Territorial distribution of immigrants in Europe

University of the Basque Country UPV/EHU Master in Economics: Empirical Applications and Policies Territorial Distribution of Immigrants in Europe Author: Aiert Gurrutxaga Urzaa Supervisors: María José Gutiérrez & Susan Orbe 2019/20 Index Abstract ......................................................................................................................................... 4 1.- Introduction ............................................................................................................................. 5 2.- Data ........................................................................................................................................ 12 3.- Segregation metrics ............................................................................................................... 12 3.1.- Indices ............................................................................................................................. 12 3.2.- Results ............................................................................................................................ 17 4.- Determinants of segregation metrics ................................................................................... 23 5.- Conclusions ............................................................................................................................ 40 6.- Bibliography ........................................................................................................................... 42 Figures Index Figure 1: Example of the Kuznets curve between segregation and GDP ...................................... 6 Figure 2: Percentage of immigrant population at regional level (NUTS2), 1999-2019 ................. 9 Figure 3: Social network of a secondary education school. Source: “Networks, Crowds and Markets. Reasoning about a highly connected world” ............................................................... 11 Figure 4:Evolution of dissimilarity, entropy and Atkinson 0.5 indices at European level, 19992019 ............................................................................................................................................. 18 Figure 5:Dissimilarity index for each country, 1999-2019 .......................................................... 19 Figure 6: Dissimilarity index for Romania, Spain and Ireland, 1999-2019 .................................. 20 Figure 7: Atkinson 0.5 index for Romania, Spain and Ireland, 1999-2019 .................................. 21 Figure 8: Entropy index for Romania, Spain and Ireland, 1999-2019 ......................................... 22 Figure 9:Evolution of different macroeconomic variables, 1999-2019 ...................................... 25 Figure 10: Evolution of density, 1999-2019 ................................................................................ 25 Tables Index Table 1: Top 5 countries with the highest percentage of immigrant population in EU-15, 19992019 ............................................................................................................................................... 7 Table 2: Top 5 regions with the highest percentage of immigrant population in EU-15, 19992019 ............................................................................................................................................... 8 Table 3:Definition and formula of dissimilarity, entropy and Atkinson indices .......................... 16 Table 4: Macroeconomic variables and their measure ............................................................... 24 Table 5: Estimation of the empirical model for the dissimilarity index ...................................... 27 Table 6. Econometric model of dissimilarity index ..................................................................... 29 Table 7: Estimation for the empirical model for the Atkinson 0.5 index .................................... 32 Table 8: Econometrical model of the Atkinson 0.5 index ........................................................... 34 Table 9: Estimation for the empirical model for the entropy index ........................................... 36 Table 10: Econometric model of entropy index .......................................................................... 38 Table 11: Dissimilarity, entropy and Atkinson 0.5 indices' value and number index at European level, 1999-2019 .......................................................................................................................... 44 Abstract Uneven distribution of immigrant population has diverse consequences in countries and societies. We call immigrants those that have a nationality different to the reporting country. Some European regions have an important percentage of immigrant population. This articles aims to quantify this unevenness in the distribution from 1999 to 2019. For this we use segregation indices of the evenness dimension. The three indices we use show that immigrant population is unevenly distributed at European level and at country level. However, this heterogeneity has decreased. We also look for macroeconomic determinants of these indices. We find a very interesting relation between GDP and the indices, as there is Kuznets effect. Being aware that there are inequalities and how different socioeconomic variables can affect is helpful in order to adopt successful policies to enjoy the positive effects immigration can provide. Keywords: NUTS2 regions; Immigrants; Segregation indices; Inequality measures; Kuznets effect 5 1.