The Effects of Immigrant s Voting Rights: Evidence from a Natural Experiment. Simona Fiore

2016 The Effects of Immigrant s Voting Rights: Evidence from a Natural Experiment Simona Fiore

The effects of immigrant s voting rights: evidence from a natural experiment Simona Fiore Preliminary Draft June 20, 2016 Abstract This paper studies the political effect of immigration in the receiving country, by focusing on how political and economic outcomes change as a consequence of an increase in political power of ethnic minorities, given by the extension of voting rights to non citizens of non EU origin. A reform introduced in Belgium in 2004 is exploited as exogenous variation in the share of immigrant potential voters. In the spirit of a differencesin-differences strategy, election results of municipalities with more immigrants entitled to vote are compared with municipalities with a smaller share of immigrant voters (intensity of the treatment), before and after the reform. Significant effects on elections results are found in the region of Flanders and Wallonia. Interestingly, the effect goes in opposite direction: while in Flanders the pro-immigration parties are losing votes in the post reform elections in municipalities with a higher share of immigrants potential voters, the very pro-immigration parties are gaining votes in Wallonia. The magnitude of the estimated coefficients suggests that most of the effect is coming from a change of natives voting behavior. University of Bologna. Department of Economics. simona.fiore3@unibo.it. 1

1 Introduction During the last decades, immigration to European countries has increased exponentially, contributing to the creation of more ethnically heterogenous societies. Together with the increase in immigration, politics in Europe has taken a rightward turn, driven by the success of nationalist and anti immigrants parties. Among scholars, it has risen questions concerning the relationship between ethnic diversity and political outcomes. Understanding the nature and the effect of this relationship is important in light of the fact that the political effect of immigration can produce long lasting consequences in receiving countries through changes in preferences and policies. A growing literature is studying the effect of the mere presence of immigrants and the increasing ethnic diversity on policy outcomes in Europe, through the changes in native voters political attitudes and voting behavior. This is only one of the many channels through which immigrants can play an effect on receiving countries political life. The contribution of this paper is to show how political and economic outcomes change as a consequence of an increase in political power of ethnic minorities, given by the extension of voting rights to immigrants (non citizens) of non EU origin. Belgium has recently extended the right to vote in local elections to immigrants from outside the European Union with at least five years of legal residence in the country. The act providing for non citizen voting was adopted by the Belgian Parliament as the Law of 19 March 2004, and went into effect in 2006. This natural experiment is exploited here to study how the new group of enfranchised affects political outcomes at municipal level, by looking at electoral results for Belgian municipalities during the period 1988-2012. There are two channels through which the 2004 reform can influence Belgian local political outcomes: a direct channel, driven by creation of a new group of voters, i.e. non citizens of non Eu origin; and an indirect channel, driven by natives reaction to immigrants enfranchisement. If natives perceive immigration as an important issue, they can shift their vote to parties whose immigration policy increases their utility. Since the 70s, extending the right to vote to immigrants in local elections has been a sensitive issue of the political debate in Europe. The active political participation of noncitizens is widely recognized as a stimulating factor for immigrants integration in the host society. Supporters of the expansion of voting rights claim that all residents who pay taxes, contribute to the social security and take part at the community life should have the right to contribute to the selection of representatives. On the other hand, there is a common fear that the immigrants votes could tilt the political balance with potentially unfavorable outcomes for natives, e.g. with respect to the level of redistribution. In Belgium, Flemish politicians were the most resistant to the enfranchisement of immigrants, arguing that the introduction of a new group of voters would shatter the delicate balance of power between Dutch and French-speaking communities and possibly create a disproportionate benefit for Francophone political parties. Behind anecdotal evidence, there is very little systematic evidence on the socio-economic consequences of enfranchisement of immigrants. The aim of this paper 2

