as percentage of GDP, increases income inequality.

While the findings of this paper are in line with the theory, much remains unclear. The study has the following limitations: First, the significance levels of human capital inequality regressed against income inequality are low, as they are significant on the 10 percent level, while in the period 1996-2010 no significant association is found. Secondly, most of the control variables are insignificant in the analysis, indicating that most of the variation in within-country income inequality remains unexplained. Thirdly, when focusing on continental differences, a significant positive association is found for Latin-America and North-America, while a negative association is found in Australia. The association of the latter might be explainable by the low sample size, yet it remains unclear why no association is found for Asia and Africa, as in these continents, compared to Latin-America and North-America, the sharpest declines in human capital inequality occurred. Finally, whether the results are completely robust against reverse causality remains doubtful, as argued by Bellemare et al. (2017) and Reed (2015). Combined, the four limitations provided above do cast some doubt in the overall robustness of the analysis.

(36)

Latin-36 America and North-America, while being absent in the other continents. Perhaps the correct way to research this would be to focus on one continent at a time, as much heterogeneity is present between them. Moreover, future research should focus into providing additional robustness test for reverse causality, for instance by using an instrumental variable approach. Finally, it may be doubtful whether the levels of human capital inequality can decrease beyond the perfect equality of opportunity equilibrium. Personal characteristics such as ability and risk tolerance lie at the base of human capital inequality, indicating that human capital inequality will always exists.

(37)

37

Bibliography

Achhoff, B., & Schmidt, T. (2008). Empirical evidence on the success of R&D cooperation -- happy together? Review of Industrial Organization, 33(1), 41-62.

Alderson, A., & Nielsen, F. (2002). Globalization and the Great U-Turn: Income Inequality Trends in 16 OECD Countries. American Journal of Sociology, 102(4), 1244-1299.

Becker, G., & Chiswick, B. (1966). Education and the distribution of Earnings. The American Economic

Review, 56(1), 358-369.

Bellemare, M., Masaki, T., & Pepinsky, T. (2017). Lagged explanatory variables and the estimation of causal effect. The Journal of Politics, 73(3), 949-963.

Benhabib, J., & Spiegel, M. (1994). The role of human capital in economic development: Evidence from aggregate cross-country data. Journal of Monetary Economics(34), 143-173.

Breusch, T., & Pagan, A. (1979). A Simple Test for Heteroscedasticity and Random Coefficient Variation. Econometrica, 47(5), 1287-1294.

Brinks, D., & Coppedge, M. (2006). Diffusion is No Illusion: neighbor emulation in the third wave of democracy. Comparative political studies, 39(4), 463-489.

Card, D., & DiNardo, J. E. (2002). Skill-Biased Technological change and Rising Wage Inequality: Some Problems and puzzles. Journal of Labor Economics, 20(4), 733-774.

Castelló, A., & Doménech, R. (2002). Human capital inequality and economic growth: Some new evidence. The Economic Journal, 112(2), 187-200.

Castelló, A., & Doménech, R. (2002). Human Capital Inequality and Economic Growth: Some new evidence. The economic Journal, 112, 187-200.

Castelló-Climent, A., & Domenech, R. (2020). Human Capital and Income Inequality: Some facts and

Some puzzles. Retrieved from Retrieved from BBVA Research: https://www. bbvaresearch.

com/wpcontent/uploads/migrados/WP_1228_tcm348-430101. pdf

Chani, M. I., Jan, S. A., Pervaiz, Z., & Chaudhary, A. R. (2014). Human Capital Inequality and Income Inequality: testing for causality. Quality and Quantity, 48, 149-156.

Chinn, M. D., & Ito, H. (2006). What matters for financial development? Capital controls, institutions, and interactions. Journal of Development Economics(81), pp. 163-192.

Dalton, H. (1920). The Measurement of the Inequality of Incomes. The Economic Journal, 30(119), 348-361.

Davis, G. F., & Cobb, J. A. (2010). Corporations and economic inequality around the world: The paradox of Hierarchy. Research in Organizational behaviour, 30, 35-53.

de Gregorio, J., & Lee, J.-W. (2002). Education and Income Inequality: New evidence from cross-country Data. Review of Income and Wealth, 3(48), 395-416.

Drukker, D. M. (2003). Testing for serial correlation in linear panel-data models. The Stata Journal,

(38)

38 Erosa, A., Koreshkova, T., & Restruccia, D. (2010). How Important Is Human Capital? A Quantitative

Theory Assessment of World Income Inequality. Review of Economic Studies(77), 1421-1449. EROSA, A., KORESHKOVA, T., & RESTUCCIA, D. (n.d.).

Feenstra, R. C., Inklaar, R., & Timmer, M. P. (2015). The Next Generation of the Penn World Table.

