The impact of economic growth on gender income inequality

BSc Thesis Economics and Business

The Impact of Economic Growth on Gender Income Inequality

Author: William Smith Supervisor: Naomi J. Leefmans

Student number: 11107456 January 2018

Statement of Originality

This document is written by William Smith who declares to take full responsibility for the contents of this document. I declare that the text and the work presented in this document is original and that no sources other than those mentioned in the text and its references have been used in creating it. The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.

Abstract

In every country in the world the GNI per capita of females is lower than the GNI per capita of males. This article aims to analyse the determinants of gender income inequality using Kuznets’ inverted-U hypothesis. This is done by the estimation of a fixed effects model using panel data from 168 countries between 2000 and 2015. The results suggest GDP is not a significant factor in determining gender income inequality and thus meaning the inverted-U relationship does not apply to gender income inequality. The analyses do show that gender inequalities in schooling and labour force participation rates and the Gini coefficient have a significant effect on gender income inequality.

Contents

1. Introduction 1

2. Literature Review

2.1 The Kuznets Curve 3

2.2 Previous Studies of the Relationship Between GDP and Gender Income 4 Inequality

2.3 Mechanisms Behind the Relation Between Gender Income Inequality 6 and Economic Growth

2.4 Reversed Causality 10

2.5 Other Determinants of Gender Income Inequality 12

3. Methodology and data description

3.1 Methodology 16

3.2 Data Description 19

4. Results

4.1 BPLM and Hausman Tests 21

4.2 Fixed Effects Model Results 22

4.3 Discussion of Other Regression Models 25

5. Conclusion 26

Bibliography 30

1. Introduction

Although the progression towards gender equality has been substantial in developed countries over the past decade, the gross national income per capita in every country remains lower for females than males. As seen in Figure 1 the general trend of gender income inequality in EU countries has been decreasing. In the United Kingdom (the line which starts at the earliest date), the gender wage gap (defined here by the difference between the median earnings of men and women relative to median earnings of men) has been consistently decreasing, dropping from 44.58% in 1970 to 16.84% in 2016. Gender inequality is a pertinent topic as reducing it is one of the main aims of the United Nations Development Programme (UNDP, 2016, p.41). Also, it has been part of the European Employment Strategy and policy since 1999 and, in 2003, targets were formulated “to achieve by 2010 a substantial reduction in the gender pay gap in each Member State” (Plantega & Remery, 2006, p.4).

Figure 1: Gender wage gap for EU countries. Source: https://data.oecd.org/earnwage/gender-wage-gap.htm

In European countries, economic development is relatively high. Kuznets’ inverted-U hypothesis states that, in developed countries, societal income inequality is expected to decrease but, in less developed countries, economic growth brings rising income inequality (Kuznets, 1955, pp.7-18). Lantican, Gladwin, and Seale (1996, p.250) later proposed that this hypothesis could also be used to model gender income inequality. This paper will look at a wide selection of countries at varying stages of development whilst, at the same time, expanding on previous research into the effects of economic development on gender income inequality by investigating the question: to what extent does economic growth affect gender income inequality and can the effects be modelled by Kuznets’ Inverted-U hypothesis? The paper will aim to answer this question by using a fixed effects regression model to analyse panel data from 168 countries over eight time periods spanning from 2000 to 2015. This paper adds value to previous research as the regression uses more recent data from a wider range of countries than used in previous research. Also, the model used in this paper controls for gender labour force participation rate inequality which is not done in previous research. As outlined in section 2.3, Evenson (1982, p.213) and Tam (2011, p.141) suggest the correlation between economic growth and gender income inequality is caused by the U-shaped relationship between GDP and gender labour force participation rate inequality. Controlling for this will indicate whether this mechanism is plausible reasoning for the correlations found in previous research. Answering my research question will lead to an insight into the determinants of gender income inequality by analysing the correlation between GDP per capita (from here onwards GDP will refer to GDP per capita, unless specified otherwise) and GNI per capita for males and females. The use of these variables to describe the level of economic development and gender income inequality was decided due to the extensive availability of data on these measures. If economic development is found to be significant, it will highlight its disadvantages. This will serve to

better inform governments when producing policies to prevent the development of gender income inequality.

The structure for this paper is as follows. Section 2 will discuss in-depth the previous research relating to this paper’s research question. Section 3 will introduce the data, method, and specify the hypothesis for this research. In section 4 the results of the study are presented and discussed. In section 5 a conclusion will be made containing theoretical implications as well as limitations of this research and future research ideas.

2. Literature Review

In this section, research concerning the Kuznets curve will be presented followed by research linking it to gender income inequality. Secondly, there will be a review of previous research similar to that of this paper. Then, there will be a discussion about the mechanisms behind this correlation and finally, research into other factors that affect gender income inequality.

2.1 The Kuznets Curve

Kuznets (1955, pp.7-18) first proposed the relationship between economic development and initially rising income inequality and subsequently, in developed countries, the inequality’s narrowing again. The proposed cause for this is change in the industrial structure; the shift from low-productivity sectors to high-productivity sectors during time of initial growth, thus increasing income inequality (Kuznets, 1955, p.7). Later in the economic development, high-productivity sectors begin to dominate and wage inequality decreases (Thornton, 2001, p. 15). When the economy is at an early stage of development people migrate from rural areas to urban areas where income is higher but less evenly distributed (Lantican et al., 1996, p.236). Over

time income in these urban areas will become more equal as the migrants’ descendants become integrated into the workforce (Lantican et al., 1996, p.236).

Some authors, however, suggest that Kuznets’ hypothesis does not hold when tested using time-series data. Papenek and Kyn (1986, p.61) found that there is a relatively fast flattening of the Kuznets curve over time. Li, Squire, and Zou (1998, p.42) found using panel data from 84 countries, that income inequality was relatively stable a 40-year period. This suggests there is little relationship between economic development and income inequality. Tam (2008, p.387) and Bruno, Ravallion, and Squire (1998, p.21) also found Kuznets’ hypothesis failed to hold when using panel data.

Lantican, Gladwin, and Seale (1996, p.250) proposed that Kuznets’ inverted-U hypothesis could be used to model gender income inequality. The explanation offered for this is that women are heavily discriminated against in the early stages of economic development, meaning that job opportunities for women rise more slowly than for men and consequently resulting in a lower labour force participation rate for women (Lantican et al., 1996, p.239). As countries reach the later stages of the Kuznets curve, overall wages increase causing women to invest in skills and enter full-time employment reducing the income inequality. The authors performed a regression of gender inequality on GNP per capita for a set of Asian countries at different stages of economic development which ascertained that gender inequalities in urban areas did fit Kuznets’ inverted-U curve (Lantican et al., 1996, p.256).

