THE RELATIONSHIP BETWEEN INCOME INEQUALITY AND CORRUPTION: THE ROLE OF FEMALE REPRESENTATION IN GOVERNMENT

(1)

THE RELATIONSHIP BETWEEN

INCOME INEQUALITY AND

CORRUPTION: THE ROLE OF FEMALE

REPRESENTATION IN GOVERNMENT

Master Thesis

Msc Economic Development & Globalization

ABSTRACT

Using a fixed effects model and panel data from 104 countries over the period of 1998-2017, this paper investigate the relationship between corruption and income inequality. Furthermore, this study introduces a new potential channel through which corruption influences inequality, namely, through the share of women in parliament. The empirical findings, show that corruption is positively associated with income inequality, globally. However, this relationship becomes statistically insignificant when the model is regressed using the GMM estimator. This result should therefore be interpreted with caution and these limitations should be borne in mind. Furthermore, the comprehensive dataset reveals heterogeneity in the data. Findings show a positive statically significant relationship between corruption and inequality for Sub Saharan Africa countries and OECD countries. These findings are robust to different measures of inequality, additional control variables and different estimation procedures. Moreover, this study finds no evidence regarding the proposed channel of transmission.

Keywords: Corruption, Income inequality, Female representation in government, Panel data

Name: Kalina Petrovski

Student number: 1870017

Supervisor: dr. M.V. Nikolova

Co-assessor: dr. M.D. Laméris

(2)

1 Introduction

The great discrepancy between the wealthiest and the rest of the population has received overwhelming attention. In recent decades, countries have experienced an increase in the income share of the top percentiles of the income distribution, also known as the “wealthiest” or the “rich” (Piketty, 2007). The share of income held by the top percentiles are important determinants of the widespread observed increase in overall income inequality within countries. Empirical studies on the topic have increased due to the fact that data on the top income share is now available for many countries covering many years (Piketty, 2001; Atkinson, 2004; Alvaredo et al. 2013). Piketty’s book named Capital in the Twenty-First Century (2014) brought the debate regarding income inequality to a broader public. The main issue of this debate is that the rich people have been becoming richer over the last couple of years. The increasing income gap within countries creates questions, particularly with respect to the reasons for the persistence of income disparity. However, a key potential cause of the persistence in income inequality, of which we have limited knowledge of, is the effect of corruption on the top income share.

The destructive effect of corruption on society and economy, is something that has been frequently described within the literature of economics. Moreover, the World Bank classifies corruption as the greatest barrier to political, economic and social development (World Bank, 2003). Furthermore, the World bank estimates that, globally, bribes by itself costs more than 1 trillion USD each year and that corruption in public procurement involves additional costs of 1.5 trillion USD (Samanta et al., 2016). This confirms that corruption represents a major issue globally. Corruption has indirect and direct effects on governance and economic factors, which in turn exacerbates inequality. The literature establishes several channels through which corruption influence or aggravates inequality. Nevertheless, no academic research so far has investigated whether the share of women in government influences the relationship between corruption and inequality. This is unfortunate since it seriously restrict our knowledge of the direct and indirect influences corruption might have on income inequality and the possible policy actions to control this.

(4)

“Does representation of women in government mediate the relationship between corruption and inequality?”

This research will contribute to the existing literature by shedding light on the influence of women representation in government on the relationship between corruption and income inequality. To the best of my knowledge, this paper will be the first study examining in-depth this relationship. Moreover, previous studies on the corruption-inequality nexus are very limited and, mainly due to lack of data, have usually been focused on small samples of countries with a regional focus. What this thesis adds to the existing empirical literature is a much greater dataset, I examine 104 countries over the period 1998-2017 and therefore capture a more divers group of countries across a longer time period than any previous research. The estimation results in this study are thus more representative than previous studies The larger sample size and number of countries increases the external validity of the findings of this paper. Another contribution this paper makes is it uses the top income share as a dependent variable while all previous research on the relationship between corruption and income inequality used the Gini coefficient as a proxy of income inequality.

Results show that corruption it positively associated with income inequality globally. However, this relationship becomes statistically insignificant when the model is regressed using the GMM estimator. This result should therefore be interpreted with these limitations in mind. Furthermore, the comprehensive dataset reveals heterogeneity in the data. Findings show a positive statically significant relationship between corruption and inequality for Sub Saharan Africa countries and OECD countries. These findings are robust to different measures of inequality, additional control variables and different estimation procedures. Moreover, this study find no evidence with respect to the proposed channel of transmission.

(5)

2 Literature Review

Even though the impacts of corruption have been largely seen as a barrier to economic development, a large number of studies on this matter has concentrated on the relationship between corruption and economic growth (Mauro, 1997; Tanzi, 1998; Mo, 2001; Méon and Sekkat, 2005; Aidt et al., 2008; Swaleheen, 2011; Gründler and Potrafke, 2019, among others). Surprisingly, only a limited number of studies investigate the distributional impacts of corruption.

2.1 The Relationship Between Corruption and Income Inequality

Economic literature has tested the relationship between corruption and income inequality using a variety of data sources and techniques, but the research in this field is still limited. The majority of the studies that examined this relationship demonstrated a positive relationship, which means that a higher level of corruption is correlated with a higher level of income inequality within countries. Moreover, the argument is that the rich get away with more in comparison with the poor, in that way increase the income gap and consequently make the poor even poorer (Tanzi, 1995).

In one of the first research, Gupta et al. (2002) examined the effect of corruption on income inequality for a small sample of developing countries, using different indicators as proxies for corruption and the Gini coefficient as a proxy for income inequality. They used data from 1980–1997. Their results demonstrated that corruption has a positive effect on income inequality. Additionally, Batabyal, and Chowdhury (2015), Policardo et al. (2019), and Li et al. (2000) found the same result, that corruption increases income inequality. Meanwhile, Gyimah-Brempong (2002) examined the effect of corruption on income inequality and economic growth, using OLS and IV estimation methods. The panel dataset consisted of 21 African countries over the time period of 1993–1999, with the results showing that corruption positively affected income inequality. Conforming the findings of these prior studies, Dwiputri et al. (2018) studied the effect of corruption on income inequality in Asian countries and found that corruption has a significant effect on income inequality. Based on OLS and 2SLS methods, their results showed that a higher level of corruption is associated with a higher level of income inequality.

(6)

corrupt practices, or both (Gyimag-Brempong and Munoz de Camacho, 2006). Since the size of Latin American and African economies is relatively small, it is possible that the negative effects of corruption are relatively large compared to the negative effects corruption has on the larger economies of Asian and OECD countries. In another study, Mehrara et al. (2011) examined the effect of corruption on inequality using a panel dataset from several OECD and OPEC countries during the time period of 2000–2007. The results demonstrated that corruptions positively affects income inequality in OECD countries and has no impact on income inequality in OPEC countries, perhaps because of high oil revenues.

