Immigration and income inequality in the EU from 2004-2014

(1)

1

Immigration and income inequality in

the EU from 2004-2014

Bachelor Thesis By Ieva Augustinaite 10554645 Supervised by N.Ciurila January 2017 University of Amsterdam BSc Economics & Finance

(2)

2

Statement of originality

This document is written by Ieva Augustinaite 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.

(3)

3 Abstract

The paper analyses the effect of immigration on income inequality for a sample of 23 EU countries covering a period between 2004 and 2014. Theoretical section analyzes the effect immigration on labour market outcomes and discusses recent trends in Europe. The empirical part aims to estimate the effect of immigration by employing fixed effects model for cross-country time series, all-the-while controlling for factors shaping income inequality. The study finds that newly admitted high-skilled immigrants have a positive effect on income inequality. However, the overall results suggest that no relationship between overall foreign-born population, low-skilled immigrants, high-skilled immigrants and income inequality exists. Inconclusive results can be a consequence of low variation in the variables of interest over the time frame investigated leading to poor empirical performance.

(4)

4

2.7 Empirical evidence ... 13 3. Econometric analysis ... 16 3.1 The data ... 16 3.2 The model ... 16 3.3 Dependent Variables... 17 3.4 Independent variables ... 17 3.5 Control variables ... 18 3.6 Hypothesis ... 19 3.7 Assumptions ... 19 3.8 Panel diagnostics ... 19 4. Results ... 22 4.1 Descriptive statistics ... 22

4.2 Panel model results ... 22

5. Discussion ... 25

6. Conclusion ... 26

References ... 27

(5)

5

1. Introduction

During the last two decades earnings inequality has increased in many industrialized countries (Smeeding, 2006, OECD, 2011). According to Piketty (3013), inequality in many European countries has reached its historic height and has been a subject of debate among the policy makers. Simultaneously, the share of immigrants in these countries has been continuously growing (Piketty, 2013). Attitude surveys provide consistent evidence about the rising concern and negativity towards foreign-born population (Page, 2009; Hainmueller et al., 2014). IOM (2015) claims that “European residents appear to be, on average, the most negative globally towards immigration, with the majority believing immigration levels should be decreased”.

Among the determinants of income inequality, both political and economic, immigration has been one of the most controversial (Xu, 2015). According to Federal Reserve Bank of New York (Figure 1), immigration is responsible for 5 to 10 % increase in earnings inequality in the Unites States.

Figure 1. Average % contribution to income inequality in the U.S

Source: Federal Reserve Bank of New York

Scholars argue that the link exists via depressing the wages for low-skilled workers, adverse employment effect and the tendency of immigrants to display larger income dispersion than the natives, thus directly increasing in inequality in the host country (Borjas, 1995; Card, 2009). So far, existing literature has focused mainly on the labour market effects, primarily investigating

Technological Other International Decline in Decline Rising change trade real minimum in immigration wage unionization

(6)

6

immigration the United States. Most of the studies found that the effect of immigration was either moderate or absent (Altonji and Card, 1991, 2000; Borjas, 2003; Card, 1990). However, literature directly linking immigration and income inequality is rather limited. This is especially a case for Europe as increasing immigration is a relatively new concern for most countries (Dustmann and Frattini, 2011). Although public often holds immigrants responsible for increasing income gap and possibly harmful labour market externalities for some native workers, so far the evidence is rather inconclusive (Dustmann et al., 2005). It seems that negativity towards immigrants has been primarily caused by social media and lacks empirical credibility (Boomgaarden et al., 2009).

Rising concern over inequality and immigration in European countries provides a motivation to investigate a potential relationship empirically. Thus, the aim of this paper is to estimate the impact of immigration on income inequality in the EU.

The paper is constructed as following. Section 2 provides a theoretical framework and existing literature regarding income inequality and immigration. In section 3, econometric analysis is provided. The data, the model employed and methodology are explained. Section 4 presents the results. Discussion of the results and limitations are provided in section 5. Finally, section 6 concludes the paper and provides suggestions for the future research.

2. Literature review and theoretical framework

The following section presents the most relevant economic theory and existing evidence linking immigration and income inequality. Furthermore, in order to obtain better insights about the relationship, a brief overview of trends for each phenomenon are discussed separately.

2.1 Brief immigration history in Europe

Large-scale immigration in many European countries is a more recent phenomenon compared to the United States, Canada or Australia (Dustmann and Frattini, 2011). Until the early 1970s, many current net migration European countries, such as the United Kingdom, Spain, Norway, Ireland or Italy were the primary source of emigrants (Hall, 2000). However, due to increasing standard of living and economic growth, these trends reversed. European countries started attracting immigrants from less developed regions, in particular, the ones linked through colonial history. Some scholars argue the phenomenon could have arisen due to inequalities between European countries and their former colonies.

(7)

7

Historic events had high significance shaping European immigration. In 1957, the establishment of European Economic Community (EEC) granted its citizens free movement of capital, goods and people within the Member States. This lead to the first major immigration wave to the Western and Northern Europe. Due to economic expansion and labor shortage, immigrants were largely encouraged and welcomed by the receiving countries. The breakdown of Soviet Union together with the Balkan Wars in 1990s led to many civil intra-Europe conflicts causing large refugee and asylum movements that started triggering Southern European countries (Hall, 2000). According to Raess and Burgoon (2015), such an unfavourable immigrant flow could have imposed various negative labour market externalities on the receiving countries’ populations. Finally, an enlargement of the European Union in 2004 and integration of former Soviet Union countries significantly increased the intra-EU immigration leading to the third major wave. The latter was primarily driven by labour market reasons. Thus, it becomes interesting and relevant to investigate the period following the major EU enlargement in 2004.

2.2 Composition of immigrants in the European Union

The following section summarizes the main “Immigration: The European experience” findings by Dustmann and Frattini (2012).

In 2009, immigrants accounted for more than 11% of the total European population, which is only 1% less than in the U.S. According to the authors, composition of European immigrants is very heterogeneous. Foreign-born population across countries differs significantly in terms of origin or educational background (detailed summaries and tables are provided in Appendix 1). As shown, most of the immigrants come from inside the EU15, accounting for 20.61% of total foreign-born population. For these immigrants, Belgium and Ireland are the most popular destinations. Immigrants to Spain, Portugal and France originate mostly from their former colonies, in particular, Latin America and North African regions, arguably, due to the cultural ties.

On average, immigrants in EU countries tend to be less educated than the natives. The share of immigrants (natives) with tertiary education accounts for 24% (26%). Exceptions are Northern American and Oceanian immigrants with the share of tertiary education reaching approximately 50 %. Immigrants originating from North-African and Non- EU tend to be least educated with approximately 50 % of immigrants with lower than the secondary education. Foreign-born population coming from the EU15 closely resembles educational composition of the natives.