is to fill this gap by exploiting the 2004 Belgian law. Belgium is an interesting setting to study the impact of enfranchising non citizens of non Eu origin. First, Belgium has historically attracted significant immigration flows and the number of immigrants has steadily increased since the 80s. Second, the timing of the reform is peculiar: it passed at a time when the debate about immigration in Europe was not focused on enlarging immigrants rights. Third, according to the Belgian Constitution, local governments can take any initiatives that is beneficial to local interests and no other government has legal responsibility for the concerned field of action. Local governments have great autonomy and responsibility on a number of important issues, including expenditures on education, security and social policies, culture, urbanism and environment. Moreover, Belgium is a country of different linguistic and political realities. The three regions of the Flanders, Wallonia and Brussels differ in many dimensions, including political preferences and immigration policy (see Section 2.1). These differences offer the opportunity to explore heterogeneous effects of the 2004 Law and can provide key insights on the interpretation of the results. The empirical strategy adopted here exploits two sources of variation: time variation, given by the introduction of the reform and cross-sectional variation, given by the differences in the share of non citizens potential voters (intensity of the treatment) across municipalities. In the spirit of a difference-in-difference strategy, municipalities with different intensity of treatment are compared, before and after the reform. The estimation of a fully flexible model (see section 3.1 for more details) ensures the non violation of common trend assumptions; moreover, results of some placebo tests performed by using only pre reform elections suggest that the results can then be interpreted as causal. The reform has significant effects on elections results only in the region of Flanders and Wallonia. Interestingly, the effect goes in opposite direction: while in Flanders the pro-immigration parties are losing votes in the post reform elections in municipalities with a higher share of immigrants potential voters, the very pro-immigration parties are gaining votes in Wallonia. The magnitude of the estimated coefficients suggests that most of the effect is coming from a change of natives voting behavior. As more extensively discussed below, Flanders and Wallonia adopt very different immigration policies and citizens of the two regions have different political preferences (as shown by elections results) and different attitudes towards immigrants (as shown by the analysis of some questions related to immigration asked in the European Social Survey, see section 4.4). These heterogeneities can explain why the reactions of natives to the enfranchisement of immigrants goes in opposite directions in the two regions. Moreover, the decrease in votes for the pro-immigration parties in the region of Flanders does not seem to be driven by worries about the adverse effect of immigrants in the economy and the labor market: Belgians leaving in Flanders are significantly more likely than Belgians living in Wallonia to think that immigrants are good for the ecomomy and they can help to create jobs. This paper relates to two strands of literature. The first one takes a political economy approach and examines the effects of the extension of voting rights on the size of government 3

(Husted and Kenny 1997, Lott 1999, Cascio and Washington 2014). The main message from this literature is that the enfranchisement of minorities (e.g. black, poorer or women) determines an increase in welfare or government expenditures, mainly driven by the change of the median voter. The second strand of literature examines the relationship between the mere presence of immigrants (and immigration-driven increases in ethnic diversity) and natives voting behavior or political attitudes (Razin et al. 2002, Böheim and Mayr 2005, Dahlberg et al. 2012, Harmon 2013, Barone et al. 2014, Mayda et al. 2015). The negative relationship between ethnic diversity and left-wing political preferences, as well as the level of public spending at local level, is confirmed in most studies. The paper is structured as follows: in Section 2, the institutional setting of Belgium is discussed; Section 3 presents data and the empirical strategy; in Section 4 results of the empirical analysis are presented; Section 5 provides some robustness checks and Section 6 concludes. 2 Institutional setting of Belgium Specific features of the Belgian immigration policy and political system are particularly relevant for the empirical analysis. This section gives a brief overview of these two aspects of the Belgian context. 2.1 The Belgian immigration policy In Belgium, competencies over immigration issues are split between the federal state and the regions. The federal state is responsible for admissions, removals, residence rights and laws on citizenship of immigrants; while regional authorities have jurisdiction over the reception and integration of immigrants, by promoting their participation in the host society (as established by the law of 8 August 1980). The regions of Flanders and Wallonia adopted very different immigrants integration policies, inspired by diverse values and pursuing contrasting objectives. It results in immigrants having different rights and duties, according to their place of residence. Flanders policy highlights the importance of cultural diversity and ethnic identities and combines an assimilationist policy targeting new comers with a multicultural policy addressed to established ethic minority groups. A large network of implementing actors, coordinated by a centralized organization, administers language courses and a civic integration course (focusing on rules and regulations, common norms and values), as part of the compulsory integration trajectories. Importantly, non attendance of the courses is punished by the means of an administrative fine. The Francophone Belgium, instead, puts emphasis on economic and social inclusion and promotes it through color-blind policies, as a good strategy to promote assimilation, opposite to targeted or multiculturalist policies, responsible for stressing differences among natives and immigrants. The Walloon integration process is very decentralized, and most 4