In contrast to the majority of findings in the literature, some authors have found a negative relationship between corruption and income inequality.1 Dobson and Ramlogan-Dobson (2010) used panel data for Latin American countries and found that a decrease in corruption leads to an increase in income inequality. This means they found evidence of a trade-off between corruption and income inequality for Latin America, and they argue that this finding is mainly due to the presence of a large informal economy that mostly employs those in the lowest quintile of income distribution. They argue that the way that corruption decreases inequality is by making it possible for entrepreneurs to bypass institutional barriers in starting and operating businesses, in particular in the informal sector.

Dobson and Ramlogan-Dobson (2012) extended upon their previous study by using a sample of developing and developed countries from several regions and found that the marginal effect of corruption becomes negative when the informal sector within a country becomes large.

Finally, Policardo and Carrera (2018) examined a panel of 50 countries from 1995–2015 and robustly found that corruption is not a significant determinant of income inequality, which conflicts with the majority of the empirical literature on this subject.

The empirical findings concerning the relationship between income inequality and corruption are ultimately heterogenous, and therefore additional work is needed to examine the link between corruption and income inequality. In addition, the use of a large country coverage makes it possible to explore more global patterns.

In general, the majority of the literature, especially those papers with largest country samples, found that corruption generates greater income inequality. Therefore the following hypothesis is constructed:

Hypothesis 1: Corruption has a significant positive effect on income inequality within countries.

2.2 Channels of the Effect of Corruption on Income Inequality

A large body of literature shows that corruption can worsen income inequality mainly through the misallocation of resources, which creates strong economic and social disturbances. Corruption influences inequality through several channels (Gupta et al., 2002).

First, corruption can lead to biased tax systems via tax evasion, defective tax administration, and exemptions, which simultaneously results in lower tax revenues. Given that the recipients of tax exemptions and evasion are mainly the wealthy and well-connected, the

1_{Some authors have also suggested a quadratic relationship between corruption and income inequality (e.g., Li}

(7)

tax burden will almost fall entirely on the poor, leading to an increase in income inequality. Furthermore, lower tax revenues will decrease the ability of the government to invest in social programs to improve income inequality (Dincer and Gunalp, 2012).

Second, corruption can impact the expenditures of governments on social programs and the quantity of social programs. According to Gupta et al. (2002), the elites within a country might bribe public officials to force the government to extend benefits to elites instead of using funds for poverty-alleviating programs. This will decrease the effect of social programs on the distribution of income. Even if corruption does not affect the number of social programs, corruption may still affect the composition of social spending in such a way that benefits the wealthy at the expense of the poor, for instance, by focusing expenditure on higher education instead of primary and secondary education, which more greatly favors the poor (Dobson and Andres, 2011). Similarly, healthcare spending could be skewed toward the construction of very advanced hospitals dedicated only to the wealthy instead of investing in preventive healthcare that benefits the poor.

Third, Gupta et al. (2000) point out that corruption can affect inequality via inequality in asset ownership. The level of asset ownership can influence government policy and increase unequal income distribution. If a large number of assets is owned by the well-connected wealthy, this enables them to use their wealth to lobby for favorable policies to increase returns on their assets, while lowering the returns on the assets of the less wealthy. Similarly, Tanzi (1998) notes that only the well-connected people get the government projects which are most profitable. This gives them resources to bribe public officials and further increase their share of assets, thereby causing greater inequality in asset ownership.

Fourth, corruption can influence income distribution through its effect on the distribution of human capital and human capital formation (Gupa et al., 2000). A higher level of corruption is considered to be linked to a lower level of spending on health and education, which further increase income inequalities (Mauro, 1998). Corruption has an effect on the operating costs of the government via decreases in public revenues, which consequently decreases effective public expenditures, including expenditures on education (Gupta et al., 2002). Various studies claim that corruption leads to a reduction in investment in human and physical capital (Tanzi and Davoodi, 1998; Wei, 2000; Gupta et al., 2002).

Lastly, corruption increases uncertainty and risk for low-income groups and those who are not well-connected. Thereby, these groups face higher risk when making investment decisions about land, human capital, or physical capital. Because of this unequal allocation of risk, corruption is likely to maintain income inequality.

2.3 The Role of Women in Government

(8)

relationship between gender inequality in economic and political life and the corruption level (Dollar et al., 2001; Swamy et al., 2001) The first reserach was conducted by Dollar et al. (2001) and showed that women are less corrupt, less individually oriented (selfish), and have higher ethical and moral standards compared to their male colleagues. This study demonstrated that lower levels of corruption were linked to a higher share of female representation in parliaments. If women tend to be less self-serving and base their behavior on higher ethical and moral standards, it means that having women in essential economic and political positions could lead to less corruption and in turn (potentially) less income inequality. The second study by Swamy et al. (2001) found that women tend to be more risk averse and honest compared to their male counterparts. Assuming that women are more risk averse, then women should be less likely to engage in corruption when it is risky compared to men. This study further showed that countries that have a high share of female representation in the labor force and a large share of women representation in government bureaucracies tend to be less corrupt. Later studies also confirmed that women tend to be less corrupt (Frank et al., 2011; Rivas, 2012). Additionally, Paweenawat (2018) found evidence that is consistent with the findings of Swamy et al. (2001) and Dollar et al. (2001), finding a negative relationship between corruption and a higher share of female representation in parliament. Consistent with these previous findings, Hao et al. (2017) showed that a higher share of women representation in the legislative branch is crucial in reducing corruption within a country. The share of female representation in the legislative branch can affect corruption because legislative corruption is a crucial factor in public corruption (Swamy et al., 2001). Furthermore, when the parliament consists of a higher share of women, it can lead to decreased bureaucratic corruption via the development of legal rules and regulations. Consequently, legal rules that lead to less corruption within a country also affect income inequality within a country. Watson and Moreland (2014) found that women’s substantive and descriptive representation are linked to less corruption. They used health expenditures as a proportion of government spending as a proxy for the substantive measure of women’s representation. Health issues are often related to women’s interests (Schwindt-Bayer, 2006; Htun and Power, 2006; Jones, 1997). Elsa Chaney (2014) argued that women considered political office as an extension of their roles as a wife or mother and feel an obligation to concentrate on matters based on those roles, such as healthcare, children and family concerns, and the education and welfare of women. If female legislators place higher priority on these social matters relative to male legislators, it could lead to higher levels of spending on health and education (social programs), which will decrease skill inequalities and by consequence income inequality. Other previous studies also confirmed that greater female participation in government is related with improved health and education outcomes (Tech, 2018). The World Inequality Report (2018) highlights that government expenditures are essential in education, environmental protection, and health to combat existing income inequality as well as to avoid further increases in inequality. Additionally, a higher level of governmental spending on education and health is linked to less corruption (Mauro, 1995).