- Introduction The aim of this article is to study the territorial distribution of immigrants across Europe in recent years. This is a very important socioeconomic subject in nowadays Europe, partly because, according to Sardadvar & Rocha (2016), over the past quarter century, Europe has experienced three important events which eased migration within the continent: the dissolution of the Comecon, the Treaty on European Union and the European Union accession of new countries. One of the reasons of why population exoduses are really important phenomena is that they have a huge impact in the destination country. People leave their country of origin for many reasons, such as improving living condition, or, simply, to escape social distress (Jones, 2015), or from natural disasters (López-Carr, Marter-Kenyon, 2015) or war. Once they make the decision to leave, they search for what might be the best destination. Macroeconomic variables of the destination country, such as, employment, education level or how rich the country is may affect the immigrant’s selection. On the other hand, immigrants also affect destination countries in several dimensions, for example, they increase the number of available labor, propelling the economic growth. However, we can see that for migrant population, or at least for those with some non-native characteristics is much more difficult to obtain certain jobs, as explain Scheve and Slaughter (2001) and Mayda (2006) . Another example is the effect they have on demography. Usually immigrants are young and end up contributing to increasing birth rates in destination countries. An example of negative effect could be, for example, that the economic gap between the origin and destiny countries can increase, especially if those leaving the country are the most qualified ones. Also a part of the population does not like foreigners coming to their country and having jobs, they claim that these individuals steal the jobs (Scheve & Slaughter, 2001; Mayda, 2006) and that they are criminals. The truth is that they have less job opportunities, as said before. As a consequence to this statements made by many populist political parties, in some countries with a high share of foreigners there has been a rise in the racist and xenophobic behaviours and ultra-rightist political parties’ power. In the second part of the article we focus 6 on how some these macroeconomic variables affect different segregation indices. The most important thing we want to study when searching for determinants of these segregation metrics is the possible effect of the Gross Domestic Product. One of the objectives will be to see if there is a Kuznets curve between the GDP and different segregation indices used. If this result is obtained we will see that the segregation of immigrants increases as countries become richer, but only until a point, once the country has a per capita GDP higher than that, the relation between indices and GDP will be negative. We can see this evolution on Figure 1. Figure 1: Example of the Kuznets curve between segregation and GDP It is because of segregations importance, in all aspects, and because “migration is a ubiquitous phenomenon” (Mazzoli et al, 2020) that there are several works in relation to segregation and different ways of studying it. The importance of the subject has an increasing tendency, and even more with the possible crisis that is coming. According to Eurostat, around 9.29% of the population of the European Union (15 countries, EU-15) had a different nationality to the respective country in 2019; this number doubles that of 1999 (4.5%). This shows that the weight of non-national population has increased in Europe during these two decades. However, it must be taken into account that these figures may not represent the 7 real numbers of immigrants for several reasons. Among others, illegal immigrants are difficult to be recorded, as they do not always appear in any register. Table 1: Top 5 countries with the highest percentage of immigrant population in EU-15, 1999-2019 Table 1 shows the five countries of EU-15 with the highest percentage of immigrant population for years 1999 and 2019. We can clearly see that; like Czaika and Di Lillo (2018) point out immigrants are making up a continuously growing proportion of the European population. For example, Belgium is the country with the highest percentage of immigrant population in 1999, but still has a lower percentage than the fifth country in 2019, shows this. During these twenty years the relative importance of France and Sweden in this aspect has decreased, since they do not appear in the top five of 2019, being substituted by Spain and Ireland. As previously said, in some countries ultra-right movements have increased together with the increase of the percentage of immigrant population. This is the case of Austria, where we can see that the percentage has nearly doubled during the time period studied, Austria is one of the countries where the far-right movements more have increased too, having the Freedom Party Austria (FPÖ) a high percentage of votes. This phenomenon is not exclusive to Austria, it has been observed in countries like France and Greece. However, this data is at country level, so differences in internal level are perfectly possible, as we see in the next table. Top 5countries in EU-15 1999 2019 Belgium 7.39% 12.89% Austria Germany 6.74% 11.36% Ireland Austria 6.48% 10.57% Germany France 4.63% 9.05% Spain Sweden 3.63% 8.89% Belgium 8 Top 5 regions in EU-15 1999 2019 Country Regions % % Regions Country Belgium Région de BruxellesCapitale / Brussels Hoofdstedelijk Gewest 26.98% 33.74% Comunidad de Madrid Spain United Kingdom Inner London 21.32% 33.31% Région