of the initiatives promoting integration are taken at municipality level and implemented by local organizations. Immigrants are not obliged to attend integration courses and the the total budget devoted to integration policy is much lower in Wallonia with respect to Flanders (Adam and Jacobs 2014). The region of Brussels combines together Flemish and Francophone integration policies, led by local organizations, mainly in poor and deprived neighborhoods. Another relevant difference at regional level concerns parties position regarding immigration and immigrants rights. On the one hand, Francophone political parties are willing to promote naturalization as a natural mean for foreigners integration; on the other, Flemish parties claim the importance of integration before naturalization and ask for a stricter immigration policy. 2.2 The Belgian political system The Belgian electoral system is divided according to linguistic lines: Flemish parties compete for votes in the region of Flanders, while Francophone parties run in the region of Wallonia. Only in the region of Brussels there are both parties. During the 70s the major statewide parties split along regional lines, and all parties created after that are organized at community level and only represent the interest of part of the population. The Belgian political system can then be classified as a two party systems (Dandoy 2014). Immigration is a relevant issue in the electoral platforms of Belgian parties, and it has been highly politicized. The attitudes of local parties toward immigration is recovered by looking at the manifestos of their national counterparts; more specifically, by looking at the analysis conducted by Dandoy (2014), who studied the electoral platforms of the main Belgian national parties between 1977 and 2007. Belgian local politics is highly nationalized, with many national parties taking part at local elections. It happens more in Flanders than the Walloon and Brussels regions. As Figure 1 reveal, for the local elections in the period 1988-2012, in Flanders, more than 80% of the votes was gained by parties with a national counterparts (or parties that can be labeled as belonging to one major party family with a national counterpart); the share is much lower (about 40%) in the other two regions, where municipal elections still remain more localized, i.e. there are many local parties with no national counterparts running for elections. Many significant aspects come out from the analysis of national parties political manifestos and allow constructing a rank of parties over immigration attitudes. First, it emerges that, overall, Flemish parties allocate more attention to immigration than Francophone parties. Second, as expected, extreme right parties manifestos dedicate more space to immigration issues than other parties, with an exception. When looking at each single election in the period 1977-2007, it emerges that during the 90s the Flemish liberal party focused on the issue of immigration more than any other party. Third, by looking at a specific policy sector of immigration, namely immigration integration, it appears that extreme-right and liberal parties dedicate more attention to the issue with respect to socialist and Christian Democrat 5

parties. Last, when looking at the specific issues of the migration debate a clear pattern emerges in both regions: while Christian Democrat and socialist parties give more space to the discussion of democracy and rights, including the debate on voting rights; extreme-right and liberal parties focus on law and order, with special emphasis on the relation between immigrants and crime and the implications for public order. Given these elements of the parties manifestos, main parties participating at local election are ranked according to four categories from very-pro to very-anti immigration parties. In Table 1, the list of parties according to their attitude toward immigration is reported. 3 Data and key variables Data come from multiple sources. As for the outcomes variables, the focus is on both election results and expenditures at municipal level. Data on five election rounds (1988,1994 2000 before the reform and 2006, 2012 after the reform) have been collected from the web, using Python 1. The regional authorities of Flanders provided data on municipal expenditures and revenues at local level (for the period 2003-2013). The Ministry of Interior released data on potential and registered non Eu voters for local elections in 2006 and 2012. Data on the characteristics of the Belgian municipalities over time are (mainly) available online through the Statistics Belgium 2. The variable of interest in order to study the effect of enfranchising non citizens of non Eu origin is defined as follows: pot_voters = NC_M V (1) where NC_M is the number of potential voters of non Eu origin who have no Belgian citizenship and V is the remaining population of potential voters (i.e. Belgians and other Eu citizens). Table 2 shows the share of potential non Eu non citizens voters measured in 2006 and 2012 elections. 3 Interestingly, this share has increased over time, above all in Flanders. Moreover, it is important to notice that potential non Eu voters do not spread equally across regions: the municipalities of the Brussels region have a much higher share, while Flanders and Wallonia have a smaller and similar share of non Eu potential voters. The final sample used for the analysis consists of 580 municipalities, observed in five points in time, corresponding to each municipal election during the period 1988-2012. 4 In Table 3, summary statistics on data across the five rounds of elections are reported. In the region of Flanders, the pro-immigration party is the one that gained the majority of votes, 1 Sources: all regions 1988-2000; Region of Wallonia 2000-2006, 2012; Region of Flanders 2006 2012; Region of Bruxelles 2006-2012. 2 http://statbel.fgov.be/. 3 The variable as measured in 2006 is considered in the empirical analysis, however results are similar when the 2012 share is used. 4 The nine municipalities of the German-speaking community are not considered in the analysis. 6