On the other hand, corruption might lead to fewer women being elected, which leads to less legislation related to healthcare and education. Less legislation in these areas drives inequalities in opportunity, which affects wealth and income inequality.

(9)

inequality. Given the positive relationship between corruption and income inequality, a high level of gender equality in politics can influence the level of corruption and therefore decrease income inequality through public policy. In summary, women’s participation in politics constrains corruption and improves policies outcomes. This paper makes a first attempt at evaluating the interplay between female participation, corruption, and income inequality according to the following hypothesis:

(10)

3 Data and Methodology

3.1 Model Specification

This paper examines the effect of corruption on income inequality for a wide variety of countries from 1998 to 2017, by focusing on changes in income inequality within countries. The Hausman test was used to select the appropriate econometric technique for the panel data, with the choice being between fixed and random effects models. The estimated results of the test demonstrated in Table B2 (Appendix) led to the rejection of the null hypothesis, meaning that the effects model was most suitable. The estimation of the coefficients in the fixed-effect regression model are based on the variation of the (in)dependent variables within the entity.

The sample contains 104 countries, including countries of varying levels of corruption and economic development. The choice of countries and time period was constrained by the availability of data. The complete overview of countries is given in Appendix A. Using panel data allows for the examination of both cross-sectional and time-series dimensions of the dataset, using the variation between and within countries to investigate how different levels of corruption and forms of economic development impact income inequality. To empirically analyze the impact of corruption on income inequality, the following regression model is estimated and considered most suitable based on preliminary tests:

Yit = β0 + β1Corruptit + β2Xit + δt + αi + εit. (1) The dependent variable (Yit) is a measure of income inequality, the top-1% income share. The

corruption measure (Corrupt) is the Control of Corruption Index (CCI), and Xit is a vector of control variables outlined in section 3.1.4. β0 is the intercept, β1 is the slope parameter for corruption, and β3 indicates the slope parameters for the control variables. δt denotes time effects because of time-variant omitted-variable biases and stochastic shocks that could be common to all countries (Stern and Common, 2001; Stern, 2004). αi controls for time-invariant country heterogeneity due to differences among countries with respect to cultural, geographical, and institutional factors. ε is the error term while i and t indicate country and year, respectively. The mediator variable was included in the second model to test for a mediation effect from the mediating variable (share of women in parliament) on the relationship between the independent variable (corruption) and the dependent variable (income inequality). If a mediation effect exists, the effect of the independent variable (corruption) on the dependent variable (income inequality) must be less than in the first model (partial mediation). Perfect mediation exists in the event that the independent variable (corruption) has no effect after adding the mediator variable (Baron and Kenny, 1986). The mediation regression specification for the fixed-effects model is

(11)

where Women denotes the share of women in parliament. The slope parameter of the key independent variable in (1), β1, demonstrates the impact of a higher level of corruption on income inequality. In line with hypothesis 1, β1 should not only show a positive association between corruption and inequality (more corruption is related to higher inequality levels) but should also be significantly different from zero (corruption determines inequality). A positive coefficient sign entails that higher CCI values (i.e., higher corruption) correspond with higher values of the top-1% income share (i.e., more inequality).

According to Hypothesis 2, it is expected that the magnitude of β1 in (1) is larger than the magnitude of β1 in (2), which suggests that the effect of corruption on income inequality is greater in (1) than in (2). If both expectations prove to be correct, there is evidence that corruption affects income inequality through the share of women in parliament.

An important issue in estimating the models is that endogeneity may play a role. Even though a fixed-effects regression model, with time- and country-fixed effects included, partially controls for endogeneity, it does not control for all sources of endogeneity. First, endogeneity may occur from omitted variables. As it is impossible to control for all theoretically relevant factors, I attempted to mitigate this problem by including a number of control variables in the robustness checks. These control variables consist of the total natural resource rents and foreign direct investment. Second, reverse causality is possible. A number of studies examine the possibility that income inequality may affect corruption (Jong-Sung and Khagram, 2005; Dwiputri et al., 2018; Sulemana and Kpienbaareh, 2018). Income inequality may be responsible for fostering corruption, which could be a response to a perceived unequal distribution of income (Policarda and Carrera, 2018). To address the problem of reverse causality, I applied the Arellano and Bond (1991) generalized method of moments (GMM) approach in the robustness checks. Third, measurement errors may play a role. Considering that corruption is a hidden phenomenon, studies are dependent on indirect measures based on perceptions of corruption. Therefore, it is likely that these measurements have some limitations and suffer from biases, meaning the perception of corruption does not correspond to the reality of corruption. The instrumental variable (IV) estimators could correct for measurement errors; however, given the shortcomings with potential instruments for corruption, I decided to reject the IV estimation.

3.1.1.Measuring Income Inequality

(12)

Data on the top-10% income share is used for robustness checks.

The data on the top income share was obtained from the World Inequality Database (WID) and multiplied by 100 to obtain percentages as a unit of measurement. An important advantage of the WID is that it uses several types of datasets, including fiscal data, whereas most other inequality databases solely use household surveys to examine inequality dynamics. Household surveys are mainly based on self-reported information that does not give adequate information on wealth and income levels, particularly concerning the wealthiest people of a population. Therefore, information regarding the top income share is generally underestimated. Additionally, most often the sample included in household surveys is very limited. Taking into account that the group of extremely wealthy individuals is very small, they are most likely not be involved in such surveys. It is worth mentioning that working with fiscal data is not without limitations. The use of fiscal data does not completely solve the issue concerning the underestimation of the top income share because of challenges like tax avoidance and evasion. Additionally, tax data is subject to differences in fiscal policies across countries. Despite these constraints, using fiscal data has been found to substantially improve and increase the precision of measuring income inequality.

3.1.2 Measurement of Corruption

The difficulty of defining corruption also makes it very complicated to measure. On the one hand, what a normally accepted practice is in one country for a particular type of corruption can vary between time periods or countries. Besides the difficulties of defining and measuring corruption, there is also the challenge that most corrupt practices are unrecorded and hidden by nature, which makes the objective measuring of corruption very challenging. Because there are no objective measures of the absolute level of corruption, empirical research relies on indicators that are perception based.

To measure corruption, I used the CCI as a proxy for corruption. This index has been widely used in the literature and, compared to other corruption indices, is more extensive and covers more countries. This index was established by Kaufmann, Kraay, and Mastruzzi (2011) and obtained from the World Bank’s World Governance Indicator dataset. The CCI assesses the overall level of corruption in a country and measures “the extent to which public power is exercised for private gain, including both petty and grand forms of corruption as well as ‘capture’ of the state by elites and private interests” (World Bank, 2019). The CCI index is measured by averaging data from different data sources. These sources of data include surveys from public sector organizations, households, firms, non-governmental information providers, and commercial business information providers.