2.2 Previous Studies of the Relationship Between GDP and Gender Income Inequality In this section, previous papers which conducted similar regressions to the one in this paper will be discussed.

Lantican et al (1996, p.252) investigated the effect of GNP on different aspects of gender inequality in Bangladesh, the Philippines and South Korea; a set of Asian countries all

having developed to different extents over the past three decades. When looking at education inequality they found that GNP and the square of GNP were significant determinants when controlling for the percentage of the male-female ratio of the labour force working in agriculture, fitting Kuznets’ inverted-U hypothesis. They controlled for the gender inequality in the labour force participation rate in the agriculture sector as it reflects the level of structural transformation of the economy away from agriculture and towards manufacturing which should in turn reduce inequality (Lantican et al., 1996, pp.245). The given mechanism behind this was that poor Asian parents traditionally gave priority to schooling of their male children but in the latter stages of economic development the offering free elementary schooling expanded meaning the monetary constraint on parents educating their children was loosened. This lead to an increase in the female primary enrolment rate, reducing the enrolment gap (Lantican et al., 1996, pp.252-523). This favouritism and change in schooling structure may not be reflected outside of these countries so there is limited external validity when applying this mechanism to the data used in this paper.

Ndinga (2011, p.44) used panel data on a sample of 41 African countries to observe the effect of GDP and GDP squared on gender income inequality whilst controlling for inequality in schooling, overall income inequality and poverty, finding GDP and GDP squared to be significant at a 1% level. Ndinga (2011, p.45) suggests that there are two types of labour market in Africa; the informal market (lower wages) and the formal market (higher wages). When economic development is low, men and women are both forced into the informal market due to lack of opportunities in the formal market. When the economy grows, men gain more opportunities than women in the formal market due to discrimination causing gender income inequality to rise. Despite this mechanism, in the regression run, the coefficient of GDP was negative and GDP squared was positive, meaning the relationship cannot be modelled by the Kuznets curve (Ndinga, 2011, p. 46). This suggests that income inequalities will not reduce

over time, only increase. The reason given for this pattern is that as the economy develops, the paradigm shifts from men working whilst women are at home to men working for higher pay whilst women work for lower pay therefore the income gap fails to reduce (Ndinga, 2011, p. 46). This study is limited by only including African countries, of which most are at a low stage of development compared to countries used in this paper. Including data from countries that are at a further stage of economic development may cause a different relationship to be found.

Haas (2007, p.128) conducted a similar regression, expanding away from African countries, using cross-sectional data on 121 countries from 2005. The regression controlled for the disparity in literacy rates for males and females and the Gini coefficient. Haas (2007, p.133) found only the Gini coefficient significant at a 1% significance level, the mechanism behind this will be explained in section 2.5. GDP and GDP squared were almost significant at a 5% significance level (p-values of 0.55 and 0.67 respectively) and the coefficient of GDP was positive and GDP squared was negative which shows the relationship fits that of the Kuznets curve (Haas, 2007, p.133). The mechanism used for their significance is that women often produce homemade items for sale and, as the economy grows, this industry declines to give way to large scale manufacturing, creating more opportunities for men and raising their wages whilst causing women to become unemployed and lowering the female labour force participation rate (Haas, 2007, p.127). This relates to the informal job market Ndinga (2011, p.45) suggested. As the economy continues to grow, the service sector provides more job opportunities for women, increasing their labour force participation rate and reducing the gender income inequality (Haas, 2007, p.127).

2.3 Mechanisms Behind the Relation Between Gender Income Inequality and Economic Growth

In this section, mechanisms explaining the causality of the relationship between GDP and gender income inequality will be discussed. In the first two paragraphs, Oostendorp (2009, p.142) looks at the effects of globalisation. Then, Evenson (1982, p.213) and Tam (2011, p.141) evaluate the effects of the female labour force participation ratio. Finally, Saito, Mekonnen, and Spurling (1994, p.45) postulate the effect of technological advancement.

Oostendorp (2009, p.142) states that globalisation increases the gender wage gap for developed countries. For developed countries, trade theory suggests that when trade becomes freer due to globalisation, the price of scarce factors of production falls. In the export sector, where female labour is a scarce factor, female wages fall (Oostendorp, 2009, p.142). The Stolper-Samuelson theorem, based on the assumptions of the Heckscher-Ohlin model of trade, states that the prices of factors of production will be fully equalised between trading countries when there is free trade (Burtless, 1995, p.804). This means wages will be equalised for equivalent work. When there is no free trade, workers in developing countries have a lower wage than the equivalently skilled workers in developed countries. Burtless (1995, p.804) states that the skills of workers in developing countries are equivalent to the lower skilled workers’ in developed countries. Based on this assumption, if female workers are less skilled than male workers, female wages in developed countries will decrease more dramatically as trade increases with developing countries (Oostendorp, 2009, p.142). Hence, the gender wage inequality increases. This mechanism suggests that when developed countries increase trade, gender income inequality increases. Thus implying that the Kuznets curve does not hold for gender income inequality.

Oostendorp (2009, p.142) also posits that, due to the increased competition caused by globalisation, wages will fall especially for females if they are disproportionately highly employed in sectors that face competition from foreign “cheap” labour. This is concurred by research done by Wood (1991, p.171), finding that, in developing countries between the early

1960s and the mid 1980s, the demand for female workers in the manufacturing sector increased due to pressure on wages from global competition. This is also presented to be the case by Seguino (2000, p.1211). In her regression, she finds that gender inequality is positively correlated with economic growth in countries where the export sector employs predominantly female workers. In this paper, however, it is assumed that the causality is reversed and gender inequality encourages economic growth (Seguino, 2000, p.1223). Ertürk and Çağatay (1995, p.1971) state that, when some export-oriented countries reach later stages of development they start to produce more sophisticated products and compete less on price. This requires higher skilled workers meaning that women are substituted by men in the labour force, thus reducing women’s labour force participation rate and increasing gender income inequality (Ertürk & Çağatay, 1995, p.1971).

Evenson (1982, p.213) posits that the relationship between GDP and gender income inequality can be explained through females’ time allocation. The author states that in the early stages of development wages rise, but women are strongly discriminated against and a female’s real opportunities rise slower than a male’s. This causes a relative fall in women’s wages compared to men’s, leading to a fall in the female labour participation rate in this period (Evenson, 1982, p.214). This leads to lower wages for females relative to males as there are fewer women committing to market-orientated and on-the-job training (Mincer & Polachek, 1974, p.80). Also, it causes women to obtain less labour market experience causing female wages for decrease relative to male wages (Blau & Kahn, 2000, p.81). A more rapid or continuing period of economic growth creates greater expectations for the future and increases the perceived value of time (Evenson, 1982, p.214). This raises the perceived opportunity cost of time causing women to forgo larger families to invest more in skills and pursue full-time employment leading to an increase in the female labour force participation, raising relative wages compared to men (Lantican et al., 1996, pp.239-240).