however also the anti and very-anti immigration parties have a considerable share of votes. The very-pro immigration parties have the majority of votes in Wallonia where, differently from Flanders, the anti-immigration parties only gain a small share of votes. The two regions have comparable and very high citizens turnout, while the registration rate among non Eu 5 is much lower in Flanders than in Wallonia, about 12% and 25% respectively. In the region of Brussels, the votes across parties are distributed more evenly, with the anti-immigration party having a slightly higher share of votes with respect to the very pro-immigration parties. The turnout among Belgians is a bot lower if compared to the other two regions and the registration rate among non Eu is also low. The winning margin, i.e. the difference in the share of votes between the first and the second party, is pretty high in the three regions; close elections are not common at local level in Belgium. Table 4 report statistics on demographic characteristics of Belgian municipalities. On average, municipalities in Flanders and Wallonia have less then twenty thousand inhabitants, while municipalities in Brussels region are bigger. Unemployment rate is especially high in Wallonia, where also the share of foreigners is higher with respect to Flanders. 3.1 Empirical strategy The empirical strategy exploits two sources of variation: the first is time variation coming from the introduction of the reform; the second source of variation is cross-sectional and arises from the differences in municipalities share of non Eu immigrants allowed to vote. In the spirit of a differences-in-differences strategy, election results of municipalities with more immigrants entitled to vote are compared with municipalities with a smaller share of immigrant voters (intensity of the treatment), before and after the reform. The introduction of municipality and time-period fixed effects controls for all timeinvariant differences across municipalities and secular changes over time. The strategy relies on the absence of any other shocks occurred around the same time the reform was introduced and correlated with the share of potential immigrant voters. The latter identification concern is addressed by controlling for time and municipality-varying factors that may bias the estimates, such as population density, population size (in logarithmic form), the share of foreign population and the unemployment rate. In addition, controls for turnout, a dummy for winning incumbent and a dummy for being incumbent are added. The analysis will start with a focus on the parameter δ in equation (2) below: y mt = α + η m + γ t + δ pot_voters m post t + ε mt (2) where y mt is an electoral (or economic) outcomes, η m is a municipality fixed effect, γ t an election (or year, depending on having a political or economic outcome) fixed effect, pot_voters m 5 The registration rate among non Eu can be considered a good proxy for turnout: non citizens who want to vote are obliged to register prior to the election and, once registered, if they do not show up at the pools they have to pay an administrative fine. 7

is the fraction of non Eu non citizens allowed to vote, as it is measured in 2006 (the variable that captures the treatment intensity). The availability of two or more pre- and post-treatment periods allows to estimate a flexible model that includes leads and lags of the treatment. This allows to assess the presence of anticipatory effects or other violations of the common trend assumption. Formally, the model in equation (3) below is estimated: y mt = α + η m + γ t + T δ j pot_voters m It j + βx mt + ε mt (3) j=t+1 where everything is defined as above, with the exception that the effect of the treatment is identified in each election going from t+1 to T (with t being the reference category) and a set of time-varying municipality controls (as listed above) is added. Equation (3) imposes no parametric assumptions on the pre-treatment dynamics and allows for a the test of the null hypothesis of no common pre-treatment trends (H 0 : δ j = 0 for all pre-treatment periods). Moreover, it also allows the implementation of tests on the dynamics of the treatment effect, i.e, it is possible to test whether the effect is constant in the post-treatment period. 4 Results This section presents the results of the effect of the reform on political and economic outcomes, by region. For what concerns political outcomes, first, all labeled parties are considered, then the analysis will focus on outcomes for which the reform was effective. 4.1 Flanders Table 5 show results on the share of votes taken by each family of parties in Flanders. The reform significantly affects only the pro-immigrant parties in the region, which are losing power in the post reform period. The magnitude of the coefficient indicates that an increase of 1 p.p. (percentage point) in the treatment determines a decrease of more than 2 p.p. in the share of votes to the pro-immigrant party. There are no significant results on the outcomes of the other parties. From now on, the focus will be on the share of votes to the pro-immigrant parties. In order to give a causal interpretation to the effect of the reform, municipalities with different intensity of the treatment must have similar pre reform trend in the outcome variable. This hypothesis can be tested by estimating a fully flexible model, as expressed in equation (3). Results of the flexible estimates for Flanders are reported in Table 6. In all columns, the coefficients associated to the pre-treatment periods are small and non significantly different from zero; while there is a negative and increasing (in absolute value) effect in the post treatment election rounds. These results are also robust to the inclusion of district specific time trend. Figure 2 plots the coefficients of the interactions between the intensity of the treatment 8