The index includes 214 countries and is originally scaled from approximately -2.5 to 2.5, with a higher score representing less corruption. To make the coefficients of the CCI more easy and logical to interpret, the index was converted so that a higher corruption coefficient indicates a higher level of corruption.

(13)

differently. In addition, data based on the perceptions of respondents in the survey or subjective assessments from experts can be biased and not fully accurate. This makes systemic bias and measurement error a particular concern. The results should therefore be interpreted with these limitations in mind. Even though there are some complaints against using subjective measurements of corruption (e.g., Olken, 2009), Kaufmann et al. (2010) discusses several arguments in favor of using perception-based indicators. What is important is that non-perception-based indicators of corruption are virtually not existent for inter-country comparisons.

3.1.3 Measurement of Women’s Representation in Government

This thesis uses an econometric model in which only one mediator is posited. A variable acts as a mediator in the sense that it influences the relationship between the dependent and independent variable. The mediator explains in what way or why a particular effect occurs (Baron & Kenny, 1986). As described in Section 2.3., the mediator variable in this study is a proxy for the representation of women in government. Following Swamy et al. (2001) and Dollar et al. (2001), the representation of women in government is measured as the proportion of seats held by women in national parliaments. Data for the percentage of women in parliament is collected by the Inter-Parliamentary Union (IPU) and has been obtained from the World Bank. It shows the percentage of seats held by women in a single or lower chamber.

In summary, there are at least two ways in which women in parliament can influence the aggregate level of corruption. First, legislative corruption is an essential aspect of governmental corruption. A higher share of women in parliament can influence governmental corruption. Since women are generally less sensitive to bribes, more women in parliament should reduce the incidence of corruption in legislation. Second, women in parliament can influence judicial and governmental corruption by adopting legislation to prevent bribery or even to influence the executive and judiciary institutions in some countries or bring corruption to the public’s attention and stimulate the press and media organizations to discuss and highlight the problem (Swamy et al., 2001). However, corruption could result in less women being elected, which might lead to less healthcare and education-related legislation, which are drivers of inequality of opportunity. Inequality of opportunity then impacts inequality of outcome, which essentially is wealth and income inequality.

3.1.4 Measurement of Control Variables

This paper includes a number of control variables to minimize the omitted-variable bias. By including several essential control variables, their impact on other variables in the statistical model is controlled for, consequently preventing both specification bias and an incorrect specification of the regression model (Kolnes, 2016; Wooldridge, 2016).

As in other papers on income inequality in the literature (Gupta et al, 2002; Gyimag-Brempong et al., 2002; Gyimag-Gyimag-Brempong et al., 2006) the regression model in this paper follows the literature and includes the following control variables: gross domestic product (GDP) per capita income, secondary gross school enrollment rates, government expenditure, and the unemployment rate.

(14)

more people will be looking for a job compared to the number of available jobs. This will put a downward pressure on wages, and this is most likely to happen in the low-skill employment, where employees are easy to replace. Hence, unemployment primarily hurts those with less education from low-income groups and will therefore increase income inequality (Policardo et al., 2018).

To control for the different levels of economic development, the natural log of GDP per

capita, adjusted to the Purchasing Power Parity (PPP) is included in the model as a proxy for

economic development. Roine et al. (2009) argues that mainly the richest favor from periods of high economic growth. The income share of the top percentile has increased strongly in periods when GDP per capita growth within a country has been above average. This finding is in line with the top income holders being more sensitive to growth (for instance, through compensation being related to profits).

As already mentioned, human capital supply is of great importance in determining income inequality. Therefore, a proxy for education is also included as a control variable in the analyses. Although there are several measures available with respect to educational attainment, for instance literacy rates or the distribution of education, these measures are mostly not available for a large number of countries and years. A frequently used and widely available measurement of education is the gross enrollment ratio, which measures the level of participation for a certain educational level. A high ratio of enrollment in education suggests a great level of participation at a particular level of education. In a similar vein, it also shows if a certain country is able to accommodate all qualified children in society and for this reason it could be seen as a great measurement for the capacity of the educational system. In line with Ullah and Ahmad (2016) and Dobson (2012), the measurement gross enrollment ratio of secondary education will be used in the analysis of this paper.

Government expenditures is also included to control for the distribution effects of

government spending and the size of the government. Even though, it would be more desirable to include data on social spending, but unfortunate that kind of data is not widely available for many developing countries. Therefore, government expenditures are used as a proxy for social spending. Public expenditures and transfers on social programs can create an important income source for poor people. This is why expenditures on social programs which will benefit the poor, are assumed to have a positive effect on income distribution.

Data on the control variables are obtained from the WDI database (2019) developed by the World Bank. An overview of the definition of all main variables as well as the source and expected sign can be found in Appendix A.

3.2 Descriptive Statistics and Diagnostic Tests

As part of the preliminary analysis of the data, scatterplots were created for each variable to identify any outliers that could/might distort the regression results. Datapoints that were extremely far away from the greater part of the data are excluded from the dataset.

(15)

percent of the population earned 31.7% of the total income, belongs to Malawi in 2002. In contrast, Slovak Republic has the lowest top income share of 0.04, indicating a more equal income distribution in the population. These differences in inequality scores highlight the different levels of economic development of the countries included in the sample. The ten highest top1 income share scores differ between .3171 and .3086 and experienced mainly in the 2010s. The ten lowest top1 income share scores experienced mainly in the beginning of 2010s and differ between 0.0445 and 0.0547. The dataset also includes countries that differ in the representation of women in government. The highest observation of the share of women in parliament is observed in Rwanda, namely 63.8 percent, in 2015. The lowest observation of gender equality, 0, mainly occurs in Middle-east countries such as, Jordan, Kuwait, Yemen, Qatar, Saudi Arabia and the United Arab Emirates.

Different variables also show a large variation within the country. Most of the wide variation within the country with regard to gender equality occurs in Ethiopia , it differs between 2 and 38.8 for the period 1998-2017

Table 1: Descriptive Statistics

Variable Obs Mean Std. Dev. Min Max

Top 1% 1,285 13.886 5.421 4.45 31.71 CCI 1,285 -0.313 1.080 -2.470 1.614 Share Women 1,285 19.582 11.950 0 63.8 Log GDP 1,285 9.3825 1.290 6.434 11.695 Education 1,285 81.960 33.006 6.197 163.934 Government 1,285 17.245 4.964 1.340 40.444 Unemployment 1,285 8.461 6.221 .317 37.25

Furthermore, histograms for all variables were created to check for normality. The histograms of GDP per capita revealed that it was not normally distributed and showed a skewed distribution. Consistent with the econometric literature, the data of the variables have undergone a natural logarithmic transformation and show an improved bell-shape form.