de BruxellesCapitale / Brussels Hoofdstedelijk Gewest Belgium France Corse 16.86% 31.29% Région lémanique Switzerland Austria Wien 15.58% 29.06% Wien Austria Belgium Prov. Hainaut 14.88% 27.57% Ticino Switzerland Table 2: Top 5 regions with the highest percentage of immigrant population in EU-15, 1999-2019 In Table 2 we see the top 5 regions from EU-15. We observe that there are differences between countries and regions; for example, the second region with the highest percentage of immigrants is Inner London, but United Kingdom does not appear in Table 1. In 2019 the same happens, Swiss regions of Ticino and Région Iémanique. We also observe that Comunidad de Madrid is the region with the highest percentage of immirants in 2019, being Spain the second country in this year. Table 2 shows the fact that countries change when taking a look at regions. Apart from that, Table 2, like happened with Table 1, shows that there has been an increase in the weight of immigrant population. Like happened in Table 1, the first region in 1999 has a lower percentage of immigrants than the fifth region in 2019. 9 Figure 2: Percentage of immigrant population at regional level (NUTS2), 1999-2019 Figure 2 shows the evolution and distribution of the immigrants at regional level (NUTS2). For this we have maps for 1999 and 2019. The maps show the percentage of population with a nationality different to the one of the reporting country. We can observe that the immigrants are heterogeneously divided across countries, and the same happens inside the countries, as Table 1 and Table 2 showed. In the case of these maps we have intervals of percentage of immigrants. Those areas with the light green represent countries where the immigrant population is five percent or less of total population. Areas gets darker when there is more percentage of immigrants. The second lightest green shows countries where immigrant population is between five and ten percent of total population. The second darkest green are countries where the percentage of immigrant population is between fifteen and twenty percent. Lastly we have the dark green, which represents countries where immigrant population is above twenty percent. Notice that data in 2019 covers more countries than data from 1999, we have countries such as Italy or Switzerland, for which we have no data in 1999. We also observe a higher immigrant concentration in 1999. For example, we see that in Spain there is no region that stands out more than the rest; however, this can be because of the intervals selected to colour the map. We can also see that in 2019 countries, and especially regions, are much more differentiated than in 1999. 1999 2019 16 Table 3:Definition and formula of dissimilarity, entropy and Atkinson indices DEFINITION / CHARACTERISTICS FORMULA Dissimilarity Index or Delta (D) The most widely used index. Measures the percentage of the group’s population that would have to change residence for each region to have the same percentage of the groups as the country overall. ∑[|(−)|]   [2(1−)] Entropy Index or Information Index (H) It measures the weighted average deviation of each region from the country diversity. It is bigger when each group is equally represented in the area.     (  −   )     = 1 !+(1−) 1 1−!  =  1  ! + ( 1 −  )  1 1 −  ! Atkinson Index (A) Allows to weight regions at different points on the distribution. Permitting that where minorities are under or over-represented contribute in a more important way to the overall index. 1−  1−!&(1−)'((     & '( 17 These indexes satisfy some of the four criteria that were established by James and Taeuber (1985). These properties are characteristics to have a good index; this is, indices that fulfil the four principles are good indices. This are the transfer principle, which says that a measure should be sensitive to redistribution of the immigrant population among countries with immigrant proportions above or below the region’s. The second criterion is the compositional invariance, so the relative size of the immigrant population should not affect the index. Third we have the size invariance, which states that the measure should not be affected if the number of immigrant and non-immigrant population is multiplied by a constant. And fourth, the organizational equivalence, which means that the index should be unaffected by aggregating units with the same minority composition. The three indices that we study do not fulfil all the principles. In the case of the dissimilarity index, it fails to satisfy the transfer principle. On the other hand, the same happens with the compositional invariance in the case of the entropy index. The Atkinson index is the only index, out of those we use, that satisfies without any problem the four criteria. Massey & Denton worked with a total of 20 indices, but the reduced the list to five, just one index for each one of the dimensions. So, to summarize, in the evenness case, that is the one in which we are interested, the best option is the dissimilarity index by Duncan and Duncan as it contains most information that the other indices can provide. Apart from that, there are continuity and ease reasons, as most of previously written literature uses this index. As said we will be using two more indices, the entropy index, which is an index that works well too, and it can be divided to see the weight of each one of the groups analysed. This decomposition might be useful to see to which extent the inequalities are generated within groups and to which extent are caused because of differences between groups. However, we do not use that property in this paper; and the Atkinson index, which is also a great choice, since it fulfils the four principles previously explained. That is why we decide to focus on these three indices. 