Tam (2011, p.141) also offered reasoning as to why the labour force participation rate causes gender wage inequality. The female labour force participation rate was regressed on the log of GDP and the log of the square of GDP and it was found that there was a U-shaped relationship (Tam, 2011, p.142). When the overall level of income is low, the agricultural sector dominates the market and women therefore have a high labour force participation rate and gender income inequality is low (Tam, 2011, p.141). Standing (1978, p.514) proposes an income effect which states that a woman’s wage is a negative function of overall wages and that, when income grows beyond a certain level, women’s participation will fall. This is explained by an increase in the fertility rate (Tam, 2011, p.141). Also, as machinery is introduced to the farming sector due to economic growth, men displace women in the labour market causing a substitution effect (Tam, 2011, p.141). As development continues, mental human capital becomes more important than physical human capital, reversing the substitution effect and raising females’ relative wages (Tam, 2011, p.141). As mentioned above, due to the higher wages, women forgo larger families, reducing the fertility rate and reversing the income effect (Lantican et al., 1996, p.239).

Saito, Mekonnen, and Spurling (1994, p.45) suggest that females may be less productive due to their limited access to technology. This was applied to farming jobs which are low paid and traditionally dominating in countries on the left side of the Kuznets curve. The explanation behind this is that economic development leads to increasing specialisation of labour which changes farming from traditionally female to more male farming systems in low development countries as hand hoes are replaced with ploughs and farming becomes more intensified (Lantican et al., 1996, p.239). This displaces women from the agricultural sector leaving them at the bottom of the labour market hierarchy due to their lower levels of formal education and training (Lantican et al., 1996, p.239). In accordance to the Solow growth model technological progress leads to economic growth (Solow, 1994, p.48). With the greater use of

the technology, male productivity will outstrip female productivity as the economy grows (Saito et al., 1994, p.45). Thus, lower productivity for women leads to a lower wage (Benavot, 1989, p.15).

2.4 Reversed Causality

As mentioned previously, research by Seguino (2000, p.1211) and Ertürk and Çağatay (1995, p.1969) suggests gender income inequality causes economic growth. The mechanism through which this occurs is as follows. Given women’s high concentration in the export market, there is a low share of income in that market being payed to workers which indicates profitability for the firm (Seguino, 2000, p.1214). In Kaleckian models, investment is a function of profitability and output, implying that a low share of income going to workers stimulates investment (Seguino, 2000, p.1214). As the export sector has high price elasticities, a reduction in women’s wages lowers the unit cost of goods produced. This in turn increases demand whilst lowering the share of income going to the workers, stimulating investments and therefore leading to economic growth (Seguino, 2000, p.1214). To test this Seguino (2000, p.1221) regressed investment as a share of GDP on difference between the log of female earnings and the log of male earnings whilst controlling for the variance of GDP growth and inflation rates. The two control variables were found to have significant, negative effects on investment due to their reflection of economic uncertainty and the gender wage gap had a significant, positive effect on investment (Seguino, 2000, p.1221). This contrasts with research which suggests that income inequality discourages investment as it causes social conflict which creates uncertainty (Seguino, 2000, p.1222). Seguino (2000, p.1222) posits that the latter is not applicable in the case of gender income inequality in the sample of countries used because women are less inclined than men to protest income inequality sufficiently to discourage

investment. A criticism of Seguino’s work extends the sample he used, commenting that differences in male and female productivities were not accounted for when looking at the wage gap; continuing to say that, if no correlation between gender income inequality and economic growth exists, restricting women’s education and access to the labour market in general (which increases the gender income differences) is detrimental for economic growth as it restricts the pool of talent (Schober & Winter-Ebmer, 2009, p.1481).

Ertürk and Çağatay (1995, p.1975) argue that females being concentrated in the export sector will only result in economic growth if the effect of this concentration on investment is higher than its effect on savings. In the Kaldorian model, savings is determined by the rate of capacity utilisation and the intensity of female household labour (Ertürk & Çağatay, 1995, p.1972). The intensity of female household labour moves in tandem with feminisation of the labour force because participation in the labour force usually results in women working harder in both the market and the household (Ertürk & Çağatay, 1995, pp. 1969-1973). Also, the intensity of female household labour is inversely related to household income (Ertürk & Çağatay, 1995, p.1973). This means that, when the real wage is low, the intensity of female household labour is high, resulting in savings being high. During economic expansion defeminisation occurs, the intensity of female household labour decreases and savings decline. Investment demand is derived from changes in the rate of capacity utilisation in the short term. In the long term, investment demand is cost-determined (Ertürk & Çağatay, 1995, p.1972). A high rate of feminisation in the export market is associated with lower labour costs and, therefore, positively related to investment demand in the long term. When economic development is low, female labour force participation is high which stimulates investment. This encourages the propensity to invest to rise above the propensity to save, leading to economic growth. Simultaneously, the high female labour force participation rate leads to a high intensity of female household labour which positively effects savings, depressing

economic growth (Ertürk & Çağatay, 1995, p.1973). Females being concentrated in the export sector will only cause economic growth if investment rises more than savings.

The aforementioned mechanisms propose the risk of simultaneous causality to the regression in this paper. The values of GDP and the square of GDP will therefore be lagged to prevent this from causing bias and inconsistent estimators. As the level of GDP of a previous period is a determinant of GDP in the current period it is a suitable instrument to use as an indicator of the effect of the current level of GDP on gender income inequality. The level of gender income inequality cannot affect the level of GDP in previous periods and therefore the risk of simultaneous causality is removed. The same reasoning can be applied for the of lagging the square of GDP.

2.5 Other Determinants of Gender Income Inequality

This section describes previous research conducted on other factors that influence gender income inequality of which the first two will be used as control variables in the regression in this paper. The first is gender inequality in schooling. The second is the overall income inequality of a country. Last is the effect of discrimination.

Data suggests enrolment ratios at all levels of schooling are lower for females in developing countries (Hill & King, 1995, p.21). Education increases cognitive ability and therefore productivity of workers (Benavot, 1989, p.15). In countries where females have less expected education their productivity is relatively lower resulting in lower wages (Benavot, 1989, p.15).