(8)

8

As shown in Figure 2, distribution of earnings between natives and immigrants is very uneven. A trend of decreasing share of immigrants as earnings increase is visible, both for the EU and non-EU immigrants.

Figure 2. Earnings distribution for natives and immigrants1

Source: Dustmann and Frattini (2012)

On average, immigrants coming from non-EU have lower earnings and approximately 15 % belong to the lowest earnings decile. People coming from Latin America are at the biggest disadvantage with approximately 1 out of 5 people clustering at the bottom of the distribution. According to the source, income distribution for EU15 immigrants resembles the U-shape, implying a high density of immigrants at the lowest or the highest percentiles of the distribution. Furthermore, it is concluded that trends in immigrant earnings follows similar pattern to that of educational background.

In terms of employment prospects, immigrants are at the disadvantage compared to the natives and, on average, hold lower ranked occupations in all EU countries. This is especially true for non-EU immigrants. The latter group is associated with lowest activity rates and highest unemployment rates peaking at 2013 compared to other immigrant groups (Eurostat, 2014).

(9)

9

In conclusion, deriving trends for overall EU immigrants is rather complicated due to great diversity of foreign born population across countries. Nevertheless, the findings will be relevant for the empirical part of the paper and will help to form the expectations regarding the consequences of immigrants on income inequality.

2.3 Income inequality. Definition and measurement

According to Equality Trust, income inequality is the extent to which income is distributed unevenly among a group of people. It reflects the gap between the poorest and richest individuals/ households, nationally/globally. Income inequality is recognized as one of economic and social inequality forms. Thus, policy makers cannot combat issues such as poverty without looking into existing inequalities in the society. Consequently, information regarding inequality became essential for measuring poverty, crime rates or social exclusion (Lederman et al. ,2002; Lichbach, 1989). According to the latter, income inequality is not only linked to social conflicts, but is positively correlated to political problems and unfairness.

Given the importance of inequality for policy makes and society, determining relevant causes is essential. The most common way of presenting uneven income distribution is the Gini Index (Deininger and Squire, 1996). The measure is based on the Lorenz Curve which plots the % of total income (y axis) accumulated by the bottom % of the population (x axis). As shown in Figure 3, the 45 degree line represents perfect equality. The Gini can be interpreted as the ratio of areas _A+BA .

Figure 3. The Lorenz Curve2

2_Note:_{Gastwirth, J. L. (1972). The estimation of the Lorenz curve and Gini index. The review of economics and statistics,} 306-316.

(10)

10

The Gini measure is attractive due to its interpretational ease (Deininger and Squire, 1996). However, due to its computational simplicity relative to alternative measures, such as Theil Index, the coefficient might lead to inaccurate and deceptive results. It fails to differentiate between types of inequalities and is rather sensitive to changes in the middle part of the income distribution (Mellor, 1989). Furthermore, the index only plots the number of income properties, but the diagram is not based on any distribution process model. Consequently, same coefficient can occur for two very different income distributions. Despite its drawbacks, the Gini index remains the most popular estimation method in existing income inequality literature.

2.4 Recent trends in the EU income inequality

According to the report “Wage and Income Inequality in the European Union”, evolution of income inequality goes in line with wage inequality. Although wage is just a part of household income, it contributes significantly shaping inequality in European countries. Appendix 2. provides detailed summary of income inequality changes measured by Gini Index between 2004 and 2014. The results suggest that about two-thirds of EU Member States have experienced an increase in income inequality over the last decade. By the end of the period, Gini coefficient has risen by approximately 3%-3.5% in Luxembourg, Germany and Cyprus. France, Denmark and Spain experienced a 2% increase. These countries, with the exception of Cyprus, have been long perceived as net migration countries. On the contrary, the UK, Poland and Portugal experienced 3%-5% decline. In other countries, Gini Index remained either stable or there was too little evidence to draw conclusions. Furthermore, the overall average income has been growing in most of European countries ,however, the poorest 10% of the population have failed to capture any considerable share of the rising income OECD(2011). The richest 10% have experienced a much higher income growth than the residual population for the last 25 years.

2.5 Factors causing income inequality in the EU

Dreger et al. (2015) concludes that income inequality causes in the EU resemble those of other industrialized countries, especially the U.S. The author suggests that technological change, globalization, openness to trade, crowding out of routinized work and financial crisis have been the core factors driving recent income inequality in EU countries. The impact of the EU enlargement has been the most recent subject of discussion. The membership granted its citizens free movement and provided various labour market opportunities for workers in more developed European countries. After the expansion of the Union in mid-90s, overall Gini coefficient increased from 0.29 to 0.3 for

(11)

11

the pooled EU sample. In 2004, after Hungary, Lithuania, Poland, Slovakia, Slovenia and Czech Republic joined the Union, Gini Index further rose to 0.33.The latter EU expansion was followed by the third major immigration wave in Europe. Although the impact of EU expansion on earnings distribution has been recognized, the role of increased immigration on income inequality remains unclear.

2.6 Immigration and income inequality. Theory and relationship

Generally, the impact of immigration on income distribution can be split into direct and indirect effects. Lemos (2014) and Card (2009) argue that immigrants themselves tend to cluster in the lowest/highest percentiles of income distribution, thus directly contributing to increasing the earnings gap. More uncertainty arises analysing the indirect effect that occurs due to possible adverse wage/employment effect on the existing workers. Theoretical predictions about the consequences of immigration crucially depend on assumptions and framework applied. The following section presents the dominating approach modelling labour market effects of immigration. Theoretical framework is based on work by Borjas (1995), Dustmann et al., (2005) and Card (2001). Classical labour market theory considers one output good, constant returns to scale technology (CRS) economy where capital is fixed. A distinction is made between skilled and unskilled labour that are either immigrants or existing workers. Labour is assumed to be perfectly inelastic, meaning that workers are willing to accept whatever wage is offered. The model specifies that immigrants and existing workers are substitutable within a skill group and complementary otherwise, implying low-skilled workers being complements to high-low-skilled workers. In a case of a substitution, short-run economic model predicts an adverse wage effect due to an increased number immigrants, and thus excess labour supply in the particular skill sector. The final assumption states that, in order to experience any distributional effect, the skill composition of immigrants and existing workers must differ. In case of identical skill composition, overall output increases and no wage/employment effect occurs as relative skill composition is not affected.

For illustration, an extreme condition where all immigrants are low-skilled is presented. In such a case, an economy experiences excess supply in a low-skill sector. To restore the labour market equilibrium, firms thus are able to offer lower wages to all workers . An increase in low-skilled labour supply creates a relative scarcity of skilled workers and drives up their earnings.