3.2.- Results We are not interested only in the results obtained by using the indices, we also want to observe the evolution they had during the time period studied, to see how they all evolve in the same way, even if they make the measures in different ways. 18 We calculate the uniformity in the distribution of immigrants in two levels. First of all, we measure it at European level, using every NUTS2 region from our country sample. Second, we measure it at country level, using every NUTS2 region of each country. This analysis is made for each one of the years of our sample, 1999-2019. Figure 4:Evolution of dissimilarity, entropy and Atkinson 0.5 indices at European level, 1999-2019 Figure 4 shows the evolution of the dissimilarity index, the entropy index and the Atkinson index with 0.5 parameter for Europe. In this case, in which we study the indices at European level, we analyse Europe as if it was a country as a whole, with all the NUTS2 regions taken inside. In order to see results in a clearer way, the values obtained have been normalized, using 1999 as the base year. It can be observed that the three indices have a similar evolution, and, in general, a negative trend. What means that the immigrant population is more evenly distributed in 2019 than in 1999. These results corroborate the intuition generated from the first glance. This is, the intuition that segregation decreases with time. This could happen because, with the passage of time, immigrants try to move to countries where work possibilities exist but where the first waves of migration did not arrive. We observe an increase in the indices’ value in the years 2004 and 2005; this may happen because we obtain data for some countries at first time in 80 85 90 95 100 105 110 115 120 Europe Dissimilarity Entropy Atkinson 0.5 19 this period. Examples of these countries are Italy or Romania, which have not data until these years. If we take a look to the European indices of dissimilarity, we can see that the dissimilarity index is the one that shows highest values until 2003. After this year it is the one below for every other year. It is also the index with the most uniform trend. With this information we can say that, according to the dissimilarity index, the segregation has been reduced over time. The other two indices will give a similar result, but they show a smaller reduction of the segregation than the dissimilarity index, since during these two decades the value of the dissimilarity index has reduced in 18.1% against the reduction of 16.35% and 16.63% of the entropy and Atkinson indices respectively. Figure 5:Dissimilarity index for each country, 1999-2019 First of all, we calculate the dissimilarity index, which is the most widely one used. Figure 5 shows the index for the 1999-2019 period for the European countries for which we have available data. We can see that the values are below 0.4 in most cases. The highest value is obtained by Romania, being around 0.55. At the same time, it evidences differences in segregation across countries; which mean that inside each country “immigrants” are not homogeneously distributed. We also observe that the evolution of the index changes between countries. However, during these two decades the value of the index has decreased in most of these 0 0,1 0,2 0,3 0,4 0,5 0,6 199920002001200220032004200520062007200820092010201120122013201420152016201720182019 Dissimilarity index for each country Belgium Bulgaria Czechia Germany Ireland Greece Spain France Italy Netherlands Austria Portugal Romania Slovenia Slovakia Finland Sweden Unitd Kingdom Norway Switzerland 20 countries, which means that heterogeneity in the territorial distribution has decreased according to this index. In this Figure 5 we observe that some countries are missing, this happens because countries such as Luxembourg or Malta have only one regions, and, as a consequence, the index value they obtain is always zero. Figure 6: Dissimilarity index for Romania, Spain and Ireland, 1999-2019 Figure 6 shows the evolution of the dissimilarity index in the three chosen countries. In this case we have chosen Romania (blue line), Spain (red line) and Ireland (green line). We can see that, for example, in Romania immigrants are more unevenly distributed than in Spain or Ireland, that is all it shows, it has nothing to do with the number of immigrants. In the case of Romania, we see an irregular increase. It is the country that reaches the highest values and it ends being the country with the highest immigrant concentration. In the case of Spain, we see values that show that it is neither the country where the immigrants more segregated are nor the country with the most homogeneous geographical distribution. We also see that it has a relatively flat curve, which means that in terms of concentration it has not changed too much. Last we have the case of Ireland, where we observe an increase in the unevenness distribution of 0 0,1 0,2 0,3 0,4 0,5 0,6 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Dissimilarity index Romania, Spain and Ireland Belgium Bulgaria Czechia Germany Ireland Greece Spain France Italy Netherlands Austria Portugal Romania Slovenia Slovakia Finland Sweden Unitd Kingdom Norway Switzerland 21 immigrants. During most part of the period is the country analysed with the lowest index value. There can be many reasons to make this happen, for example, it could be that most immigrants in Romania are there for work reasons and that specific types of jobs are