In addition, I also checked for the assumption of multicollinearity, regarding the correlation between the variables, which is present if there is a high correlation (.9 and above) among the independent variables. Table 2 provides the correlation matrix for the main variables. The table reveals a high correlation between GDP and CCI (-0.758), but not completely collinear. As expected, CCI is positively correlated with the measure of income inequality: the correlation coefficient between CCI and Top1 is 0.362. Also in line with the expectation, the measure of women’s representation in government is negatively correlated with income inequality (-0.267). Among the control variables, the table demonstrates the highest correlation between education and log GDP (0.893), suggesting a positive link.2

(16)

Table 2: Correlation Matrix

An assumption behind panel data regression models is that the errors are uncorrelated and homoscedastic. There is a possibility that these assumptions will not be met since the sample consist of both time-series and cross sectional aspects. Therefore, tests for heteroskedasticity and serial correlation is carried out. The Modified Wald test is undertaken to test for the assumption of homoskedasticity. The results, demonstrated in Appendix table B3, imply a rejection of the null hypotheses, meaning that heteroskedasticity is present. Finally, the Wooldrigde test was carried out to check for the presence of serial correlation in the models. The outcome, shown in Appendix table B4 indicates the presence of serial correlation. To correct for the data that shows problems of heteroskedasticity in the error terms, the model will be estimated using robust standard errors clustered by country.

Variables Top1 CCI Share

Women

(17)

4 Empirical results

To visualize the relationship between corruption and income inequality over the period 1998– 2017, Figure 1 shows a scatterplot of the average of the CCI index and the average of the top-1% coefficient. The scatter plot indicates a positive linear bivariate relationship, meaning with more corruption, there is a higher level of inequality. Thus, the figure supports there being a linear relationship between corruption and income inequality.

Figure 1: The positive correlation between corruption and inequality.

To get a sense of the relationship between the mediator variable (share of women in parliament) and the dependent variable (income inequality), Figure 2 (left side) displays a scatterplot of this relationship. The scatterplot demonstrates a clear negative linear relationship, which means that as the share of women in parliament increases, inequality decreases.

The scatterplot also reveals variation in the data. Examples of countries with a relatively high share of women in parliament and a high level of inequality include, among others, South Africa, and Namibia. Countries with a relatively low share of women in parliament and a low level of inequality include countries such as Hungary, Ukraine, Malta, and Japan. On the other hand, Sweden is the country with the highest share of women in parliament and the lowest level of inequality.

(18)

Figure 2: The negative correlation between the average share of women in parliament and average income inequality (left) and average CCI index (right) .

Having produced these descriptive findings, I tested whether these correlations survive the regression analyses. Table 3 shows the results for the entire sample for regressions (1) and (2) with the top-1% income share as the dependent variable. Column 1 reports the results of (1), and column 2 reports the results of (2). The explanatory variables in all regressions are GDP per capita, education, government consumption, and the employment rate.

The most interesting results come from the variables CCI and share of women in parliament in column 2 of Table 3. Holding constant the other control variables, corruption is positively and statistically significantly associated with income inequality in both models. The positive sign indicates that a higher level of corruption exacerbates income inequality, which supports the expectations set out in Section 2.1 and is in line with the earlier findings of Gupta et al. (2002). Corruption is significant at the 5% significance level in both model and testing whether the coefficient for corruption is zero in the equations yields F statistics of 3.80 and 4.92, as shown in columns 1 and 2, respectively.

Column 2 indicates that an increase of 1 index point in the CCI is associated with an average 1.570 percentage points increase in the income share of the top 1%. The coefficient for corruption (CCI) does not substantially change in Model 2, where the variable Women in Parliament is added, and the results remain statistically significant at the same significance level. These results suggest that corruption has a positive effect on income inequality, but not through the share of women in parliament. This implies no mediation effect from the mediating variable (share of women in parliament) on the relationship between the independent variable (corruption) and the dependent variable (income inequality), and therefore the share of women in parliament cannot be considered as a mediator in this sample.

(19)

payments are more responsive to economic growth than regular wages, which are a more stable form of income with relatively few bonusses (Roine et al., 2009).

Education seems to be negatively associated with the top-1% income share and hence lower income inequality. This is consistent with the widely accepted view that gaining access to public services such as education serves to reduce income inequality (Chu, 2000). The coefficients for Unemployment Rate and Government were not found to be significant.

Table 3: Fixed effects panel estimation with cluster robust standard errors Dependent variable is the income share of top 1

(1) (2)

VARIABLES Model 1 Model 2

CCI 1.448* 1.570**

(0.743) (0.708) Share Women Parliament

Log GDP per capita 3.914***

-0.045** (0.022) 4.052*** (1.485) (1.501) Education -0.037*** -0.035*** (0.012) (0.012) Government 0.026 0.031 (0.055) (0.056) Unemployment -0.002 0.001 (0.043) (0.046) Constant -22.639 -23.834 (14.839) (15.073) Observations 1,335 1,285 Number of countries 104 104 Within R-squared 0.076 0.092 Adjusted R-squared 0.061 0.076 Notes: Cluster robust standard errors in parentheses. Country- and year-fixed effects are included in all regressions.

*** p<0.01, ** p<0.05, * p<0.1

In general, globally, the outcomes reported in Table 3 provide no evidence of a mediation effect from the share of women in parliament on the nexus between corruption and inequality, and even though the coefficient on corruption is significant and has the expected sign, the effect of corruption on income inequality appears to be limited. The dataset includes a great number of countries, which implies a lot of heterogeneity in the data, and it might be possible that the results reported in Table 3 differ if examined more deeply. Therefore, I conducted a more in-depth analysis to explore the heterogeneity of different world regions and levels of development.

(20)

20 Europe, and Latin America (LAM).3_{The causes, consequences, and nature of corruption might}

differ geographically, depending on culture, historical context, political organization, and legal systems (Warf et al., 2019). To explore heterogeneity based on levels of development, the dataset is split into two sub-samples: OECD countries versus non-OECD countries. Members of the OECD are known as the “developed countries club” or “rich countries club” since the OECD countries are mainly developed, high-income countries (Clifton and Diaz-Fuentes, 2015).4 Moreover, less-developed countries largely suffer from higher levels of corruption and inequality.

Table 4 presents the results in order to examine whether the main “average” effects from the global analysis (Table 3) are heterogeneous across different regions. In general, the magnitudes of the coefficients do not differ much compared with the main results in Table 3. Overall, there is no support for a mediation effect from the share of women in parliament on the relationship between corruption and income inequality. The coefficient of Women in Parliament show insignificant results and is not statistically different from zero throughout all columns in Table 4. Furthermore, adding the mediator variable into the model does not lead to a significant quantitative change in the coefficient of corruption in any column.