Herz and Khandker (1991, p.59), in a study focusing on Peru, found that a one percent decrease in the differences between male and female enrolment rates at the primary education

level reduced the gender wage gap by five percent in Peru as a whole and a one percent reduction at the secondary education level reduced the gender wage gap by seven percent in urban areas. It is argued that if females’ education level increases it will increase their commitment to formal labour, causing them to have a more continuous employment and work more hours which would reduce the gender pay gap (Benavot, 1989, p.15). Herz and Khandker (1991, p.49) found that the probability that women would join the labour market increased by more than five percent if both men and women completed secondary education. This would increase the incentive to employers to offer more on-the-job training to females, increasing their wages (Mincer & Polachek, 1974, p.80). Herz and Khandker (1991, p.54) found that private rates of returns to schooling are higher for women than for men. Having a secondary education increases women’s wages by 17 percent in urban areas and 13 percent in rural areas whereas men’s wages only increase by nine percent in rural areas (Herz & Khandker, 1991, p.53). Behrman and Deolalikar (1995, p.115) found that women in Indonesia also had a higher return to schooling than men. This may be because of the disequilibrium in the labour market caused by lower investment in female education so a female with secondary education is disproportionally highly educated compared to other females (Behrman & Deolalikar, 1995, p.115). Hill and King (1995, p.21) showed that the female-to-male enrolment ratio is higher for more developed countries. As increased schooling for females has a positive correlation with economic growth, as countries develop, women’s education levels increase and women’s wages grow faster than men’s, thus lowering the gender income gap (Hill & King, 1995, p.29). Another variable that effects gender income inequality is the overall income inequality in a country. The variation in wages paid for various skills or employment in specific sectors, the “wage structure”, reflects the overall income inequality in a country but also effects the gender income inequality (Blau & Kahn, 1996, p.29). For example, the higher the return on experience in a country’s wage structure, as the human capital model suggests- assuming

women have less experience than men (due to their shorter, more discontinuous participation in the labour force according to Mincer and Polachek (1974, p.80))- the larger gender income inequality will be (Blau & Kahn, 2000, p.81). Another factor is the premium paid on specific occupations. If higher premiums are paid on male dominated jobs in one country than another then that country will have a higher gender pay gap (Blau & Kahn, 1996, p.30). Blau and Kahn (2000, p.79) investigated the influence of female dominated jobs on the mean hourly female to male wage ratio of full time workers. Using data from 1978, 1988 and 1998 in the United States, the authors presented that this ratio increased as the concentration of women in specific, traditionally female-dominated, relatively low-payed, jobs decreased. The female segregation into these jobs could be attributed to a couple of causes. Traditionally, women in the household spend more time on housework than men and these extra hours worked at home may decrease the energy put into work so women may have lower relative productivities to men (Becker, 1985, p.36). Women, on average, accumulate less labour market experience than men as they have shorter, more discontinuous employment (Mincer & Polachek, 1974, p.80). There are therefore lower incentives for employers to offer females training and lower incentives for females to commit to market-orientated and on-the-job training leading hence to lower wages relative to men (Mincer & Polachek, 1974, p.80). Blau and Kahn (2000, p.79) presented that, in the US in the 1970s, 53% of women worked in predominantly female jobs with less than 20% in managerial roles. In 1999, 41% of women worked in these predominantly female jobs and 45% of those worked in managerial roles (Blau & Kahn, 2000, p.79). During this time, the female to male hourly wage ratio increased by 0.117 for 18 to 24 year olds, 0.148 for 25 to 34 year olds, 0.172 for 35 to 44 year olds and 0.134 for 45 to 54 year olds showing a significant increase in women’s relative wages not only due to the entry of new cohorts of female workers but also because of a relative wage increase for older, longer incumbents of the labour market (Blau & Kahn, 2000, p.78).

Blau and Kahn (1996, p.30) suggested that the more unionised an economy is the lower the variation in wages for two reasons. Firstly, the use of collective bargaining tends to reduce inter-industry and interfirm variation in wages, reducing overall wage inequality and indirectly reducing gender wage inequality. Secondly, as women’s wage distributions are below men’s, if a centralised system raises minimum wage it will cause the gender income gap to decrease. The authors conducted a regression focusing on the US, as it has a higher overall income inequality than other countries in the sample. They looked at the effect of male wage inequality and the decentralisation of male wage setting on the gender pay gap and found that both had a significant effect (Blau & Kahn, 1996, p.56). They concluded that gender wage inequality increased with the level of wage inequality in the US and that the level of gender inequality would be similar to the level in Sweden as the level of overall wage inequality was the same for both countries (Blau & Kahn, 1996, p.56).

Another factor that could have an impact on gender income inequality is discrimination against women. Wright and Ermisch (1991, p.518) used data from Britain in 1980 to investigate the effect of discrimination against women in the labour market. They found that a minimum of 44% of the wage gap could be attributed to discrimination and women’s wages would be 20% higher if there were no discrimination (Wright & Ermisch, 1991, p.519). This is influenced by women’s having a lower probability of being promoted (Blau & DeVaro, 2007, p.511). This suggests that women’s wages are relatively lower than men’s because they are less represented in high paying jobs (Lazear & Rosen, 1990, p.108). One explanation could be that women are less likely to be appointed to jobs with high promotion prospects and therefore tend to work in ‘dead-end jobs’ (Groot & Brink, 1996, p.225). This links to the previous point of female dominated jobs being lower paid because of the exclusion of women from the high promotion prospect jobs leading to an excess supply of labour for female dominated jobs and thus decreasing female wages (Blau & Kahn, 2000, p.81). Lazear and Rosen (1990, p.106)

suggest the discrimination comes when choosing who is awarded a promotion, not what jobs women work in. They propose that women require a higher skill level than men to get a promotion, therefore leading to lower skilled men being in the same level of job and thus, assuming there is no difference in gender ability distribution, it is more likely males will be awarded promotions (Lazear & Rosen, 1990, pp.106-108). As discrimination is not easily measurable, the use of a fixed effects regression is intended to control for the differences in the level of discrimination found in each country.

3. Methodology and Data description

In this section the methodology used in the statistical analysis will be described and information on the data for the variables used in this research will be presented and defined.

3.1 Methodology

This method expands on previous research by using more and newer data whilst also taking several econometric concerns into consideration. The model used is more comprehensive than previous studies as it controls for more variables in an attempt to prevent omitted variable bias and contains lagged variables to avoid simultaneous causality. The relationship between gender income inequality and GDP will be modelled by the fixed effects regression:

Iit = αi + β1GDPi,t-1 + β2(GDPi,t-1)2 + β3Si,t + β4Li,t + β5Ginii,t + uit

Where the variables are determined by:

I=Male Gross National Income per capita Female Gross National Income per capita3

L=Labour force participation rate male Labour force participation rate female3 GDP, GDP2_{, and Gini are self-explanatory.}

The expected signs for these variables are given in Figure 2. The expected signs dictate that an increase in GDP will increase the ratio of male to female GNI per capita until GDP is high enough that the value of the square of GDP will to cause I to decrease, giving the inverted-U relationship.