concentrated, causing an irregular distribution of these immigrants. Figure 7: Atkinson 0.5 index for Romania, Spain and Ireland, 1999-2019 In Figure 7 we have the same analysis for the Atkinson index. This index allows us to choose the value of the parameter b. Figure 7 shows the calculated index for b=0.5. This is, as said by Iceland et al. (2002), the index will be more sensitive to changes in the middle part of the distribution, when underrepresented and overrepresented areas contribute equally. In this case we obtain lower values than in the dissimilarity case. Romania has the peak value too, and also the highest value of 2019, while Sweden has the lowest value of this year. We can see that the evolution of different countries is virtually the same that what we have just seen with the dissimilarity index. If we take a look to the same three countries previously projected, we see in a clear way that results are, although quantitatively different, qualitatively similar. 0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Atkinson index 0.5 for Romania, Spain and Ireland Belgium Bulgaria Czechia Germany Ireland Greece Spain France Italy Netherlands Austria Portugal Romania Slovenia Slovakia finland Sweden United Kingdom Norway Switzerland 22 Figure 8: Entropy index for Romania, Spain and Ireland, 1999-2019 Lastly, we have the entropy index, also known as the information index. As we can see in Figure 8, the values obtained are different to the ones obtained above, since they are calculated in a different way. However, we see that in all cases the values are not really high for any country. Once again we observe similar quantitative results than with the other indices used. In the three cases we observe an irregular increase for Romania, a pretty similar trend in Spain and low values with a little increase in Ireland. To summarize, and as a conclusion of this first part, when analysing at European level, we can see that, in general, the evolution during the period is very similar for the different indices used. Taking a look to the values of the calculated segregation indices presented in Table 11 in appendix, we can see that segregation has decreased at European level between 1999 and 2019. However, with each index the intensity of this trend is different. This strength is bigger in the case of the dissimilarity index, where the value of the index reduces in more or less 18% while in the case of the other two indices we can see that the value is around 16% smaller in 2019 than in 1999. 0 0,02 0,04 0,06 0,08 0,1 0,12 0,14 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Entropy index for Romania, Spain and France Belgium Bulgaria Czechia Germany Ireland Greece Spain France Italy Netherlands Austria Portugal Romania Slovenia Slovakia Finland Sweden United Kingdom Norway Switzerland 23 In the case of the country level analysis, we see that even if the values obtained with each index are different, they show virtually the same. Data shows heterogeneity at country level. Most of the countries show a decreasing trend throughout the years, fulfilling the expectation in this case too. In some countries immigrants are much more concentrated, for example in Romania and Italy, while in other countries the distribution is nearly uniform, the cases of Ireland and Sweden. So with all the three indices we reach the conclusion that immigrants are geographically unevenly distributed. For example, immigrant population is more homogeneously distributed in the territory in Ireland than in Romania. However, the most important and remarkable point is that the results are qualitatively similar even if they are quantitatively different. 4.- Determinants of segregation metrics In this second part of the paper we want to see what macroeconomic variables have an effect on the segregation indices and how affect them. For this we use all data together taking into account its panel structure. In order to analyse the effect that some macroeconomic variables have over the segregation indices. A fixed effects model is specified, ) ,+ =,  +-. ,+ +/ ,+ where the dependent variable, Y i,t, is the segregation index. We have an index for each country and year. Subscript i is used to indicate the country, and subindex t indicates the year, we have 19 countries and 21 years in this case. In this part we work with 19 countries because we do not use countries with only one region, like Luxembourg or Estonia, because the value of their indices is always zero. X i,t is the regressor. These regressors are some macroeconomic variables that can be a priori determinants of the segregation. Last, u i,t is the error term. With these fixed country and time effects, Bentivogli and Pagano (1999) state that regional net migration is sensitive to changes in income disparities, but unresponsive to changes in the relative unemployment rates. 