(12)

12

Figure 4. The impact of low-skilled immigration inflow on low-skilled workers

Figure 5. The impact of low-skilled immigration inflow on high-skilled workers

Source: Dustmann et al., 2005

As demonstrated in Figure 4., in the pre-immigration period existing workers (N) are employed at 𝑤0 with an equilibrium at point A. Immigration shock (M) shifts labour supply (perfectly elastic) to the right and all unskilled workers (N+M) are now employed at a lower wage level 𝑤1. Due to skill complementarity assumption, Figure 5. shows that the relative demand for skilled workers increases, shifting the demand curve to the right and driving up high-skilled earnings. Moreover, the model

(13)

13

suggests that an increase in the number of unskilled workers creates an immigration surplus, denoted by the area (ABC) in Figure 4. Similarly, due to the same assumption of skill group complementarity, the surplus is absorbed by the skilled workers. Therefore, immigrant surplus process (see Borjas, 1995 for extensive discussion) creates earnings redistribution both at the top and the bottom of the distribution.

Until now, it was assumed that wages solely restore the labour market equilibrium. If an assumption of labour inelasticity is relaxed, low-skilled immigration shock can cause voluntary unemployment among existing low-skilled workers as their wages fall and people are not willing to accept the lower wages offered in the market. Thurow (1987) argues that unemployment rate is positively correlated with income inequality as a higher fraction of unemployed people are associated with lower income, thus further increasing dispersion in earnings distribution.

Dustmann (2005) argues that such labour market modelling is rather unrealistic and not applicable for nowadays economies. It relies on simplified assumptions that fail to capture all possible adjustment mechanism aspects to the immigrant inflow. For example, it assumes limited flexibility of output mix and openness to trade that are more applicable for current economies. Nevertheless, the model remains a core for much empirical work.

2.7 Empirical evidence

Empirical application has proven to be more complex than the underlying economic theory. So far, no consensus regarding the appropriate estimation framework has been reached. Most of existing research has focused on immigration consequences on the natives’ wages and natives wage distribution. Overall evidence suggests that the redistributional effect, if any, comes primarily from the downwards pressure on the wages for the low-skilled native workers. Although the wage effect on the natives is shown to be rather small or absent, the effects are more significant for the overall earnings distribution. Most relevant studies are presented in the following section. Furthermore, a few studies analyzing the relationship between immigration and income inequality directly are discussed.

According to Friedberg (1995), one of the most important decisions in modelling distributional effects of immigration is the degree of substitutability among immigrants and existing workers. There have been several attempts trying to estimate the elasticities of substitution. Card (2009) conducted a cross-city time series study in order to measure the labour market effects of

(14)

14

immigration for the US cities. He concluded that: (1) immigrants and natives without high-school education are perfect substitutes to those with a high-school degree, (2) high-school graduates and college graduates are imperfect substitutes. Furthermore, Card found that immigration increased wage inequality for the U.S native workers only moderately. However, the overall earnings effect was more significant and it was shown that immigrants were responsible for approximately 5 % increase in inequality during period 1980 and 2000. The results are obtained using both OLS and IV regression methodologies and the change in the variance of log wages as a measure of inequality.

Altonji and Card (1991) applied the local labour market approach in order to estimate the wage/employment effect on the opportunities of low-skilled natives in the United States. The study compared the differences in labour force market participation across 120 U.S. cities during 1970 and 1980. Average wage levels for specific participant groups based on sex/race were regressed against the fraction of immigrants within a group, controlling for city-specific factors, such as education and age. The main results of their research showed that 1% increase in less-skilled immigration rate lead to a 1.2 % decrease in the wages for low-skilled workers. However, the study failed to fully control for bias arising due to internal migration of the U.S natives themselves, thus the results must be treated cautiously.

Hartog and Zorlu (2005) extended Altonji and Card’s research by including the third skill group and distinguishing among low-, medium- and high-skilled labour. Such methodology was expected to provide more econometric accuracy. The study attempted to measure the impact of immigration from a reduced form wage equation for several European countries: the United Kingdom, Norway and the Netherlands during the 1990’s. Hartog and Zorlu concluded that a 10 % increase in the low-skilled non-EU immigrants in the Netherlands depressed the less-low-skilled natives’ wages by 0.42% and pushed up the wages of skilled workers by 0.21%. Although the results were significant, the estimates were rather negligible. No such effect was found for medium-skilled workers. Estimation results regarding Norwegian and British labour markets were more complex. Immigrants to Norway were shown to have a very moderate positive effect on low- and medium-skill workers. Black and Pakistani immigrants to the UK were shown to have a negative wage effect on the natives’ earnings for all three skill categories. Overall, the study concluded that the impact of immigrants was nearly 0.

Dustmann et al. (2013) took a different approach and estimated the wage-effect along the earnings distribution of the natives without pre-allocating immigrants to a particular skill category for the UK labour market. Author argues that even though the overall adverse effect to natives wages is close

(15)

15

to 0, the impact of immigration should be felt differently along the distribution. Differences over time for percentiles of log wages were regressed on the fraction of immigrants in a total population, time dummies and the ratio of high-to-low skilled natives. The research confirmed the author’s concern: OLS and IV regression estimates suggest that an increase in immigration rate equal to 1% of total population is associated with 0.6% decrease in the 5th_{percentile wages, 0.5% decrease in the}

10th_{percentile wages, 0.6% increase in median wages and 0.4% increase in the 10}th_percentile

wages. Both estimation methods suggest a positive effect of immigration around median wages and a sizeable negative effect at the lowest wage distribution percentiles for the natives.

Although most of the studies examined the impact of immigration on income distribution looking at the variances of hourly wages, several studies focused on inequality measured by Gini Index. Hibbs and Hong (2015) concluded that immigration was responsible for approximately 24 % increase in income inequality for U.S metropolitan areas- the largest effect found in literature. The effect was captured by regressing the changes in Gini coefficient on the share of immigrants who arrived between 1990 and 2000, all while controlling for the number of college graduates and workers employed in manufacturing sector. The authors believed that immigrants were more likely to settle in cities that offer higher wages in the occupational sectors they were most likely to be employed in. Therefore, in order to avoid the possibility of endogenous and non-random destination choice of immigrants, instrumental regression methodology was used. The overall regression results suggested that 1% increase in immigration share relative to the total population in metropolitan areas was associated with an increase in Gini coefficient by 0.66.