American Economic Review, 105(10), 3150-3182.

Freedom House. (2018). Freedom in the world 1973-2018. Retrieved from Freedomhouse.org: https://freedomhouse.org/report/freedom-world

Galbraith, J., Halbach, B., Malinowska, A., Shams, A., & Zhang, W. (2015). The Utip Global Inequality Datasets: 1963-2008. WIDER Working Paper 2015/019, Helsinki: UNU-WIDER.

Galor, O., & Zeira, J. (1993). Income Distribution and Macroeconomics. Review of Economic Studies,

60, 35-52.

Griliches, Z., & Lichtenberg, F. R. (1984). R&D and Productivity Growth at the Industry Level: Is There Still a Relationship? In Z. Griliches, R & D, Patents, and Productivity (pp. 465-502). Chicago: University of Chicago Press.

Heckman, J., & Todd, L. L. (2006). Earnings function, rates of return and treatment effects: The mincer equation and beyond. Handbook of the Economics of Education , 1, 307-458. Hill, R. C., Griffiths, W. E., & Lim, G. C. (2012). Principles of Econometrics. Asia: John Wiley & Sons. Hill, R. C., Griffiths, W., & Lim, G. (2012). Principles of Econometrics. Asia: John Wiley & Sons. Hsieh, C.-T., & Klenow, P. J. (2010). Development accounting. American Economic Journal:

Macroeconomics, 2(1), 207-223.

Hulten, C. R. (2001). Total factor productivity: a short biography. In C. R. Hulten, New developments

in productivity analysis (pp. 1-54). Chicago: University of Chicago Press.

Jarque, C. M., & Bera, A. K. (1987). A Test for Normality of Observations and Regression Residuals.

International statistical review, 55(2), 163-172.

Jaumotte, F., Lall, S., & Papageorgiou, C. (2013). Rising Income Inequality: Technology, or Trade and Financial Globalizaiton. IMF economic review, 61(2), pp. 271-309.

Kanbur, R. (2000). Income Distribution and Development. Handbook of Income Distribution: Volume

1, 791-841.

Kuznets, S. (1955). Economic Growth and Income Inequality. American Economic Review, 45(1), 1-28. Kuznets, S. (1955). Economic Growth and Income Inequality. The American Economic Review, 45(1),

1-28.

Leeuwen, B. v., Leeuwen-Li, J. v., & Foldvari, P. (2013). Educational Inequality Gini-coefficient. Retrieved from Clio.infra: http://hdl.handle.net/10622/KORKQW

Madsen, J. (2007). Technology spillover through trade and TFP convergence: 135 of evidence for the OECD countries. Journal of International Economics, 72(2), 464-480.

(39)

39 Meltzer, A., & Richard, S. (1981). A rational theory on the size of the government. Journal of political

economy, 89(5), 914-927.

Milanovic, B. (2016). Income Inequality is Cyclical. Nature(537), 479-482.

Nielsen, F., & Anderson, A. (1995). Income inequality, Development, and Dualism: Results from an Unbalanced Cross-National Panel. American Sociological review, 60(5), 674-701.

OECD. (2018). Income Inequality. Retrieved from Data.oecd.org: https://data.oecd.org/inequality/income-inequality.htm

Palma, J., & Stiglitz, J. (2016). Do nations just get the inequality they deserve? The 'Palma Ratio' re-examined. In S. J. Basu K., Inequality and Growth: Patterns and Policy. International

Economics Association (pp. 35-94). London: Palgrave Macmillan.

Ray, D. (1998). Development Economics. Princeton University Press.

Reed, W. R. (2015). On the practice of lagging variables to avoid simultaneity. Oxford bulletin of

Economics and Statistics, 77(6), 897-905.

Saint-Paul, G., & Verdier, T. (1993). Education, democracy and growth. Journal of Development

Economics, 42, 399-407.

Snower, D. (1999). Causes of Changing Earnings Inequality. IZA discussion paper, No. 29.

Solow, R. M. (1956). A Contribution to the theory of economic Growth. The Quarterly Journal of

economics, 70(1), 65-94.

Solt, F. (2019). Measuring Income Inequality Across Countries and Over Time: The Standardized World Income Inequality Database. SWIID Version 8.2.

Sylwester, K. (2002). Can education expenditures reduce income inequality? Economics of Education

Review(21), 43-52.

Thomas, V., Wang, Y., & Fan, X. (2000). Measuring Education Inequality - gini coefficients of education. Policy, Research working paper, WPS 2525.

UNRISD. (2012). Inequalities and the Post-2015 Development Agenda: A Concept Note. Inequalities

and the Post-2015 Development Agenda (pp. 1-4). Geneva: UNRISD.

UNU-WIDER. (2019). World Income Inequality Database (WIID4).