24 The macroeconomic variables we will be using are: Unemployment rate, Density, Elderly population, R&D expenditure, Gross Domestic Product, Foreigners, Fertility rate, New citizenship, Education and Government expenditure. We can see how each one is measured in Table 4. Variables Measures Unemployment rate Percentage of labor force that is jobless Density Population per square kilometre Elderly population Percentage of population with age over 65 R&D Expenditure Percentage of gross domestic product used in R&D GDP In thousands per capita Foreigners Percentage of non-national population Fertility rate Average number of children that would be born to a woman over her lifetime. New citizenship The population that each year has received the nationality of the reporting country as percentage of total population Education Adult population with tertiary education as percentage of total population Government expenditure Percentage of the Gross Domestic Product Table 4: Macroeconomic variables and their measure 25 Figure 9:Evolution of different macroeconomic variables, 1999-2019 Figure 10: Evolution of density, 1999-2019 In Figure 9 we can see the evolution of most of these macroeconomic variables. Taking a look to Figure 9, we can see three groups. In the first group we have those variables that have a little change during this period. In this first group we have “New citizenship”, “R&D Expenditure”, “Fertility rate” and “Unemployment rate”. Then we have variables that have increased through time, mostly in a uniform way. Examples of this are “Foreign population”, “Elderly population” and 0 5 10 15 20 25 30 35 40 45 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 Macroeconomic variable Pop.Over65 Unemployment R&DExpenditure Foreigners Fertility Citizenship Education GDP/1000 130 132 134 136 138 140 142 144 Density 32 Table 7: Estimation for the empirical model for the Atkinson 0.5 index Model 1 Model 2 Coefficient Significance R 2 , error criteria and p value Coefficient Significance R 2 , error criteria and p value Elderly population 0.0005 R 2 = 0.787739 Schwarz criteria= -1,282.430 Akaike criteria= -1,397.298 Hannan-Quinn criteria= - 1,351.528 p-value=1.25e-055 0.0004 R 2 = 0.789555 Schwarz criteria= -1,313.996 Akaike criteria= -1,425.710 Hannan-Quinn criteria= -1,381.234 p-value=3.02e-057 Unemployment rate -0.0009 -0.0013 * GDP 0.0012 ** 0.0011 ** GDP_sq -9.41e-06 ** -9.20e-06 ** R&D Expenditure 0.0078 0.0085 Density -0.0018 *** -0.0018 *** Foreigners -0.0043 ** -0.0042 ** Fertility rate 0.0088 0.0087 New citizenship -0.0027 Gov. Expenditure 0.0002 0.0003 Education 0.0017 0.0018 33 Model 3 Model 4 Coefficient Significance R 2 , error criteria and p value Coefficient Significance R 2 , error criteria and p value Elderly population R 2 = 0.789536 Schwarz criteria= -1,319.817 Akaike criteria= -1,427.678 Hannan-Quinn criteria= -1,384.737 p-value= 2.92e-055 R 2 = 0.789425 Schwarz criteria= -1,325.486 Akaike criteria= -1,429.495 Hannan-Quinn criteria= -1,388.087 p-value=6.13e-056 Unemployment rate -0.0013 * -0.0014 ** GDP 0.0012 ** 0.0013 *** GDP_sq -9.41e-06 ** -1.01e-05 *** R&D Expenditure 0.0087 0.0094 * Density -0.0019 *** -0.0019 *** Foreigners -0.0044 *** -0.0045 *** Fertility rate 0.0078 New citizenship Gov. Expenditure 0.0003 0.0003 Education 0.0020 * 0.0020 * 34 Fixed effects, using 348 observations Dependent variable: Atkinson 0.5 Index Beck-Katz standard deviations Coefficient t-statistic Constant 0.3099 9.544 *** Unemployment rate -0.0013 -2.302 ** GDP 0.0013 3.510 *** GDP_sq -1.07e-05 -3.686 *** R&D Expenditure 0.0099 1.897 * Density -0.0019 -6.663 *** Foreigners -0.0043 -3.444 *** Education 0.0019 1.869 * R-square MCVF (LSDV) 0.789250 Log-verosimilitud 741.6032 Akaike criteria -1431.206 Schwarz criteria -1331.049 Hannan-Quinn criteria -1391.332 rho 0.450756 Durbin-Watson 0.997583 Robust contrast for different intercepts in groups - Null hypothesis: Groups have a common intercept Contrast statistic: Welch F (18, 119.5) = 77.0527 with p value = P (F (18, 119.5) > 77.0527) = 6.51e-057 Table 8: Econometrical model of the Atkinson 0.5 index As before, many specifications have been estimated and compared in terms of error criteria. The estimation results for the selected specification is shown in Table 8. Some significant variables are familiar from the dissimilarity index analysis: unemployment rate, density and foreigners. These variables maintain the same sign they had before; this is; the three variables affect in a negative way. So, if unemployment increases in one percent, the immigrant population will be more evenly distributed, as the value of the index is lower. In the same way, if density increases, or the population of foreigners increases in one percent, the immigrant population will be more homogeneously distributed in the territory. We also see that the adjustment is good, since 78.925% of the variability of the Atkinson index with parameter 0.5 is explained by the variability of these independent variables. Like happened in the analysis of the dissimilarity index, the log-likelihood obtained by this model is the highest out of the specifications used, also the values of the criteria are the lowest. 35 In this case we have new significant variables, the “R&D Expenditure” and “Education”. In both cases the effect is positive. So, if the expenditure made in R&D, as percentage of GDP, increases in one percent, the estimated increase in the expected segregation index is 0.0099, so the distribution of immigrant population will be more uneven. This may be because Research and Development may concentrate qualified work. In the case of education, if adult population with tertiary education, as percentage of total population, increases in one unit, the estimated increase in the expected segregation index is 0.0019, causing a more heterogeneous distribution of immigrant population. This might be caused due to qualified jobs requiring this type of education being geographically concentrated. If we take a look to the GDP and its’ square we can see that the Kuznets effect we saw in the dissimilarity index case is still present. So, those countries with GDP above minimum threshold have a negative relation between GDP and the index. This is, when GDP increases the value of the index decreases in these countries. Following the same process previously done, we get that the minimum threshold GDP is, in this case, 22.067. This means that the relation between these