and each election s dummies, as reported in column (5). There is a zero and flat effect before the 2004 reform and a negative and significant effect starting from 2006 election. Figure 3 plots the coefficients of a fully flexible estimates with all controls where the dependent variable is the probability of having a very anti-immigration mayor. Municipalities with a higher share of immigrant potential voters have a higher probability of having an anti-immigration mayor in the post reform elections. In Table 7 equation (2) is estimated by using a restricted sample, accounting only for 1988 to 2000 elections. In this subsample a placebo effect of the reform is estimated, by assuming that the reform took place in 1994 or 2000. Given that the reform was effective only for the 2006 election onwards, finding significant effects in 1994 or 2000 elections would suggest violations of the identifying assumptions, since they could not be attributed to the real reform. Conversely, finding no effects on this subsample suggest that the baseline estimates can be interpreted as causal. In column (1) of Table 7 the sample is restricted to three rounds of election, and the treatment is imposed in 1994 and 2000. In column (2), 2000 is used as a fake post treatment period. In column (3) only elections in 1994 and 2000 are considered and 2000 is the post reform period. The coefficients of the interaction between the post dummy and the treatment intensity are always much smaller than in Table 5 and they are never significantly different from zero. The placebo experiments suggest no evidence of a differential relationship between share of votes to the pro-immigration parties and the treatment intensity in the pre reform elections. If voting behavior reflects voters demand for public goods, changes in municipalities expenditures and revenues should be expected as a consequence of the changes in parties share of votes. In the case of Flanders, given that pro-immigrantion parties also favor more redistribution, less public expenditures and revenues are expected as a result of the reducing power of the pro-immigration parties. Results in Table 8 confirm this expectations and show that municipalities with a higher treatment experience a higher decrease in total per capita expenditures and per capita revenues in the post reform period. 4.2 Wallonia When looking at the results for the region of Wallonia, an opposite effect of the reform with respect to Flanders emerges: the very pro-immigrantion parties gain votes in municipalities where the share of potential non Eu voters is higher. More precisely, as shown in Table 9, an increase of 1 p.p. in the treatment determines an increase of about 3 p.p in the share of votes to the very pro-immigrant parties. In order to interpret causally the previous result, there should be no violation of the common trend assumption. The fully flexible estimates, reported in Table 10, point to similar pre reform trends in the outcome variable: the interacted coefficients are never significantly different from zero in the pre-reform elections. Figure 4 reports the interacted coefficients from column (5) in table 10, for a better visual inspection of the effect. The placebo experiments for Wallonia are reported in Table 11. As for Flanders, only a 9