A study conducted by Xu et al. (2015) found a strong positive effect of foreign-born population on state-level income inequality in United States for period 1996-2008. The study regressed the income inequality, measured by Gini-Index, on the immigration rate for 3 foreign-born population subsets, distinguishing among newly admitted legal permanent residents, low-skilled and high-skilled immigrants. Authors employed static OLS for pooled panel data and dynamic Error Correlation Model (DCM) to determine the long-run effect of immigration. Both models used panel-correlated standard errors in order to avoid cross-sectional dependence. For the robustness of results, 90/10, 50/10 and 90/50 decile ratios were included as alternative measures for income inequality. All the regressions included multiple economic, demographic and political factors in order to control for other income inequality causes and produce more accurate results. A key finding from the static OLS regression implies that an increase in overall foreign-born population has a significant effect on the overall income inequality. Furthermore, the estimates from The ECM model suggest it is the

(16)

low-16

skilled workers that are responsible for an increase in the long-term state-level income inequality. In fact, moderate negative effect of high-skilled immigrants on income inequality was found.

Given the composition of EU immigrants discussed in previous sections and underlying economic theory and existing empirical evidence, either no or moderate positive effect of immigrants on income inequality is expected. Such expectations are based on three main conclusions of the section focusing on the composition of the EU immigrants: (1) immigrants differ in terms of their educational attainment from the natives and are slightly less educated than the natives, (2) on average, density of immigrants is higher at the tails of earnings distribution and (3) foreign-born population is at the disadvantage regarding labour market opportunities.

3. Econometric analysis

𝑖𝑡

where the effect of newly permitted immigrants ∆𝐼𝑀𝑖𝑡 is approximated by the change in immigrant

share with respect to previous year for all three immigration measures. For the robustness of the results, the same two regressions will be performed with an alternative income inequality measure, namely, 90/10 decile ratio. The model will be estimated using fixed effects methodology. The justification of such choice will be presented in following sections.

3.3 Dependent Variables

Income inequality (GINI). Following the discussion provided in the theoretical framework, income

inequality is approximated by the Gini coefficient. The Gini Index is based on equivalized disposable income (EDI) that includes earnings (salary/wages) and unearned income (benefits/government transfers). EDI is defined as "the total income of a household, after tax and other deductions, that is available for spending or saving, divided by the number of household members converted into equalised adults". Due to the differences in economic policy and social security programs across countries immigrants are eligible for, such measure is appropriate in order to capture the overall distributional effect (Xu, 2015).

90/10 income decile ratio (S9010). In order to show the robustness of results, an alternative income inequality measure is used. Income ratios are very common in existing literature and are especially convenient in capturing disparities arising between particular income categories (Xu et al., 2015).

3.4 Independent variables

Foreign-born population (IM). The term refers to people, born abroad, both in other Member States

of EU or non-EU countries, who are recorded to be residents in reporting country as of the 1st of January in the respective year. Immigration rate is calculated dividing the number of immigrants by the total population. The measure is further disaggregated into: (1) low-skilled immigrants; (2) high-skilled immigrants.

(18)

18

Low-skilled (LSM) and high-skilled immigrants (HSM). As shown in existing literature, depending on

the type or skill of the immigrant, different consequences on host economies are expected. Thus, it is appropriate to disaggregate the total foreign-born population into two skill categories. The skill level is approximated by educational attainment. Low-skilled immigration is defined as an immigrant share(age 15-64) in the overall foreign born population, that has obtained at most lower secondary education. High-skilled immigrants are those having obtained tertiary education. By disaggregating the overall foreign-born population into two subcategories, it is possible to investigate how these groups are linked to income inequality individually.

3.5 Control variables

Real GDP per capita growth (GDPGrowth). Contribution of economic growth and development to

income inequality had been long an on-going debate (Kuznets, 1953, 1955). If the high share of economic growth is received by those with the highest income, then economic growth is expected to be positively correlated with income inequality. The variable is expressed as a percentage change on the previous year and valued in terms of previous year prices. The calculation method used by Eurostat is called chain-linked series. Therefore, price movements do not influence the growth rate.

Education (EDU). Nielsen et al. (1997) and Jacobs et al. (1998) suggest that increasing level of

education is associated with greater income inequality. Over the recent years highly educated people experienced an increase in their income whereas the drop-outs or those without a university degree experienced otherwise (Gottschalk, 1997). It is argued that the increase in income inequality could have arisen due to the demand for high-skilled workers growing faster than the supply of low-skilled workers. A share of population, aged 25-64, having attained at least upper secondary education is used to measure the impact of education on income inequality. It is expected the variable has positive effect on inequality.

Openness to trade (TRADE). According to Richardson (1995), increasing income inequality that many

industrialized countries experienced since mid 1980s could have arisen due to increase in the

international trade. The usual proxy for trade openness is defined as a sum of exports and imports of goods and services expressed as a share of GDP. The data on the variable is extracted from World Bank national accounts and OECD National Accounts data files.

(19)

19

3.6 Hypothesis

The following hypothesis will be tested:

𝐻0: 𝛽𝐼𝑀 = 0; 𝐻1: 𝛽𝐼𝑀≠ 0

The null hypothesis will test whether the coefficient of immigration variables is 0, implying that migration has no significant impact on income inequality in European countries. The alternative hypothesis indicates a coefficient that differs from zero, thus, immigration variables having a positive or negative effect on income inequality depending on the sign of the coefficient.

3.7 Assumptions

In order to the perform the linear regressions and test the hypothesis for panel time series data, several assumptions need to be made (Stock and Watson, 2011):

1. Linearity between the dependent and explanatory variables 2. Zero conditional mean 𝐸(𝜀𝑖|𝑋𝑖) = 0

3. Statistical independence of the errors (no autocorrelation/cross-sectional dependence) 4. Homoscedasticity (variance of the error term εit is constant)

5. Normality of the error distribution 6. No perfect multicollineary

3.8 Panel diagnostics

Before presenting the regression results, it is important to check the validity of the assumptions stated above. Linearity assumption can be evaluated by plotting the residuals versus the fitted value. As shown in Figure 6, the residuals are systematically distributed around the horizontal line exhibiting no clear pattern. Although higher density is noticeable on the left side, overall it can be concluded that the data exhibits linear relationship.

(20)

20

Figure 6. Residuals versus fitted values

In order to check for the zero conditional mean, literature suggests plotting the residuals against the independent variables. The condition is satisfied if the residuals are systematically distributed around zero. The graphs are shown in Appendix 4. and it can be concluded that zero conditional mean condition is satisfied, since residuals are mainly clustered around 0 (y axis). Furthermore, a set of diagnostic tests for the baseline model are performed in order to justify the choice of methodology. Results are summarized in Table 4. Firstly, the results of Breusch and Pagan Lagrangian multiplier test imply that the variances across countries are significantly differ from zero. That is, simple OLS is not appropriate and panel effect needs to be taken into account. Secondly, Hausman test shows random effects is preferred over the fixed effects. However, the estimates remain consistent under both methods. Random effects estimation imposes multiple computational limitations, important for this research. For this particular reason, time-fixed effect model will be used. Modified Wald test implies that the hypothesis of homoscedasticity is rejected. In addition, Wooldridge test suggests the presence of first-order autocorrelation. Baltagi (2008) argues that the issue of cross-sectional dependence usually occurs investigating long time series (>20-30 years),however, results of Pesaran’s test imply the presence of cross-sectional dependence in this dataset. To mitigate these violation of several assumptions, Daniel Hoechle (2007) suggests using Driscoll and Kraay standard errors, robust to cross-sectional dependence, autocorrelation and heteroscedasticity.