two variables will be negative when the GDP per capita is bigger than 22,067. This threshold is smaller than in the previous case, so those countries plus some new will be included. These additional countries are the Czech Republic, Portugal and Slovenia. Last, we will analyse the entropy index. In the same way than in previous cases, many specifications have been estimated. Table 9 shows this process until having obtained the last model. In this case we reject the equality of the intercepts in the estimated models too. 36 Table 9: Estimation for the empirical model for the entropy index Model 1 Model 2 Model 3 Coefficient Significance R 2 , error criteria and p value Coefficient Significance R 2 , error criteria and p value Coefficient Significance R 2 , error criteria and p value Elderly population 0.0006 R 2 = 0.824461 Schwarz criteria= -1,852.645 Akaike criteria= - 1,967.425 Hannan-Quinn criteria= - 1,921.685 p-value= 1.76e049 0.0006 R 2 = 0.824437 Schwarz criteria= - 1,858.425 Akaike criteria= - 1,969.379 Hannan-Quinn criteria= - 1,925.164 p-value= 1.41e049 0.0005 R 2 = 0.822473 Schwarz criteria= - 1,901.886 Akaike criteria= - 2,009.586 Hannan-Quinn criteria= - 1,966.699 p-value= 1.39e051 Unemployment rate -0.0003 -0.0003 -0.0003 GDP 0.0009 *** 0.0009 *** 0.0009 *** GDP_sq -6.48e-06 *** -6.64e-06 *** -6.39e-06 *** R&D Expenditure 0.0107 *** 0.0109 *** 0.0106 *** Density -0.0007 *** -0.0007 *** -0.0007 *** Foreigners -0.0007 -0.0007 -0.0006 Fertility rate 0.0017 New citizenship 0.0033 0.0032 Gov. Expenditure 0.0002 0.0002 0.0002 Education -0.0006 -0.0006 -0.0006 37 Model 4 Model 5 Model 6 Coefficient Significance R 2 , error criteria and p value Coefficient Significance R 2 , error criteria and p value Coefficient Significance R 2 , error criteria and p value Elderly population 0.0004 R 2 = 0.822196 Schwarz criteria= - 1,907.191 Akaike criteria= - 2,011.045 Hannan-Quinn criteria= -1,969.690 p-value= 6.58e-061 R 2 = 0.822066 Schwarz criteria= - 1,912.786 Akaike criteria= - 2,012.793 Hannan-Quinn criteria= - 1,972.970 p-value= 3.41e059 R 2 = 0.821696 Schwarz criteria= -1,917.913 Akaike criteria= - 2,014.074 Hannan-Quinn criteria= - 1,975.782 p-value= 3.48e059 Unemployment rate -0.0002 -0.0002 -0.0003 GDP 0.0009 *** 0.0009 *** 0.0008 *** GDP_sq -6.76e-06 *** -6.89e-06 *** -6.47e-06 *** R&D Expenditure 0.0109 *** 0.0111 *** 0.0102 *** Density -0.0007 *** -0.0007 *** -0.0007 *** Foreigners -0.0005 -0.0007 -0.0009 * Fertility rate New citizenship Gov. Expenditure Education -0.0006 -0.0004 38 Fixed effects, using 350 observations Dependent variable: Entropy Index Beck-Katz standard deviations Coefficient t-statistic Constant 0.1212 10.54 *** GDP 0.0008 4.281 *** GDP_sq -5.84e-06 -4.164 *** R&D Expenditure 0.0101 3.015 *** Density -0.0008 -8.814 *** Foreigners -0.0009 -1.867 * R-square MCVF (LSDV) 0.819703 Log-likelihood 1042.936 Akaike criteria -2037.872 Schwarz criteria -1945.282 Hannan-Quinn criteria -2001.018 rho 0.398579 Durbin-Watson 1.100795 Robust contrast for different intercepts in groups - Null hypothesis: Groups have a common intercept Contrast statistic: Welch F (18, 119.8) = 80.0174 with p-value = P (F(18, 119.8) > 80.0174) = 6.49e-058 Table 10: Econometric model of entropy index Once having gone through different models, we decide that the one shown in Table 10 is the best one, in terms of log-likelihood and error criteria. We observe that 81.9703% of the entropy indexes’ variability is explained by the variability of the regressors of this definitive model. This is the best adjustment out of the three indices’ models, and this happens being this the model with least variables out of everyone that has been done. We observe an increase in log-likelihood at the same time that errors decrease, giving a higher reliability to this last model, in the case where the entropy index is the dependent variable. We see that “R&D Expenditure”, “Density” and “Foreigners” are significant variables. We are not going to explain again what this means, as they maintain the sign seen before for the other indices. In this last analysis there still is presence of Kuznets effect. So, like in the previous cases we calculate the minimum threshold of GDP, to see the relation between these two variables in different countries. So, following the procedure previously done we see that the threshold has a value of 15.7201, being the smallest threshold out of the three. this means that countries with a GDP higher than 15,720.1 will have a negative relation between GDP and the segregation index. 39 If we take a look to which countries, from the studied ones, are these, we have all the previous ones plus Greece and Slovakia. In these countries and increase of the GDP in thousand per capita will cause a decrease in the value of the index, so the immigrant population will be more unevenly distributed. This may happen because in countries rich enough they need more individuals everywhere, so immigrants will be distributed in all territory. As we see through this part of the work, different macroeconomic variables affect to segregation indices. However, a variable being significant to explain certain index does not mean that it will be significant in every other index; we observe this with the variable “Education”. Each variable affects in different ways to indices. There are some variables we thought of that happen to never be significant. This is the case of “Fertility rate”, “New citizenship” and “Government expenditure”. The other variables are significant at least once, being remarkable the cases of “Density” and “Foreigners” that are always significant. Some of the variables that are significant have