Figure 2: Expected coefficient signs.

A pooled-OLS model ignores country specific effects as there is one value of αin the regression. A suggestion for an individually specific effect in this paper may be the cultural and political factors in countries which affect discrimination as outlined in section 2.5. Fixed effects and random effects models allow for each country to have a different intercept (αi)

which differ for each country but do not change over time eliminating risk of omitted variable bias which would lead to biased and inconsistent estimators.

The Breush-Pagan-Lagrange multiplier (referred to from here as BPLM) test will be performed to ascertain if the variance of the unobserved fixed effects is significantly different to zero. The hypotheses tested are:

H0: σ2u=0

Independent variable Expected sign

GDP +

(GDP)2 _-

S +

L +

H1: σ2u ≠ 0

If H0 is rejected then there is significant evidence to suggest that the pooled-OLS model

is not an accurate model for this research and a random effects or fixed effects model should be utilised instead.

The Hausman test will be done to determine whether there is a significant difference between the random effects model and the fixed effects model. This tests whether the individual unobserved effects are correlated with the included independent variables (denoted as X). The hypotheses tested are:

H0: cov(αi,X)=0

H1: cov(αi,X)≠0

If the test is insignificant, both the estimators of the random effects model and fixed effect model will be consistent. The random effects model is more efficient so should be used. If the null hypothesis is rejected, the random effects model will give inconsistent estimators and therefore should be substituted for the fixed effects model.

This paper hypothesises that the relationship between gender wage inequality and GDP per capita will resemble the inverted-U shape proposed by Kuznets and the coefficients of GDP and the square of GDP will be significant. The hypotheses tested are:

(1) H0: β1=0 H1: β1>0

(2) H0: β2=0 H1: β2<0

The significance of the GDP coefficient will suggest the correlation between GDP and gender income inequality and a significant GDP squared term will cause the relationship to be quadratic as needed to resemble the Kuznets curve.

3.2 Data Description

All data and data definitions in this paper are retrieved from the United Nations Development Programme (UNDP), Human Development Data (2017) database as it provides the most comprehensive data for the variables in this research. This database has data on 185 countries from 1990 to 2015. This paper will use panel data from 168 countries over eight time periods (2000, 2005, 2010, 2011, 2012, 2013, 2014, 2015) due to the availability of data on the variables used. The countries span across four levels of Human Development (very high, high, medium, low) which reflect different levels of economic development as GDP is a determinant of the Human Development figure (UNDP, 2016, p.213).

Variables showing gender income inequality differ, but this research will use the ratio of gross national income per capita of males to females (denoted as I), adjusted for 2011 purchasing power parity, expressed in international dollars. It is “derived from the ratio of female to male wages, female and male shares of economically active population and gross national income” (UNDP, 2016, p.213). This data was available for 174 countries for all eight time periods.

The level of economic growth is measured by the value of GDP per capita, a measure of all goods and services produced in a country for a given year expressed in international dollars using 2011 purchasing parity rates, which is a variable commonly used to analyse economic growth. This data was collected on 185 countries from the eight time periods specified above, with some values missing due to the availability of data. As mentioned section 2.4, there is a risk of simultaneous causality in my regression which would cause bias and inconsistent estimates. Oostendorp (2009, p.142) posits that as a country opens up to international trade which increases GDP, income inequality increases implying the causation of GDP affecting gender income inequality. Ndinga (2011, p.45) and Haas (2007, p.135) apply the Kuznets curve theorem to gender income inequality and assume the same direction of

causality. However, Seguino (2000, p.1223) suggests that gender income inequality causes economic growth because the lower the female wage, the lower the costs are for exports due to the high concentration of female workers in the export sector. For this reason, GDP may be an endogenous variable so the values for GDP and GDP squared will be lagged by one period to prevent the measurement errors being autocorrelated. GDP and GDP squared will be denoted as GDP and GDP2 respectively and the lagged variables will be denoted as lGDP and lGDP2. The values for GDP were available for 185 countries with 27 values missing over the eight time periods.

Data on the education level is given by the mean years of schooling of men and women aged 25 and older (denoted by SM and SF respectively). These people are of an appropriate age to be active in the labour market so their attainment levels are a suitable indicator of the inequality in education between men and women. Schooling inequality will be defined by the ratio of mean years of schooling of males to the mean years of schooling for females, denoted by S. The data for S was available from the 168 countries, although there are 44 observations missing in the eight time periods.

The female and male labour force participation rates are given by the percentage of the working age population (15 or older) that is active in the labour market, either working or looking for work. The gender inequality in labour force participation (denoted by L) will be expressed by the ratio of male labour force participation (denoted by LPM) to the female labour force participation ratio (denoted by LPF). The data for L is collected from 175 countries for all eight time periods. Controlling for the gender inequality in labour force participation will indicate which mechanism outlined in section 2.2 is the possible causal effect. Research by Tam (2011, p.142) suggested the labour force participation of women has a U-shaped relationship with GDP and is inversely correlated with gender income inequality. Neither Haas (2007, p.132) nor Ndinga (2011, p.44) controlled for the inequality in gender labour force

participation rates. If L is found to be significant and lGDP and lGDP2 are found to be insignificant then it indicates that the changes in L with economic growth is the determining factor of gender income inequality. If lGDP or lGDP2 are found to be significant despite controlling for the labour force participation rate then it means not all of the effect of economic growth has been captured by L and, therefore, the causal effect could be one of the other mechanisms outlined in section 2.2.

The overall income inequality of the countries will be specified by the Gini coefficient which measures the deviation of the distribution of individual or household incomes within a country from a perfectly equal distribution. A value of 0 signifies absolute equality, where everyone earns the same income. A value of 1 signifies absolute inequality, where one individual obtains all of the income and all other individuals obtain none. The data on the Gini coefficient (denoted by Gini) was available on 183 countries with 24 values missing over the eight time periods.

4. Results

In this section, the results from the regression described in section 3.1 will be reported. Firstly, there will be an explanation of the tests used to determine that the fixed effects model is most suitable. Then, there will be a discussion of the results of the fixed effects model. Next, there will be a brief discussion about the results from different regression models. Significance is judged at a 1% significance level unless stated otherwise.

4.1 BPLM and Hausman Tests

As can be seen in Table 6 in the appendix, the p-value of the BPLM test is 0.0000 which is strongly significant meaning that there is sufficient evidence to reject the null hypothesis and conclude that the pooled-OLS model is not appropriate as the variance of the unobserved

individual effects is significantly different to zero and so one of the individual effect models should be used instead. The Hausman test also has a p-value of 0.0000 (as seen in Table 7 of the appendix). This means that there is sufficient evidence to reject the null hypothesis meaning the individual unobserved effects are correlated with the independent variables and the fixed effects model should be used.