The corruption variable is significant in two of the four world regions. The sign is positive at the 5% significance level in both columns 1 and 2 for Africa, suggesting that a higher level of corruption is associated with a higher level of income inequality. The finding of a positive effect of corruption on income inequality in Africa is in line with the findings of Gyimag-Brempong et al. (2006) and Gyimag-Brempong (2001). The coefficient of corruption is significantly different from zero at α= 0.05 for both columns 1 and 2. Column 2 shows that an increase in the CCI by 1 index point is associated with a 2.608 percentage points increase in the income share held by the richest 1%. Similar to Table 3, corruption in Africa has positive effects on income inequality, but not through the share of women in parliament. The coefficient for Women in Parliament is insignificant in the African sample and including the variable in column 2, does not significantly change either the magnitude or coefficient on corruption. Regarding the control variables, the coefficient for Log GDP in columns 1 and 2 is statistically significant and shows the expected sign. The coefficients of unemployment and government expenditures are significant and do not have the expected signs. The coefficient for government expenditures is positive and statistically significant, indicating the greater the expenditures of the government, the higher income inequality is. This result might be an indication of corrupt governments that mostly extend benefits to the wealthy. Consequently, these government’s increase inequality and their actions might lead to less money being available for social programs. Furthermore, education is never statistically significant.

Turning to the other results in Table 4, Corruption is also statistically significant for the LAM sample in columns 7 and 8. More interestingly, the effect of corruption on the income share of the top 1% changes direction. While corruption seems to be positively associated with the income share of the richest in the overall sample, it has a negative association for LAM

3_{The country groupings are based on the geographic regions defined by the United Nations Statistics Divison}

(UNSD). A complete list of countries included in each region can be found in Appendix C.

4_{Chile is excluded from the OECD sample as the most unequal country in the largely developed OECD and is}

(21)

countries. This indicates that a greater level of corruption is associated with a lower income share for the top 1% in LAM. The finding of an inverse relationship is in line with the findings of Andres and Ramlogan-Dobson (2011), who state that there is a trade-off between corruption and inequality in LAM due to the presence of a great informal sector. They further argue that the informal economy is relatively large in various low-income countries, and those working in the informal economy are mostly among the bottom of the income distribution. Institutional reforms and reductions in corruption are related to more regulations and compliances. This generates increased operational costs for firms in the informal sector. Additionally, they argue that it also impacts the mechanism by which the transactions are carried out. New mechanisms will have to be adopted and firms might be forced to hire more highly qualified people at the expense of those employees who are mainly at the bottom of the income distribution. Furthermore, in countries with high corruption levels, redistribution-related government projects mainly involving manual workers may be encouraged by corrupt public officials for the sole purpose of gaining political power (Andres and Ramlogan-Dobson, 2011).

The corruption variable is different from zero at α = 0.05 in both columns 7 and 8. The numerical interpretation of column 8 indicates that with every unit (index point) increase in the CCI, the income share of the top 1% decreases by an average of 4.535 percentage points. This means that corruption has the greatest impact on inequality in LAM compared to the other regions, although the impact is negative. Furthermore, the coefficients for government expenditures and unemployment are significant and carry the expected signs. Nevertheless, the sample size of the LAM region is relatively small, and therefore I do not think that an inference for the whole LAM region is justified based on these results.

Columns 3–6 show that the coefficient for corruption and the share of women in parliament are insignificant for both Asia and Europe, and therefore we cannot draw any conclusions from this. The result that corruption does not significantly increase income inequality for Asia contradicts the findings from Gyimag-Brempong et al. (2006), who found that corruption increases inequality in Asia. However, their conclusions might differ because they use a different time period (1980–1998), corruption measurement (CPI), control variables, and measurement of income inequality (Gini). To the best of my knowledge, no study has investigated the nexus on corruption and inequality for Europe, which makes me unable to examine whether my results comply with the empirical literature. Ultimately, the insignificant results for Asia and Europe indicate that the sub-samples for Asia and Europe are not driving the results in the global analysis.

(22)

VARIABLES (1) SSA (2) SSA (3) Asia (4) Asia (5) Europe (6) Europe (7) LAM (8) LAM CCI 2.615** (1.138) 2.608** (1.128) -1.014 (1.604) -1.068 (1.698) 0.713 (0.656) 0.738 (0.711) -3.977*** (0.678) -4.535*** (0.926) Women in Parliament -0.010 (0.035) 0.004 (0.062) -0.012 (0.015) -0.298 (0.161) Log GDP Per capita 8.885*** (2.662) 7.830*** (2.450) 2.931 (1.974) 5.588*** (1.434) 2.305 (1.731) 2.095 (1.947) 8.196* (3.233) 16.352 (8.077) Education 0.057 (0.045) 0.054 (0.047) -0.028 (0.035) -0.061** (0.021) -0.026** (0.012) -0.027** (0.012) 0.023 (0.015) -0.032 (0.034) Government 0.224** (0.083) 0.203** (0.078) -0.063 (0.079) -0.036 (0.090) -0.153 (0.096) -0.158 (0.112) -0.136 (0.115) -0.377* (0.133) Unemployment -0.244* (0.125) -0.221* (0.124) 0.244* (0.128) 0.311** (0.106) 0.025 (0.045) 0.016 (0.053) 0.311* (0.103) 0.437** (0.132) Constant -53.650** (21.809) -45.426** (20.614) -10.316 (18.164) -34.206** (13.645) -6.926 (18.622) -4.394 (21.283) -58.823 (28.281) -125.962 (71.476) Observations 372 355 183 168 630 616 42 41 Number of Countries 37 37 16 16 37 37 4 4 Within R-squared 0.380 0.349 0.276 0.330 0.183 0.186 0.786 0.819 Adjusted R-squared 0.341 0.304 0.176 0.223 0.153 0.155 0.583 0.62

Notes: Cluster robust standard errors in parentheses. Country- and year-fixed effects are included in all regressions.

*** p<0.01, ** p<0.05, * p<0.1

(23)

VARIABLES (1) OECD (2) OECD (3) NON-OECD (4) NON-OECD CCI 1.856** (0.902) 1.779* (0.898) 1.235 (0.987) 1.282 (0.963) Women in Parliament -0.019 (0.020) -0.060* (0.031) Log GDP per Capita 1.643

(2.381) 1.519 (2.631) 4.501** (2.048) 4.844** (2.072) Education -0.023** (0.011) -0.022* (0.011) -0.054*** (0.019) -0.053*** (0.019) Government -0.172 (0.135) -0.180 (0.150) 0.069 (0.059) 0.078 (0.058) Unemployment -0.023 (0.054) -0.020 (0.055) -0.004 (0.077) 0.001 (0.085) Constant 2.073 (26.091) 3.579 (29.037) -20.338 (17.765) -22.804 (18.162) Observations 536 531 799 754 Number of Countries 33 33 71 71 Within R-squared 0.177 0.180 0.073 0.095 Adjusted R-squared 0.142 0.143 0.046 0.066

*** p<0.01, ** p<0.05, * p<0.1

A possible explanation for there being no significant relationship between corruption and the top-1% income share in the non-OECD sample might be due to different forms of corruption in “rich” developed and “poor” less-developed countries, which, on average, experience different consequences on different parts of the income distribution. It is possible that corruption in poor countries mostly affects the daily lives of the very poorest (e.g., it makes them pay more for access to markets and services that people in rich countries have as a normal right of citizenship) and as a result, corruption increases inequality on average, by making especially the very poorest even more poor. Thus, it could be that corruption still affects overall inequality in the wider economy in less-developed (non-OECD) countries, but has a predominant influence on other income shares, such as those at the lower end of the income distribution. In developed countries, it is mostly businesses, entrepreneurs, the elite, and powerful who are more engaged in corrupt acts to favor themselves.