Normality of residuals is rejected as shown by Jarque-Bera test (graphical presentation of the residuals is provided in Appendix 5). The violation can lead to difficulties in obtaining significant estimates in the following section (Stock & Watson, 2011). Several explanations of non-normal distribution include: (1) misspecification of the model and (2) omitted explanatory variables. Finally,

-. 1 -. 0 5 0 .0 5 .1 R e si d u a ls .25 .3 .35 Fitted values

(21)

21

the correlations table presented in Appendix 6 and shows that no multicollinearity condition is satisfied.

Table 4. Summary of the diagnostic tests

Diagnostic tests

1) Breusch and Pagan Lagrangian multiplier test

for random effects H0: random effects are zero; var(αi) = 0

Result: Prob > chibar2_=0.00003

2) Hausman test H0: difference in coefficients is not _systematic

Result: Prob>chi2_{= 0.1146}

3) Test for time effects: H0: no time-fixed effects

Result: Prob > F = 0.0016

3) Modified Wald test for groupwise

heteroscedasticity H0:

variance of the error term εitis

constant

Result: Prob>chi2_{= 0.0000}

5) Wooldridge test for autocorrelation in panel

data H0: no first-order autocorrelation

Result: Prob > F = 0.0074

6) Pesaran's test of cross sectional independence H0: no cross-sectional dependence

Result: Pr = 0.0000

7) Jarque-Bera residual normality test H0: normality

Result: Chi2_{= 0.0115}

Overall, taking into account the characteristics of the panel dataset, there is sufficient evidence that fixed effects model with Driscoll-Kraay standard errors is a plausible estimation method. The method is able to mitigate several issues discussed above, however, correcting for the non-normality of the errors is beyond the scope of the research and will be disregarded.

3_Note:_{Probabilities represent the p-values for respective diagnostic tests, values lower than 0.05 imply rejecting the null}

(22)

22

4. Results

4.1 Descriptive statistics

Appendix 6 provides the summary of statistics and the table of correlations. As shown, the mean Gini coefficient is shown to be approximately 0.289, immigration rate is around 0.968% and the share of low-skilled immigrants exceeds that of high-skilled immigrants. After plotting the variables graphically, it can be seen that no clear pattern across countries occurs. Thus further econometric analysis is required. It can be noticed that the changes in the variables over the time series investigated are very small. This is especially true for Gini coefficient. Low variation within and across countries for the independent variable can induce issues in econometric estimation as it becomes difficult to capture any possible effect of explanatory variables(Baltagi, 2008). In addition, as shown in the graphs, several countries experienced a structural break in immigration starting in 2007 as the financial crisis began. Ideally, estimations for period before and after the crisis should be provided separately, however, that would lead with insufficient time frame. Therefore, only the total period of 10 years is considered. From the correlations it can be seen that real GDP per capita exhibits very lower correlation with the dependent variable than with other explanatory variables. For this particular reason, it become important to investigate the relationship using two models: including and excluding real GDP per capita from the regressions.

4.2 Panel model results

In total, 8 regressions are performed. Model 1(Table 5) presents results including real GDP per capita growth as a control variable. The effect is first measured using Gini Index as a dependent variable. The effect of immigrants and newly permitted immigrants (∆𝐼𝑀) is estimated in two separate regressions. Then, 90/10 income decile ratio is used as the dependent variable and the regressions are constructed following the same procedure. Model 2 (Table 6) provides the results excluding real GDP per capita growth. The results are as following:

(23)

23

Table 5. Model 1, fixed effects estimates

Gini Index 90/10 decile ratio

IM -.223084 (-0.83) - -2.18366 (-1.72) - LSM -.00865 (-0.62) - .02624 (0.27) - HSM . 00853 (0.38) - .00774 (0.07) - ∆𝐈𝐌 - -.00133 (-1.30) - .00408 (0.22) ∆𝐋𝐒𝐌 - .00324 (1.81) - .09080 (1.35) ∆𝐇𝐒𝐌 - .01290* (2.65) - .24647* (3.85) GDP Growth .04769 (1.65) .01985 (1.54) .12045 (1.27) .14882 (0.92) EDU -.04330* (-2.51) -.03292 (-1.57) -.21983 (-0.68) -.29323 (-0.94) TRADE -.01351** (-1.98) -.00364 (-0.63) (.01614) (0.41) .01798 (0.31) Constant .33824* (25.60) .31537* (23.62) 2.1140 * (8.97) 2.14726* (10.23) F-value 4.65* 14.16* 2.96** 3.97* R2_within _0.0498 _0.0361 _0.007 _00.0592 N 251 230 230 230 n 23 23 23 23

4_{Note: The first number is the coefficient estimate, the value in the brackets is the t-value.*indicates the significance of}

the coefficient at 0.05 level, ** indicates significance at 0.1 level. The explanatory variables of interest are presented in bold. N refers to the total number of observations for a given variable, n is the number of countries. R2_{within refers to the}

(24)

24

Table 6. Model 2, fixed effects estimates

Gini Index 90/10 decile ratio

IM -2.1798 (-1.72) - -2.06327 (-0.77) - LSM .02856 (0.3) - .14883 (0.33) - HSM .00996 (0.09) - .03742 (0.13) - ∆𝐈𝐌 - -.00048 (-0.47) - .01296 (1.27) ∆𝐋𝐒𝐌 - .00368** (2.15) - . 08746 (1.35) ∆𝐇𝐒𝐌 - .01266* (2.53) - .24168* (3.85) GDP Growth - - - - EDU -.06530* (-3.27) -.03763 (-1.79) -.1541 -0.29 -.32539 (-0.97) TRADE -.01108 (-1.52) -.00183 (-0.30) .01980 0.29 .03046 (0.45) Constant .35331* (16.17) .31701* (22.97) 2.0101* (4.34) 2.15837* (9.86) F-value 5.01* 6.11* 0.28 4.74** R2_within _0.0383 _0.0338 _0.0079 _0.0577 N 251 230 230 230 n 23 23 23 23

From the tables above, it can be noticed that excluding the real GDP growth per capita did not lead to any significant changes regarding the explanatory variables of interest. The results from both models imply that the newly permitted high-skilled immigrants have a significant positive effect on overall income inequality. 1% increase in newly permitted high-skilled immigrant share is associated with 0.01266 to 0.01290 points increase in Gini index and 0.24168 to 0.24647 points increase in 90/10 income decile ratio depending on the model used.