4.2 Fixed Effects Model Results

All values used in this section can be found in Table 5 of the appendix, unless specified otherwise. The p-values found in Table 5 of the appendix will be halved for this analysis as the hypotheses for this paper permit a one-sided t-test and the p-values found in the Stata tables are given by a two-sided t-test. Under the fixed effect model, lGDP and lGDP2 are insignificant (lGDP2 would be significant for any significance level higher than 5.15% but this is not accurate enough to be taken into account in this paper) and S, L, and Gini are significant. The rho value (0.99521763) is very high. It means that approximately 99.5% of the variation is explained by individual specific effects indicating that an individual effects model is suitable for this research.

The p-value of lGDP was 0.354 which is insignificant thus meaning GDP was not found to be a significant determinant of gender income inequality, contrary to previous research. Ndinga (2011, p.44) found that GDP was significant at a 1% level. The expansion of data to countries outside of Africa may have caused this as the countries included by Ndinga are predominantly at a relatively early stage of economic development. This is supported by Haas (2007, p.133) who found GDP to be insignificant at a 5% significance level in a regression which included data on countries from varying levels of economic development.

At a 5% significance level, lGDP2 is almost significant. If it were to be considered significant, it would imply that, when GDP increases by 10000 in a country then in the next

time period, the gender income inequality ratio would increase by 0.00596. This implies that gender income inequality increases with GDP and never decreases, meaning the relationship does not follow the inverted-U shape of the Kuznets curve. Ndinga (2011, p. 46) also found the coefficient of the square of GDP to be positive. The reason given is that when African countries develop, the shift is from men working and women doing housework to men entering well paid work and women entering the job market in lower paid jobs. This means there is a paradigm shift, but gender income inequality persists. At this significance level Haas (2007, p.133) found GDP and the square of GDP to be significant which is inconsistent with the findings of this paper. This may be caused by the use of cross-sectional data by Haas (2007, p.128) and the use of panel data in this paper. As mentioned, when using intertemporal data there is a relatively fast flattening of the Kuznets curve meaning its stability is ambiguous when using panel data.

The effect of schooling inequality has a p-value of 0.004 meaning it is significant. The implied effect is that if, within a country, the inequality in years of schooling goes up by one year, then gender income inequality in that country increases by approximately 0.132. This is in line with research conducted by Herz and Khandker (1991, p.59), which focused on Peru, in which they found that a reduction in the school enrolment inequality led to a reduction in the gender pay gap. This study expanded on that research by adding more countries and found the that effect moved in same direction, showing the external validity of said study.

The ratio of male to female labour force participation rate has a p-value of 0.000 meaning it is significant. The implied effect is that, if the ratio increases by one unit within a country, meaning the male labour force participation rises relative to female labour force participation rate, then gender income inequality rises by approximately 0.247. This is consistent with the theoretical work proposed by Mincer and Polacheck (1974, p.80), in which

they posited that a relative decrease in female labour force participation ratio results in lower wages relative to men.

The Gini coefficient has a p-value of 0.000 meaning it is significant. The implied effect is that a 0.1 increase in the Gini coefficient of a country causes a decrease of approximately 0.118 in the gender income inequality ratio. This is inconsistent with previous research in which evidence was given that the overall income inequality in a country is positively related to gender income inequality (Blau & Kahn, 1996, p.29). Haas (2007, p.133) found the Gini coefficient to be significant, also with a p-value of 0.000, but with a positive coefficient. The difference may be caused by the use of a fixed effects model correcting for omitted variable bias as Haas (2007, p.132) used an OLS regression. In the pooled-OLS regression run in this paper, the Gini coefficient was significant, with a positive coefficient (as can be seen in Table 3 of the appendix). A reason for this unexpected negative correlation may be that the coefficient is bias due to simultaneous causality. If women are concentrated in the lower paid job markets then an decrease in women’s wages would increase gender income inequality and, in turn, increase overall income inequality causing the Gini coefficient to rise.

In research done by Haas (2007, p.132) and Ndinga (2011, p.44), neither controlled for gender labour force participation rates in their regressions which may have caused omitted variable bias. Haas (2007, p.133) found both GDP and the square of GDP to be significant at a 7% significance level and Ndinga found both to be significant at a 1% significance level. In this paper both GDP and the square of GDP were found to be insignificant and the inequality in gender labour force participation was found to be significant. This indicates that the mechanism that causes the correlation between GDP and gender income inequality is based around the correlation between economic growth and female labour force participation rates. As outlined in section 2.2, this correlation was researched by Tam (2011, p.142) stating that, when there are high female labour force participation rates (which is inversely related to L),

gender income inequality is low. This is consistent with the positive correlation between I and

L found in this paper.

4.3 Discussion of Other Regression Models

Table 1 presents the coefficients from the three regression models run obtained from Tables 3,4, and 5 in the appendix. The values in brackets are the standard errors.

Table 1

Regression Results

Note. * represents significance at a 1% significance level, ** represents significance at 5%,

and *** represents significance at 10%. If there is no asterisk it means that the variable is insignificant at a 10% significance level.

I Pooled-OLS Random Effects Fixed Effects

lGDP -7.79e-06* (2.07e-06) -1.73e-06 (3.04e-86) 1.35e-06 (3.59e-06) lGDP2 9.87e-11* (l.74e-11) 6.06e-11** (2.77e-11) 5.96e-11 (3.65e-11) S 0.2540944* (.0335021) 0.0758427*** (.0452939) 0.1315911* (.0492768) L 1.174992* (.013277) 1.01956* (.0292836) 0.2471642* (.0603549) Gini 0.7439472* (.1597752) -0.0800644 (.2338609) -1.18437* (.3219811) R-Squared 0.9163 0.9125 0.4913

The first thing to note is the very small values for the coefficients of lGDP and lGDP2. This is to be expected due to the mean of I being 2.034444 and the mean of lGDP being 16848.68 (from table 2 in the appendix). The pooled-OLS model is the only model in which

lGDP is significant. Although the coefficient is negative which is inconsistent with previous

theory outlined in this paper, along with the positive value of lGPD2, it is consistent with the results of Ndinga (2011, p. 46). Ndinga (2011, p.44) used a random effects model. The coefficient signs found when using the random effects model in this paper agree with the findings of Ndinga (2011, p.44) which may indicate the author suffered from model misspecification.