(24)

these differences in highly developed and developing countries. These subjects would be interesting avenues for future research.

Nevertheless, what we can conclude from these results is that, even for countries that are highly developed and have strong institutions, corruption still plays a significant role in determining inequality. I cannot draw any conclusion as to whether this impact is lower compared to countries that are less developed and have weaker institutions since corruption is insignificant in the non-OECD sample.

Overall, the results from the fixed-effects model indicate that, globally, corruption is positively and significantly associated with income inequality. This finding is novel since previous studies were mainly based on small country samples. Deeper analysis further revealed that the association between corruption and income inequality exhibits heterogeneity. The relationship between corruption and income inequality is different by world region and level of development. Before I draw any conclusion from these results, I first need to investigate their robustness. The next section focuses on the global, SSA, and OECD samples since other samples showed mainly insignificant results. Besides, the sample size of the LAM region is too small to draw any justified conclusion.

(25)

5 Robustness Tests

5.1 Additional Explanatory Variables

There is a possibility that the results in the previous section depend on omitted-variable bias, and therefore I will add several additional control variables. These variables are added to the base model one by one, to examine whether this affects the results. As two additional explanatory variables I add: foreign direct investment (FDI) and the natural resource rents. Several studies found that corruption has a negative effect on the quality and the number of investments and decreases the rate of return (Mauro, 1995). Moreover, Wei (2000) found that especially corruption decreases the inflow of FDI. Investors who are looking for a honest business environment will prevent investing in highly corrupt countries. Excluding this variable from the model could possibly result in biased results and therefore I will use it as an additional control variable. FDI is measured as the net inflow of FDI as a percentage of GDP.

According to Gupta et al. (2002), the abundance and dependence on natural resources might increase inequality. Also Ross (2003) concluded that economies who are based on exploitation of natural resources, have higher levels of inequality and corruption and perform worse with respect to social development. Therefore, the total natural resources rents will be used as second additional variable. It is measured as the sum of oil rents, natural gas rents, coal rents, mineral rents, and forest rents as a percentage of GDP. The data for both additional variables was obtained from the World Bank (WDI).

Results for the full global sample are demonstrated in Table 6. Columns 4 and 6 show the coefficient estimates of corruption when total natural resource rents and FDI respectively, are included as extra control variables. The coefficient of corruption demonstrated in columns 4 and 6 remains positive and significantly different from zero at α = 0.05 irrespective of the extra control variables I include in the equation. The coefficient of the share of women in parliament remains also stable.

The coefficient estimates of corruption in the SSA and OECD sample (appendix table D1 and D2) are positive and significantly different from zero. Including extra control variables does slightly change the significance in the OECD sample, but the size of the coefficients remains stable.

(26)

VARIABLES (1) Model 1 (2) Model 2 (3) Model 3 + Natural Resources (4) Model 4 + Natural Resources (5) Model 5 + FDI (6) Model 6 + FDI CCI 1.448* 1.570** 1.575** 1.651** 1.641** 1.665** (0.743) (0.708) (0.743) (0.710) (0.750) (0.698) Share Women Parliament -0.045** (0.022) -0.046** (0.021) -0.047** (0.021)

Log GDP per capita 3.914*** (1.485) 4.052*** (1.501) 4.131*** (1.524) 4.148*** (1.505) 4.125*** (1.535) 4.176*** (1.494) Education -0.037*** -0.035*** -0.037*** -0.036*** -0.037*** -0.036*** (0.012) (0.012) (0.013) (0.012) (0.013) (0.012) Government 0.026 0.031 0.020 0.021 0.023 0.022 (0.055) (0.056) (0.059) (0.059) (0.060) (0.058) Unemployment -0.002 0.001 -0.002 -0.000 -0.003 0.002 (0.043) (0.046) (0.043) (0.046) (0.045) (0.044) Natural resources -0.035 (0.024) -0.033 (0.024) -0.035 (0.024) -0.033 (0.023) FDI 0.003 (0.007) 0.002 (0.006) Constant -22.639 -23.834 -20.727 -20.439 -20.689 -20.739 (14.839) (15.073) (14.310) (14.292) (14.488) (14.206) Observations 1,335 1,285 1,333 1,284 1,321 1,280 Number of countries 104 104 104 104 104 104 Within R-squared 0.076 0.092 0.083 0.097 0.084 0.098 Adjusted R-squared 0.061 0.076 0.066 0.080 0.067 0.081

*** p<0.01, ** p<0.05, * p<0.10

5.2 Alternative Measure of Income Inequality

A different way to check the robustness of the findings is by replicating the regressions by using a different measurement of income inequality, namely, the income share of the top decile (10%). An important study conducted by Piketty (2014) reveals that the top decile is very heterogeneous. Capital incomes are comparatively more relevant to the top percentile of the income distribution, while labor incomes are the more important source of income in the top decile.

(27)

corruption rises, inequality rises. This finding is in line with the results in Table 3 and reveals no significant differences. The mediation test does not show any differences compared to the baseline results using the top-1% income share as the dependent variable. Moreover, when the share of women in parliament is controlled for in column 2 of Table 7, the estimated coefficient of corruption does not become considerably smaller or lose significance. The transmission variable (women in parliament) carries the expected sign, although it is not significant and not statistically different from zero at the 90% confidence level. Therefore, we can again conclude that corruption has a positive effect on the top-decile income share, but not through the share of women in parliament in this sample.

The results demonstrated in the appendix Table D3 for the SSA and OECD sample are comparable to the results in the baseline regression specification using the top 1%. The corruption variable shows the expected sign and remains significant.