Contrary to the expectations, negative effect of total foreign born population on income inequality is found. However, these results a shown to be insignificant. The same holds for low and high-skilled immigrants. Although the coefficients are shown to be positive, none of the results are significant.

(25)

25

The overall results, imply that no relationship between immigration and income inequality exists. Therefore, the null hypothesis stated in previous section is not rejected.

Education level and trade openness, contrary to predictions, are shown to have a negative impact on inequality when Gini measure is used. GDP growth estimates imply a positive effect. This holds true for inequality measured by Gini coefficient and 90/10 decile ratio. These results are also shown to be insignificant. Similarly, the results are shown to be statistically insignificant.

Within R-squared measures for all equations are very small, implying a low statistical power of the explanatory variables included in the regressions. Several explanations for such performance are presented in the following section.

5. Discussion

A choice of the methodology could have been a major shortcoming of the study, leading to insignificant and inconclusive results. Bartels (2008) argues that fixed effect model might perform poorly for slowly changing variables over time. Although the overall Gini coefficient has increased over time in most countries, the changes could have been too small in order to capture any effect of immigration and other explanatory variables on income inequality. To get more insights, a longer time frame must be investigated, ideally, starting from 1980s when income inequality started increasing in most industrialized countries. However, no database providing information on low-skilled and high-low-skilled migration for European countries exists.

The methodology could also explain why a study by Xu et al., (2015) produced very different outcome. Their results were obtained by using pooled OLS regression with panel-correlated standard errors. However, such methodology might not have been the appropriate model specification for cross-sectional time series data as it disregards the panel effect (Baltagi, 2008). No discussion or justification regarding the panel effect was present in the paper. Nevertheless, pooled OLS methodology could solved the issue of slow variation in the dependent variable existent on this study, thus producing more significant results than the fixed effects. An alternative methodology widely used in previous literature is the IV regression. Such methodology is able to mitigate multiple issues arising due to the violations of assumptions. Therefore, IV regression could be another methodological solution, however, choosing appropriate instruments is out of the scope of this paper.

(26)

26

Furthermore, the period investigated in the study included the financial crisis in the middle of the period. As shown, during the beginning of the crisis immigration experienced a structural break in several countries and therefore the overall results could have been distorted. Unfortunately, the study failed to capture the effect of crisis for the variables of interest.

6. Conclusion

Immigration and income inequality have been one of the major EU problems for the last decade. Due to widespread negativity towards foreign-born population it is important to understand the impact of immigration on the labour market outcomes in receiving countries. Adverse wage effect predicted by economic theory provides a reason to believe that immigration has an impact on income distribution and could be responsible for a rise in income inequality. Therefore, the study aimed to estimate the effect of immigration on income inequality for 23 EU countries using a cross-sectional time series model from 2004-2014. It was done by employing fixed effects methodology with Driscoll-Kraay standard errors. Gini coeffient and 90/10 decile ratios were used as income inequality measures.

The paper found a positive effect of newly permitted high-skilled immigrants on income inequality. However, the overall results indicate that no significant effect of immigration on income inequality exists. Such inconclusive results could have arisen due to violations of several assumptions, unfavourable behaviour of the panel dataset and a choice of the methodology. Thus, empirical results have to be treated with cautious and are a subject of several limitations.

Taking the above into consideration, future research should focus on finding the appropriate estimation methodology in order to obtain the better understanding about the relationship between immigration and income inequality. Investigation of the longer time horizon could also lead to more conclusive results. As overall results suggested, a rather simplistic linear model was not enough to determine the effect of immigration on income inequality. Although the overall foreign-born population was subcategorized into two skill levels, further immigrant disaggregation, based on their demographics, such as ethnicity, gender or age, could help to evaluate how those individual groups are linked to income inequality with more accuracy.

(27)

27

References

Altonji, J. G., & Card, D. (1991). The effects of immigration on the labor market outcomes of

less-__skilled natives. In Immigration, trade, and the labor market (pp. 201-234). University of Chicago

__Press.

Alvaredo, F., Atkinson, A. B., Piketty, T., & Saez, E. (2013). The top 1 percent in international and

__historical perspective. The Journal of Economic Perspectives, 27(3), 3-20. Baltagi, B. (2008). Econometric analysis of panel data. John Wiley & Sons.

Bartels, B. (2008). Beyond" fixed versus random effects": a framework for improving substantive and

__statistical analysis of panel, time-series cross-sectional, and multilevel data. The Society for

__Political Methodology, 1-43.

Borjas, G. J. (1995). The Economic Benefits from Immigration. Journal of Economic Perspectives, 9(2),

__3-22.

Boomgaarden, H. G., & Vliegenthart, R. (2009). How news content influences anti‐immigration

__attitudes: Germany, 1993–2005. European Journal of Political Research, 48(4), 516-542..

Card, D. (2001). Immigrant inflows, native outflows, and the local labor market impacts of higher

__immigration. Journal of Labor Economics, 19(1), 22-64

Card, David. 2009. "Immigration and Inequality." American Economic Review, 99(2), 1-21.

Deininger, K., & Squire, L. (1996). A new data set measuring income inequality. The World Bank

__Economic Review, 10(3), 565-591.

Dreger, C., López-Bazo, E., Ramos, R., Vicente Royuela, V., & Suriñach, J. (2015). Wage and income

__inequality in the European Union. Policy Department Working Paper). Retrieved from

__http://www. europarl. europa. eu/RegData/etudes/STUD/2015/536294/IPOL_STU, 282015. Dustmann, C., Frattini, T., & Preston, I. P. (2013). The effect of immigration along the distribution of

__wages. The Review of Economic Studies, 80(1), 145-173.

Dustmann, C., & Frattini, T. (2011). Immigration: the European experience. Centro Studi Luca

__d'Agliano Development Studies Working Paper, (326).

Dustmann, C., Fabbri, F., & Preston, I. (2005). The impact of immigration on the British labour

__market. The Economic Journal, 115(507)..

Friedberg, R. M., & Hunt, J. (1995). The impact of immigrants on host country wages, employment

__and growth. The Journal of Economic Perspectives, 9(2), 23-44.

Hainmueller, J., & Hopkins, D. J. (2014). Public attitudes toward immigration. Annual Review of

__Political Science, 17, 225-249.

Hall, B. (2000). Immigration in the European Union: problem or solution?. Organisation for Economic

(28)

28

Hibbs, B., & Hong, G. (2015). An examination of the effect of immigration on income inequality: A

__Gini index approach. Economics Bulletin, 35(1), 650-656.

Hoechle, D. (2007). Robust standard errors for panel regressions with cross-sectional

__dependence. Stata Journal, 7(3), 281.

International Organization for Migration (2015).How the World Views Migration. Retrieved from

__https://publications.iom.int/system/files/how_the_world_gallup.pdf

Lichbach, M. I. (1989). An evaluation of “does economic inequality breed political conflict?”