The adjusted R-squared value of 0.9157, from Table 3 in the appendix, indicates that the pooled-OLS model is a good model for explaining the variance of I. From table 5 in the appendix, the R-squared within of the fixed effects model (given by the percentage of within variation that can be explained by the model) is 0.1061 means 10.61% of the variation within a country over time can be explained by the model.

5. Conclusion

This study aimed to examine how economic development influences gender income inequality. More specifically, using panel data, it focused on testing whether this relationship could be modelled by Kuznets’ inverted-U hypothesis. This is done by analysing 168 countries over eight time periods spanning from 2000 to 2015 using a fixed effects regression model.

Overall, the results of this study show that GDP does not have a direct effect on gender income inequality, contradicting previous literature conducted by Lantican, Gladwin, and Seale (1996, p.250), Haas (2007, p.133) and Ndinga (2011, p.37). The results, however, do concur with previous research into the fallacy of the Kuznets curve when applied to panel data as described by Tam (2008, p.387) and Bruno, Ravallion, and Squire (1998, p.21). The results

indicate that GDP and the square of GDP have no significant effect on gender income inequality when controlling for the gender inequalities in schooling, labour force participation rate and the overall income inequality. Therefore, the null hypotheses for hypotheses 1 and 2 are accepted, implying that gender income inequality cannot be modelled by Kuznets’ inverted-U hypothesis.

The inclusion of control variables resulted in a number of interesting points of analysis, the data for which is found in Table 5 of the appendix. Firstly, the gender inequality in schooling was found to have a significant positive effect on gender income inequality which coincides with previous research conducted by Herz and Khandker (1991, p.59) which focused on a single country, Peru. This paper contributes to the external validity of said research as it enlarged the sample size to 168 countries varying in their levels of economic development. The Gini coefficient was found to be significant but with a negative coefficient which contradicts the findings of Haas (2007, p.133) and Blau and Kahn (1996, p.29). Blau and Kahn (1996, p.56), however, used male wage inequality as a proxy for overall income inequality, whereas this paper used the Gini coefficient which may not have reflected wage structure in the same fashion.

The gender inequality in labour force participation rate was found to have a significant, positive effect which is in line with research by Evenson (1982, p.213), Tam (2011, p.141) and Lantican et al., 1996, pp.239-240). Haas (2007, p.132) found Kuznets’ hypothesis fit the relationship between GDP and gender income inequality whilst not controlling for labour force participation rates. The causality may therefore be attributed to the gender inequality in labour force participation as it has a U-shaped relationship with GDP and an inverse relationship with gender income inequality (Tam, 2011, p.142). Thus, as GDP increases, the labour force participation rate of females falls, leading to the gender income inequality gap’s widening. As GDP rises further, the value of working increases, causing women’s labour force participation

to increase, lowering gender income inequality and, therefore, giving the inverted-U relationship between GDP and gender income inequality (Lantican et al., 1996, p.239). This paper, therefore, proposes that the causality of the relationship between economic development and gender income inequality found in previous research is either caused by the mechanism proposed by Evenson (1982, p.214) or the mechanism proposed by Tam (2011, p.141) as discussed in section 2.2. Evenson (1982, p.214) proposes that when developing economies grow causing wages to rise, women are strongly discriminated against. This leads to male wages growing faster than female wages and, therefore, causing the male labour force participation rate to increase relative to that of female. When wages grow further, there is a higher perceived value of time leading to women forgoing leisure time to take regular jobs, reducing gender income inequality and producing the inverted-U relationship between GDP and gender income inequality. Alternatively, Tam (2011, p.141) explains the inverted-U relationship by the female labour force participation rate’s being high in developing countries due to the agricultural sector dominating, in which females are highly concentrated. As the economy grows, machinery is introduced into farming and men displace women in the labour market causing gender income inequality to widen. Mental human capital becomes more important than physical human capital as development continues, leading to the opposite effect and raising females’ relative wages.

A limitation of this study is the difficulty in constructing a comprehensive regression model due to the many determinants of gender income inequality of which some are hard to measure, such as discrimination. The use of a fixed effects regression model attempted to solicit for this possible omitted variable bias although it only controls for time invariant, country specific, factors. With cultural and political landscapes changing over time, discrimination may not be time invariant and would therefore cause omitted variable bias in this model.

For future research, it would be interesting to investigate the statistical evidence behind the theoretical mechanisms proposed between the gender inequality in labour force participation and gender income inequality. Tam (2011, p.141) and Evenson (1982, p.214) both focused on the relationship between income and labour force participation rates and theoretically argued this has a causal effect on gender income inequality. Statistical evidence for this is yet to be extensively researched.

Overall, this study shows that the relationship between gender income inequality and economic growth cannot be modelled by Kuznets’ inverted-U hypothesis when controlling for gender inequalities in labour force participation rate, schooling and overall income inequality. This is an important finding because it extends current literature and provides insights for governments as to the determinants of gender income inequality. From this study, it becomes clear that economic growth does not have to be dampened to prevent gender income inequality as it does not have a direct effect. Policy used to reduce inequality in the other areas analysed, however, have been shown relevant in reducing gender income inequality.

Bibliography

Becker, G. S. (1985). Human capital, effort, and the sexual division of labour. Journal of labor

economics, 3(1, Part 2), S33-S58.

Behrman, J. R., & Deolalikar, A. B. (1995). Are there differential returns to schooling by gender? The case of Indonesian labour markets. Oxford Bulletin of Economics and

Statistics, 57(1), 97-117.

Benavot, A. (1989). Education, gender, and economic development: A cross-national study. Sociology of education, 14-32.

Blau, F. D., & DeVaro, J. (2007). New evidence on gender differences in promotion rates: An empirical analysis of a sample of new hires. Industrial Relations: A Journal of Economy

and Society, 46(3), 511-550.

Blau, F. D., & Kahn, L. M. (1996). Wage structure and gender earnings differentials: an international comparison. Economica, S29-S62.

Blau, F. D., & Kahn, L. M. (2000). Gender differences in pay(No. w7732). National bureau of economic research.

Bruno, M., Ravallion, M., & Squire, L. (1998). Equity and growth in developing countries: old and new perspectives on the policy issues. Income distribution and high-quality

growth, 117-46.

Burtless, G. (1995). International trade and the rise in earnings inequality. Journal of economic

literature, 33(2), 800-816.

Ertürk, K., & Çağatay, N. (1995). Macroeconomic consequences of cyclical and secular changes in feminization: An experiment at gendered macromodeling. World

Development, 23(11), 1969-1977.

Evenson, R. E. (1982). The allocation of women's time: An international comparison. Behavior

Groot, W., & Van Den Brink, H. M. (1996). Glass ceilings or dead ends: Job promotion of men and women compared. Economics Letters, 53(2), 221-226.