Table 7: FE estimation with cluster robust standard errors Dependent variable is the income share of top 10%

(1) (2)

VARIABLES FE Model FE Model

CCI 1.807* 2.175**

(1.004) (1.011) Share Women Parliament

Log GDP per capita 4.753**

-0.039 (0.036) 3.999* (1.911) (2.116) Education -0.031 -0.017 (0.020) (0.020) Government 0.045 0.044 (0.052) (0.061) Unemployment -0.022 -0.033 (0.062) (0.069) Constant -22.639 6.980 (14.839) (19.842) Observations 1,335 1,270 Number of countries 104 104 Within R-squared 0.076 0.062 Adjusted R-squared 0.045 0.044 Notes: Cluster robust standard errors in parentheses. Country- and year-fixed effects are included in all regressions.

(28)

5.3 The Arellano-Bond GMM Difference Estimator

The baseline regression results in Table 3 reveal a statistically significant positive association between corruption and inequality. However, it is possible that this association is influenced by reverse causality. Jong-Sung and Khagram (2005) state that higher levels of income inequality could affect corruption through normative and material mechanisms.

The GMM estimator helps tackle the potential biases associated with this problem. The GMM estimator generates consistent estimates when there are endogenous regressors. By deciding whether to use the difference or system GMM specification, I followed ‘the two rules-of-thumb’ by Bond (2001). The dynamic model was first estimated by pooled OLS, where the coefficient of the lagged dependent variable was considered to be an upper bound estimate, while the equivalent fixed-effects estimate was considered to be a lower bound estimate. Since the difference GMM estimator revealed a coefficient value of the lagged dependent variable that was above the fixed-effect estimate, a difference GMM specification was preferred. In choosing between the one-step or two-step variation of the difference GMM, I used the one-step estimates since simulation studies have demonstrated that the two-step estimates offer limited efficiency gains, even if there is heteroskedasticity (Roodman, 2009). The estimator makes use of the lagged values of predetermined and endogenous variables together with current and lagged values of exogenous regressors as instruments in the differenced equation.

(29)

Table 8: Difference GMM estimator Dependent variable is the income share of top 1%

Notes: Cluster robust standard errors in parentheses. Year- and country-fixed effects are included in all regressions.

*** p<0.01, ** p<0.05, * p<0.1

Table D4 in the appendix shows the difference GMM estimates for the OECD sample. Overall, the estimates of the difference GMM estimator confirm the findings from the fixed-effects estimator. The coefficient on corruption declined slightly, but the level of significance remained stable. Corruption is significant at the 5% significance level and has a positive sign, indicating corruption is positively associated with the top-1% income share in OECD countries. Column 2 shows that an increase in the CCI by 1 index point is associated with an average 0.950 percentage points increase in the income share of the top 1%.

Table D5 (appendix) shows the GMM estimates for the SSA sample. In a first attempt regressing the model, I failed to reject the AR(1) test. After adding the second lag of inequality as an endogenous variable, the null hypothesis of no autocorrelation could be rejected at the 5% significance level. The corruption variable is significant at the 10% significance level, although its coefficient size decreased. An increase in the CCI by 1 index point is associated with an average 0.455 percentage points increase in the income share of the top 1%. This is much lower than the effect the CCI has in the OECD sample. Regarding the control variables, the impact of log GDP and government expenditures on inequality decreased but remained significant.

(1) (2) VARIABLES Difference GMM Difference GMM Top1i,t-1 0.867*** 0.797*** (0.138) (0.126) CCI 0.277 0.383 (0.256) (0.300) Women share in Parliament -0.007

(30)

(31)

6 Discussion and Concluding Remarks

Despite the increased interest in the last few years on the share of income of the wealthiest individuals, there is still little known about the effect of corruption on the top income shares. This study has been the first to concentrate on the nexus between corruption and the top-1% income share by employing a fixed-effect approach to examine the effect of corruption on the top-1% income share directly and indirectly through the share of women in parliament. This relationship was analyzed using data from 104 countries over the period of 1998–2017, and therefore this paper offers the first global perspective on the effect of corruption on inequality. There are several findings from the analysis. For the full global sample, the findings show that corruption is positively related with inequality. The finding of a positive relationship between corruption and inequality is consistent with the majority of studies related to this topic (Gupta et al., 2002; Gyimah-Brempong, 2002; Picarda et al.,2019). This result is robust after adding extra explanatory variables and using an alternative measurement of inequality. However, the coefficient on corruption lost significance when applying the difference GMM estimator. Consequently, conclusions for the global perspective based on this result should be interpreted with these limitations in mind.

A more in-depth analysis of the comprehensive dataset reveals heterogeneity in the data. Two forms of heterogeneity are found. First, corruption differently affects income inequality across world regions. The results of the analysis show that corruption has a robust and significant effect on inequality in SSA. This finding is robust to different measures of inequality, additional control variables, and a different estimation procedure. We can therefore conclude that these results confirm the first hypothesis with respect to SSA. Results based on the first difference GMM estimator indicates that an increase in the CCI by 1 index point is associated with an average 0.455 percentage points increase in the income share of the top 1% in SSA. The regional differences found regarding the impact of corruption are possibly due to differences in the character of corruption and in the size of the economies. Second, corruption is revealed to have different effects when splitting the sample according to the level of economic development. Regardless of the additional control variables, measurement of inequality and the estimation procedure, I found that corruption is positively associated with the top-1% income share in OECD countries. Therefore, the first hypothesis, can also be affirmed with respect to the OECD sample. This means that even for highly developed countries with strong institutions (to control for corruption), corruption still plays a significant role in determining inequality.

(32)

the wealthy and well-connected population groups in society, who can shape the system, most likely benefit from corruption by privilege themselves further (Acemoglu, 2002). I found this to be true for the SSA and OECD samples. However, I found insignificant relationships between corruption and the top income share for the other samples. This result again indicates the multifaceted role corruption might have on different shares of income. This is the first paper that explores the effect of corruption on the top income share, and future research is needed to examine the effect of corruption on other income shares. This in order to better understand its distortionary effects on the distribution of income and establish the most effective policies for reducing inequality.

Lastly, this study found no evidence with respect to the proposed channel of transmission. There was no evidence of mediation by the proposed mediator. This is not consistent with the theorized hypothesis, which states that the share of women in parliament mediates the relationship between corruption and inequality. A possible explanation for this result might be that the proportion of parliamentary seats held by women in single or lower chambers is not accurately capturing the representation of women in government. The results might differ by using another proxy for the representation of women in government, whereby e.g. different dimensions of women within government are taken into account (women’s presence in cabinets, sub ministerial officials, ministerial officials, parliamentarians etc.). Another explanation could be that the share of women in parliament can not be considered as a channel and that it does not mediate the corruption-inequality nexus. However, before drawing upon those conclusion, I would first like to encourage more in-depth future research on this channel. The World Bank classifies corruption as the greatest barrier to political, economic and social development, and therefore I believe it is highly important to investigate all potential channels corruption might have on development. (World Bank, 2003).

THE RELATIONSHIP BETWEEN INCOME INEQUALITY AND CORRUPTION: THE ROLE OF FEMALE REPRESENTATION IN GOVERNMENT