__studies. World politics, 41(04), 431-470.

Fajnzylber, P., Lederman, D., & Loayza, N. (2002). Inequality and violent crime. The journal of Law

__and Economics, 45(1), 1-39.

Lemos, S. (2014). Labour market effects of Eastern European migration in Wales. The Manchester

__School, 82(5), 524-548.

Mellor, J. (1989). Dramatic Poverty Reduction in the Third World: Prospects and Needed Action.

__International Food Policy Research Institute , 18-20 .

Page, B. (2009). British attitudes to immigration in the 21st century. Migration, Public Opinion and

__Politics, eds. Bertelsmann Stiftung and Migration Policy Institute. Guetersloh, Germany:

__Bertelsmann Stiftung.

Raess, D., & Burgoon, B. (2015). Flexible work and immigration in Europe. British Journal of Industrial

__Relations, 53(1), 94-111.

Richardson, J. D. (1995). Income inequality and trade: how to think, what to conclude. The Journal of

__Economic Perspectives, 9(3), 33-55.

Smeeding, T. (2006). Poor people in rich nations: The United States in comparative perspective. The

__Journal of Economic Perspectives, 20(1), 69-90.

Stock, J. H., & Watson, M. W. (2011). Introduction to econometrics.

Thorud, E., Haagensen, E., & Jølstad, F. O. (2006). SOPEMI-report for Norway.

Xu, P., Garand, J. C., & Zhu, L. (2016). Imported Inequality? Immigration and Income Inequality in the

__American States. State Politics & Policy Quarterly, 16(2), 147-171.

Zorlu, A., & Hartog, J. (2005). The effect of immigration on wages in three European

(29)

29

Appendix 1.

Table 1. Composition of immigrant population by are of origin5

% Immigrants

in total population

Composition of immigrant population by area of origin

EU15 NMS 12 Other Europe North Africa & Middle East Other Africa South and East Asia North America and Oceania Latin America Austria 15.68 17.55 18.7 51.18 3.58 1.2 5.44 1.07 1.29 Belgium 11.76 41.53 6.45 13.83 18.09 10.96 5.48 1.16 2.5 Germany 14.5 25.36 8.38 46.9 7.16 2.33 6.14 2.14 1.6 Denmark 7.98 20.05 5.39 26.27 16.12 4.76 16.75 8.04 2.63 Spain 13.09 13.83 13.76 3.89 15.13 2.86 3.28 0.65 46.6 Finland 2.71 29.86 10.51 33.75 7.16 5.08 8.89 2.73 2.02 France 10.66 27.57 2.99 6.11 40.23 12.08 6.79 1.56 2.67 Greece 7.79 5.85 12.89 61.34 11.98 1.02 4.36 2.21 0.35 Ireland* 15.59 40.16 32.66 3.21 1.54 5.71 9.59 5.6 1.53 Italy 7.41 11.37 18.11 26.72 14.03 5.48 11.27 1.81 11.2 Netherlands 10.66 17.39 3.57 16.64 17.22 5.86 17.45 2.51 19.38 Norway 8.69 30.4 5.54 14.16 11.22 7.58 20.99 4.62 5.49 Portugal 6.48 18.51 3.06 8.31 0.23 45.04 1.73 2 21.12 Sweden 15.16 26.33 8.2 21.56 20.45 4.37 10.8 1.55 6.73 UK 11.34 18.08 13.47 3.56 4.62 16.93 29.05 7.67 6.61 Total 11.27 20.61 10.63 18.91 15.39 8.34 11.25 2.83 12.03 USA 12.50 7.44 3.23 2.57 2.82 3.04 24.75 2.79 53.37

(30)

30

Table 2. Education of natives and immigrants

% with lower secondary education

Natives Immigrants

% with tertiary education

Natives Immigrants Standard deviation of lower secondary education shares across origin groups Austria 16.33 33.93 17.51 18.07 14.00 Belgium 29.03 42.72 32.8 28.4 15.92 Germany 10.47 37.53 27.02 19.31 15.93 Denmark 23.78 27.10 33.18 33.41 10.11 Spain 50.72 40.60 30.15 24.38 19.70 Finland 19.59 24.54 36.75 31.86 10.65 France 28.38 46.07 27.58 23.98 12.68 Greece 39.25 46.08 22.9 15.69 19.09 Ireland 33.04 18.51 31.32 46.34 10.43 Italy 48.36 45.32 13.62 12.85 13.19 Netherlands 27.18 37.91 31.14 25.91 12.71 Norway 19.90 27.02 34.01 38.51 12.34 Portugal 74.69 52.41 13.01 21.82 14.01 Sweden 15.31 25.18 30.9 31.94 9.19 UK 30.00 24.28 30.57 33.96 6.79 Total 31.74 38.05 25.83 23.51 15.46

6_{The sample is restricted to working age population older than 25, not in full-time education and}

(31)

31

Appendix 2.

Table 3. Gini coefficient score on income inequality AFTER taxes and transfers, 2004 - 20147