Wakelin, K. (2001). Productivity Growth and R&D expenditure in UK manufacturing firms. Research

Policy, 30(7), 1079-1090.

Wooldridge, J. M. (2012). Simultaneity Equations Models. In J. M. Wooldridge, Introductory

Econometrics - A Modern Approach (p. 554). South-Western: Cengage Learning.

World Bank. (2017). Development Research Group. Gini index (World Bank Estimate). World Bank. (2018). GDP per capita (current US$). Retrieved from data.worldbank.org:

https://data.worldbank.org/indicator/NY.GDP.PCAP.CD

(40)

40 World Bank. (2018). Research and development expenditure (% of GDP). Retrieved from World Bank:

https://data.worldbank.org/indicator/GB.XPD.RSDV.GD.ZS

Appendix A: Countries of analysis

1 Afghanistan

41 Germany

81 Nicaragua

2 Algeria

42 Ghana

82 Niger

3 Angola

43 Greece

83 Nigeria

4 Argentina

44 Guatemala

84 Norway

5 Armenia

45 Guinea

85 Pakistan

6 Australia

46 Haiti

86 Panama

7 Austria

47 Honduras

87 Paraguay

8 Azerbaijan

48 Hungary

88 Peru

9 Bangladesh

49 Iceland

89 Philippines

10 Barbados

50 India

90 Poland

11 Belarus

51 Iraq

91 Portugal

12 Belgium

52 Ireland

92 Romania

13 Benin

53 Israel

93 Rwanda

14 Botswana

54 Italy

94 Saudi Arabia

15 Brazil

55 Jamaica

95 Senegal

16 Bulgaria

56 Japan

96 Seychelles

17 Burkina Faso

57 Jordan

97 Sierra Leone

18 Burundi

58 Kazakhstan

98 Singapore

19 Cambodia

59 Kenya

99 South Africa

20 Cameroon

60 Kyrgyzstan

100 Spain

21 Canada

61 Latvia

101 Sri Lanka

22 Central African

Republic

62 Lebanon

102 Swaziland

23 Chad

63 Lesotho

103 Sweden

24 Chile

64 Liberia

104 Switzerland

25 China

65 Libya

105 Tajikistan

26 Colombia

66 Lithuania

106 Thailand

27 Congo

67 Madagascar

107 Togo

28 Costa Rica

68 Malawi

108 Trinidad and

Tobago

29 Cyprus

69 Malaysia

109 Tunisia

30 Denmark

70 Mali

110 Turkey

31 Dominican

Republic

71 Mauritania

111 Uganda

32 Ecuador

72 Mauritius

112 Ukraine

33 Egypt

73 Mexico

113 United

(41)

41

34 El Salvador

74 Morocco

114 United States

35 Estonia

75 Mozambique 115 Uruguay

36 Finland

76 Myanmar

116 Uzbekistan

37 France

77 Namibia

117 Zambia

38 Gabon

78 Nepal

118 Zimbabwe

39 Gambia

79 Netherlands

40 Georgia

80 New

Zealand

Appendix B: Measuring inequality

Anonymity principle. It should not matter who earns the income. It should at all

times be possible to arrange the income distribution so that they represent the ranking from poorest to richest, or vice versa. It should look like the following (Ray, 1998):

Y1 ≤ Y2 ≤ Y3 ≤ … ≤ Yn

Population principle. The size of the population should be irrelevant in order to

compare cross country inequality levels. The inequality level of for instance China with 1.3 billion citizens should be comparable with the inequality level of for instance the Netherlands, with 17 million citizens.

Relative income principle. Absolute levels of income should be irrelevant. A

population of 2 individuals with an income of $100,000 and $50,000 should through measurement have the same inequality level as the another population of 2 with incomes of $1,000 and $500.

Finally, there is the Dalton principle. This principle states that when a progressive or regressive transfer is made inequality should change. So when a relatively rich person

(42)

42

Appendix C: Observations per year

year Freq. Percent Cum.

(43)

43

Appendix D: Distribution of variables

Figure D3 – Distribution of GDP per capita Figure D4 – Distribution of Trade

openness

Figure D5 – Distribution of Financial Openness Figure D6 – Distribution of Democracy

(44)

44

Appendix E: Human capital inequality & Income Inequality

trends over time

Education & Income Gini 1960-1980

1960 1970 1980

Income Education Income Education Income Education

Europe Asia 32.4 21 28.6 21.9 27.5 19.1 N.A N.A 36.8 45.2 36.5 46.3 North-America N.A N.A 41.9 27.8 41.4 21.9 Latin-America 49.9 49.5 45.6 33.5 48.6 30.7

Australia N.A N.A 27.4 15.7 27.6 14.6

Africa N.A N.A 44.2 68.6 47.4 56.5

Continued 1990 2000 2010