Haas, S. (2007). Economic development and the gender wage gap. The Fifth Annual Carroll

Round, 126.

Herz, B. K., & Khandker, S. R. (1991). Women's Work, Education, and Family Welfare in

Peru. World Bank Discussion Papers 116. The World Bank, Publications Sales Unit,

Department F, 1818 H Street, NW, Washington, DC 20433..

Hill, M. A., & King, E. (1995). Women's education and economic well-being. Feminist

Economics, 1(2), 21-46.

Kuznets, S. (1955). Economic growth and income inequality. The American economic

review, 45(1), 1-28.

Lantican, C. P., Gladwin, C. H., & Seale Jr, J. L. (1996). Income and gender inequalities in Asia: Testing alternative theories of development. Economic Development and

Cultural Change, 44(2), 235-263.

Lazear, E. P., & Rosen, S. (1990). Male-female wage differentials in job ladders. Journal of

Labor Economics, 8(1, Part 2), S106-S123.

Li, H., Squire, L., & Zou, H. F. (1998). Explaining international and intertemporal variations in income inequality. The Economic Journal, 108(446), 26-43.

Mincer, J., & Polachek, S. (1974). Family investments in human capital: Earnings of women. Journal of political Economy, 82(2, Part 2), S76-S108.

Ndinga, M. M. A. (2012). Gender income inequality and development in Africa: Analysis based on Kuznets inverted U curve. Journal of African Studies and Development, 4(2), 37-50.

Oostendorp, Remco H. "Globalization and the gender wage gap." The World Bank Economic

Papanek, G. F., & Kyn, O. (1986). The effect on income distribution of development, the growth rate and economic strategy. Journal of Development Economics, 23(1), 55-65. Plantenga, J., & Remery, C. (2006). The gender pay gap. Origins and policy responses. A

comparative review of thirty European countries. Synthèse du rapport pour la

Commission Européenne, Equality Unit.

Saito, K. A., Mekonnen, H., & Spurling, D. (1994). Raising the productivity of women farmers

in Sub-Saharan Africa (Vol. 230). World Bank Publications.

Schober, T., & Winter-Ebmer, R. (2011). Gender wage inequality and economic growth: is there really a puzzle?—a comment. World Development, 39(8), 1476-1484.

Seguino, S. (2000). Gender inequality and economic growth: A cross-country analysis. World

Development, 28(7), 1211-1230.

Solow, R. M. (1994). Perspectives on growth theory. The Journal of Economic

Perspectives, 8(1), 45-54.

Standing, G. (1978). Labour force participation and development. Geneva: International Labour Office.

Tam, H. (2008). An economic or political Kuznets curve?. Public Choice, 134(3), 367-389. Tam, H. (2011). U-shaped female labor participation with economic development: Some panel

data evidence. Economics Letters, 110(2), 140-142.

Thornton, J. (2001). The Kuznets inverted-U hypothesis: panel data evidence from 96 countries. Applied Economics Letters, 8(1), 15-16.

United Nations Development Programme. (2016). Human development report 2016: Human

Development for Everyone. Retrieved from

http://hdr.undp.org/sites/default/files/2016_human_development_report.pdf

United Nations Development Programme. (2017). Estimated gross national income per capita,

PPP$) [data file], Expected years of schooling, female (years) [data file], Expected years of schooling, male (years) [data file], Income inequality, Gini coefficient [data

file], Gross domestic product (GDP) per capita (2011 PPP $) [data file], Labour force

participation rate, female (% ages 15 and older) [data file], Labour force participation rate, male (% ages 15 and older) [data file]. Retrieved from http://hdr.undp.org/en/data

Wood, A. (1991). North-South trade and female labour in manufacturing: An asymmetry. The

Journal of Development Studies, 27(2), 168-189.s

Wright, R. E., & Ermisch, J. F. (1991). Gender discrimination in the British labour market: a reassessment. The Economic Journal, 101(406), 508-522.

Appendix

Afghanistan Albania Algeria Andorra Angola Antigua and Barbuda Argentina Armenia Australia Austria Azerbaijan Bahamas Bahrain Bangladesh Barbados Belarus Belgium Belize Benin Bhutan Bolivia (Plurinational State of) Bosnia and Herzegovina Botswana Brazil Brunei Darussalam Bulgaria Burkina Faso Burundi Cabo Verde Cambodia Cameroon Canada Central African Republic Chad Chile China Colombia Comoros Congo Congo (Democratic Republic of the) Costa Rica Croatia Cuba Cyprus Czech Republic Côte d'Ivoire Denmark Djibouti Dominica Dominican Republic Ecuador Egypt El Salvador Equatorial Guinea Eritrea Estonia Ethiopia Fiji Finland France Gabon Gambia Georgia Germany Ghana Greece Grenada Guatemala Guinea Guinea-Bissau Guyana Haiti Honduras Hong Kong, China (SAR) Hungary Iceland India Indonesia

Iran (Islamic Republic of) Iraq Ireland Israel Italy Jamaica Japan Jordan Kazakhstan Kenya Kiribati Korea (Republic of) Kuwait Kyrgyzstan Lao People's Democratic Republic Latvia Lebanon Lesotho Liberia Libya Liechtenstein Lithuania Luxembourg Madagascar Malawi Malaysia Maldives Mali Malta Mauritania Mauritius Mexico Micronesia (Federated States of) Moldova (Republic of) Mongolia Montenegro Morocco Mozambique Myanmar Namibia Nepal Netherlands New Zealand Nicaragua Niger Nigeria Norway Oman Pakistan Palau Palestine, State of Panama Papua New Guinea Paraguay Peru Philippines Poland Portugal Qatar Romania Russian Federation Rwanda

Saint Kitts and Nevis

Saint Lucia

Saint Vincent and the Grenadines Samoa

Sao Tome and Principe Saudi Arabia Senegal Serbia Seychelles Sierra Leone Singapore Slovakia Slovenia Solomon Islands South Africa South Sudan Spain Sri Lanka Sudan Suriname Swaziland Sweden Switzerland Syrian Arab Republic

Tajikistan Tanzania (United Republic of) Thailand The former Yugoslav Republic of Macedonia Timor-Leste Togo Tonga Trinidad and Tobago Tunisia Turkey Turkmenistan Uganda Ukraine United Arab Emirates United Kingdom United States Uruguay Uzbekistan Vanuatu Venezuela (Bolivarian Republic of) Viet Nam Yemen Zambia Zimbabwe Table 2 Summary Statistics

Table 3

Regression Results of the Pooled-OLS Model

Table 4

Table 5

Regression Results of the Fixed Effects Model

Table 6

Table 7

The Hausman test