2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 Change over time (2005 - 2013) Austria 25.8 26.3 25.3 26.2 27.7 27.5 28.3 27.4 27.6 27.0 27.6 0.7* Belgium 26.1 28.0 27.8 26.3 27.5 26.4 26.6 26.3 26.5 25.9 -2.1 Bulgaria 26 25.0 31.2 35.3 35.9 33.4 33.2 35.0 33.6 35.4 10.4 Croatia 30 30.0 28.0 29.0 28.0 27.0 31.6 31.2 30.9 30.9 0.9* Cyprus 28.7 28.8 29.8 29.0 29.5 30.1 29.2 31.0 32.4 3.7 Czech Republic 26 25.3 25.3 24.7 25.1 24.9 25.2 24.9 24.6 -1.4 Denmark 23.9 23.9 23.7 25.2 25.1 26.9 26.9 27.8 28.1 27.5 3.6 Estonia 37.4 34.1 33.1 33.4 30.9 31.4 31.3 31.9 32.5 32.9 -1.2* Finland 25.5 26.0 25.9 26.2 26.3 25.9 25.4 25.8 25.9 25.4 25.6 -0.6* France 28.2 27.7 27.3 26.6 29.8 29.9 29.8 30.8 30.5 30.1 2.4 Germany 26.1 26.8 30.4 30.2 29.1 29.3 29.0 28.3 29.7 3.6 Greece 33 33.2 34.3 34.3 33.4 33.1 32.9 33.5 34.3 34.4 34.5 1.2* Hungary 27.6 33.3 25.6 25.2 24.7 24.1 26.8 26.9 28.0 27.9 0.4* Ireland 31.5 31.9 31.9 31.3 29.9 28.8 30.7 29.8 29.9 30.0 -1.9* Italy 33.2 32.8 32.1 32.2 31.0 31.5 31.2 31.9 31.9 32.5 32.7 -0.3* Latvia 36.2 38.9 35.4 37.5 37.5 35.9 35.1 35.7 35.2 35.5 -1.0* Lithuania 36.3 35.0 33.8 34.5 35.9 37.0 33.0 32.0 34.6 -1.7* Luxembourg 26.5 26.5 27.8 27.4 27.7 29.2 27.9 27.2 28.0 30.4 3.9 Malta 27.0 27.1 26.3 28.1 27.4 28.6 27.2 27.1 27.9 0.9* Netherlands 26.9 26.4 27.6 27.6 27.2 25.5 25.8 25.4 25.1 -1.8* Poland 35.6 33.3 32.2 32.0 31.4 31.1 31.1 30.9 30.7 -4.9 Portugal 37.8 38.1 37.7 36.8 35.8 35.4 33.7 34.2 34.5 34.2 -3.9 Romania 31 31 33 37.8 36.0 34.9 33.3 33.2 33.2 34.0 3.0 Slovakia 26.2 28.1 24.5 23.7 24.8 25.9 25.7 25.3 24.2 -2.0 Slovenia 23.8 23.7 23.2 23.4 22.7 23.8 23.8 23.7 24.4 0.6* Spain 31 32.2 31.9 31.9 31.9 32.9 33.5 34.0 34.2 33.7 34.7 1.5* Sweden 23 23.4 24.0 23.4 24.0 24.8 24.1 24.4 24.8 24.9 1.5* United Kingdom 34.6 32.5 32.6 33.9 32.4 32.9 33.0 31.3 30.2 -4.4 Source: Eurostat

7_{Retrieved from http://ec.europa.eu/eurostat/data/database. * implies that the change over time is not great enough to}

(32)

32

Appendix 3. The list of countries

1 Austria 2 Belgium 3 Cyprus 4 Czech republic 5 Denmark 6 Estonia 7 Finland 8 France 9 Germany 10 Greece 11 Hungary 12 Iceland 13 Ireland 14 Italy 15 Latvia 16 Malta 17 The Netherlands 18 Norway 19 Portugal 20 Slovenia 21 Sweden 22 Switzerland

(33)

33 -. 1 -. 0 5 0 .0 5 .1 R e si d u a ls .2 .4 .6 .8 1 EDU

Appendix 4. Residuals plotted against the independent variables

-. 1 -. 0 5 0 .0 5 .1 R e si d u a ls .1 .2 .3 .4 .5 i_HSM -. 1 -. 0 5 0 .0 5 .1 R e si d u a ls -.15 -.1 -.05 0 .05 .1 GDPGrowth -. 1 -. 0 5 0 .0 5 .1 R e si d u a ls 0 .2 .4 .6 i_LSM -. 1 -. 0 5 0 .0 5 .1 R e si d u a ls 0 1 2 3 TRADE -. 1 -. 0 5 0 .0 5 .1 R e si d u a ls 0 .01 .02 .03 .04 i_IM

(34)

34

Appendix 5. Normality of the residuals

Figure 15. Density of the residuals

Figure 15. Quintiles of the residuals

0 5 10 D e n si ty -.1 -.05 0 .05 .1 Residuals

Kernel density estimate Normal density

kernel = epanechnikov, bandwidth = 0.0106

Kernel density estimate

-. 1 -. 0 5 0 .0 5 .1 R e si d u a ls -.1 -.05 0 .05 .1 Inverse Normal

(35)

35

Appendix 6. Summary of statistics

Table 6. Correlations

GINI IM LSM HSM GDPGrowth EDU TRADE

GINI 1 IM -0.2461 1 LSM 0.0801 -0.0325 1 HSM -0.2432 0.4286 -0.5248 1 GDPGrowth -0.0873 0.1922 -0.1487 0.1902 1 EDU -0.3324 -0.0281 -0.6259 0.2157 0.0364 1 TRADE -0.2146 0.1663 -0.2382 0.2329 0.1711 -0.148 1

Figure 7. Trends in Gini coefficient (after taxes)

.2 .2 5 .3 .3 5 .4 .2 .2 5 .3 .3 5 .4 .2 .2 5 .3 .3 5 .4 .2 .2 5 .3 .3 5 .4 .2 .2 5 .3 .3 5 .4 2005 2010 2015 2005 2010 2015 2005 2010 2015 2005 2010 2015 2005 2010 2015 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 18 19 20 22 24 26 27 28 G IN I Year Graphs by country1

(36)

36

Figure 8. Trends in immigration

0 .0 1. 0 2 .0 3. 0 4 0 .0 1. 0 2 .0 3. 0 4 0 .0 1 .0 2. 0 3 .0 4 0 .0 1 .0 2. 0 3 .0 4 0 .0 1. 0 2 .0 3. 0 4 2005 2010 2015 2005 2010 2015 2005 2010 2015 2005 2010 2015 2005 2010 2015 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 18 19 20 22 24 26 27 28 i_ IM Year Graphs by country1

(37)

37

Table 6. Descriptive statistics

Variable Mean

Std. Dev. Min Max Observations

GINI 0.289045 Overall 0.038202 0.225 0.389 N = 2508 Between9_0.036269 _0.2384909 _{0.3629 n = 23} within 0.014362 0.2559357 0.354227 S90/10 7.404103 overall 2.270618 4.545455 14.6 N = 228 between 2.084563 5.021819 12.51108 n = 23 within 0.971762 5.198898 14.24255 IM 0.00968 overall 0.006618 0.0008029 0.040777 N = 250 between 0.005824 0.0022126 0.021861 n = 23 within 0.003328 0.0000493 0.028597 LSM 0.343802 overall 0.132025 0.054 0.633 N = 250 between 0.128685 0.1282 0.554364 n = 23 within 0.039727 0.1864384 0.461711 HSM 0.247072 overall 0.095022 0.065 0.485 N = 250 between 0.087043 0.106 0.438818 n = 23 within 0.040885 0.140072 0.43089 GDP Growth 0.01458 overall 0.036867 -0.147 0.119 N = 250 between 0.010411 -0.0132727 0.031 n = 23 within 0.035433 -0.1615109 0.10758 EDU 0.734552 overall 0.155773 0.249 0.932 N = 250 between 0.157272 0.3186364 0.913455 n = 23 within 0.031116 0.6506429 0.848916 TRADE 1.094534 overall 0.516414 0.4560911 3.24502 N = 250 between 0.521866 0.5283434 2.85315 n = 23 within 0.121189 0.3532638 1.486403

8_{N refers to the total number of observations for a given variable, n is the number of countries.}

Immigration and income inequality in the EU from 2004-2014