a positive effect in the index value; this is the case of “Elderly population”, “R&D Expenditure” and “Education”. An increase of these variables means an increase in the immigrant segregation, meaning that immigrant population will be geographically more unevenly distributed. We also have variables like “Unemployment rate”, “Density” and “Foreigners” that decrease the immigrant segregation. So, if these variables increase the immigrant population will be territorially distributed in a more homogeneous way. This does not mean that we have to discourage the education or the expenditure in R&D; it also does not mean that we have to adopt policies that increase unemployment. It is very important to remark the nonlinear relation of the GDP, as it shows that the effect of the GDP in the segregation index is what we were looking for, a Kuznets curve. We have seen that this Kuznets effect is present in the three models. We have also seen that this allows us to calculate where the minimum threshold is, in order to see in which countries the negative relation between GDP and the dependent variable is present. For example, we see that in Romania, where the GDP is below the threshold, if the GDP in thousand per capita, increases in one unit, the value of the index will increase, being translated in a 40 more uneven distribution of the immigrant population. However, in other countries such as Norway, where the GDP is above the threshold of the GDP, an increase of the GDP, in thousand per capita, in one unit means an estimated decrease of the segregation index; that is, the immigrant population will be more homogeneously distributed in the territory. 5.- Conclusions The gold of this paper is twofold. First, measuring the unevenness of the territorial distribution of immigrant population in Europe with regions’ aggregate data. It is also interesting to see how these indices evolved throughout the period studied, since this quantification enables the trend of the unevenness to be analysed. Second, studying the macroeconomic variables that affect the geographical unevenness of immigrants’ location. In particular, we find that the variable that most affect to immigrant territorial distribution is GDP. We analyse the relation between the GDP and the indices, focusing in trying to see if there was a Kuznets effect. Regarding with the unevenness quantification, we observe that, even if the values of the segregation indices used are different, the evolution is similar in the three cases. So, we see that the results are qualitatively similar even if they are quantitatively different. We also see in a clear way that segregation decreases over time, fulfilling the expectation. This may be because as time goes by immigrants arrive to regions where there are employment opportunities but where first wave immigrants did not arrive. We can say that at European level immigrant population is more evenly distributed in 2019 than in 1999. When the analysis is carried out at country level the conclusion is similar; unevenness in the geographical distribution of immigrant population has reduced. Regarding the search of determinants of the segregation indices used. We choose some variables that we consider representatives. In relation with the macroeconomic determinants of the geographical distribution of immigrants, we obtain some interesting results. First, we find that variables, like density, affect negatively to the indices. So that regions being denser in terms of populations will lead to a lower territorial dispersion of immigrant population. This relation may 41 reflex that when the density is higher the immigrant population has to distribute in order to fix in a better way. In the same way, others like “Elderly population” increase the value of the indices. This means that when the percentage of elderly population increases, immigrant population is more unevenly distributed. This might be caused because of the workplaces of those that just retire being more concentrated. Talking about the relation between these indices and the GDP, the relation is interesting, as the existence of the Kuznets effect in this case is proved. This is interesting for policymakers, since depending of a countries situation the priorities will change. The conclusion of the second part is especially relevant for policymakers, in order to adopt different policies to enjoy the benefits of immigration. This does not mean that it is a good idea to make drastic changes in those variables that increase the value of the indices; that is, that increase the uneven distribution. But it helps to think about other policies in order to minimize the negative effects some decisions can have in this aspect. For further research it could be interesting to use the decomposability property some indices have. This property allows us to analyse the inequality in the distribution of immigrant population in Europe, distinguishing between the inequality given between different countries and the inequality given within countries; this is, between regions inside a country. It would also be interesting to see to what extent the presence of far-right parties with some power affects to these indices, as they are winning power in many countries. For example, apart from Austria that was mentioned before, in Sweden there is Sverigedemokraterna; the Dansk Folkeparti in Denmark, the Schweizerische Volkspartei in Switzerland, the Alternative for Germany in Germany, or The United Kingdom Independence Party in United Kingdom. With so many cases it could be interesting to see how this affects to the homogeneity in the territorial distribution of immigrant population.