subsample of elections is considered and the treatment is imposed in the pre reform period. The coefficients of the interaction between the dummy post and the treatment intensity are never significantly different from zero, pointing to a non differential relationship between share of votes to the very pro-immigrants party and the treatment intensity in the pre reform elections. 4.3 Brussels Table 12 reports results of the estimation of equation (2) for the 19 municipalities of the Brussels region. The enfranchisement of non citizens of non Eu origin has a positive effect on the share of votes of the pro immigration parties, that gains a bit less than 1 p.p. of votes as the treatment intensity increases by 1 p.p. However, a deeper analysis of this effect, as shown by the fully flexible estimates in Table 13 points to no evidence of a casual effect of the reform. Specifically, when controlling for municipalities time varying characteristics, post reform coefficients are not statistically significant. 4.4 European Social Survey The European Social Survey is a cross national survey, that collects information on attitudes, beliefs and behavior, through face-to-face interviews across European countries. The 2002 and 2004 waves are used here to compare the attitudes towards immigrants as well as the fears about immigration across the three Belgian regions. 6 Only respondents who declare to be born in Belgium are considered in the analysis. Each pair of columns in the following tables report, respectively, the average response by region and the coefficient for the difference across regions. The results in Table 14 show that Belgians living in Flanders are, on average, more likely to say that immigrants are good for the economy of the country and that they contribute to the creation of new jobs more than to take jobs away. The better economic conditions of Flanders with respect to Wallonia can explain why in Flanders citizens are less worried about the economic impact of immigrants. Interestingly, however, as shown in Table 15 Belgians leaving in the Walloon region are more open to let in more immigrants from poor countries and believe that foreigners can enriched the country cultural life more than do Belgians in Flanders. Moreover, in Wallonia, Belgians are significantly more likely to have an immigrant friend or a colleague with respect to Flanders. The differences highlighted above suggest that the lost of political power of pro-immigration parties in Flanders is not driven by citizens being worried about the adverse effects of immigrants in the economy and the labor market. Rather, Flanders citizens having less interactions with immigrants and being less prone to have more immigrants can be thought as a result of the different policies pursued by each region, which determine citizens preferences 6 Only the waves before the reform are used: Belgians beliefs collected after can be biased by the reform itself. 10

and attitudes toward immigrants. 5 Robustness checks 5.1 Internal Migration The reform can potentially change the patterns of migration within Belgium. If it happens, the changes in the elections results might be due to changes in the composition of voters, rather than to the enfranchisement of a new ethnic group. In this section, data on internal migration over 25 years in Belgium are exploited in order to show that these movements do not represent a threat to the validity of previous results. Figure 5 plots the yearly average of people entering or leaving a municipality, expressed as a share of the total population, separately for the three regions over a 25 years long period. In Wallonia, internal movements are more sizable than in the Flanders and in both regions the share of people entering is bigger than the share of people moving out. The opposite is true for the Brussels region, that experiences more out migration towards the other two regions of the country. Moreover, there is a slightly increasing trend in internal movement over time but, on average, there is not a change in the movement coinciding with the introduction of the reform in 2004. In order to exclude a systematic relation between Belgian internal migration and the share of non citizens of non Eu origin, the share of people entering and exiting a municipality is used as outcome variable in an equation as in equation (3). The dummy for post takes here value 1 starting from 2004, the year of the reform. Results in Table 16 suggest that internal movements are not changing differently pre and post reform in municipalities with different share of non Eu non citizens allowed to vote. This is true for both, the share of people entering and exiting municipalities in the region of the Flanders and Wallonia. These results ensure that if any changes in the composition of the voters happen after the reform, they are not affected by the intensity of the treatment exploited in this paper. Previous results are the not biased by internal migration movements. 5.2 The Quick Citizenship Act In March 2000, the so called Quick Citizenship Act came into effect in Belgium. According to this law, foreign nationals legally resident in Belgium for three years and refugees resident for two years can submit an application for naturalization. As a result of this law, there was a huge increase in naturalization rate in Belgium and, since voting is mandatory for Belgians, the new citizens eligible as voters (i.e. citizens aged at least 18 years, people who have not been excluded from the right to vote or whose right to vote has been suspended, people who have never been given a criminal sentence) could take part at the elections. This section is devoted to show that the naturalization law does not drive or bias previous results. The focus in this section will be on election outcomes in Flanders and Wallonia for 11

which there is a significant effect of the 2004 reform. The exogenous change in naturalization driven by the 2000 law is computed as follows: nat_change = nat 2000 nat 1999 nat 1999 100 (4) Table 17 displays the naturalization growth rate for the whole country and the three regions separately: on average, between 1999 and 2000 naturalizations growth rate is more than 100%. To start with, the effect of the naturalization law is analyzed, by estimating an equation as equation (3), where the intensity of the treatment is given by the variable nat_change and the elections post reform are 2000, 2006 and 2012. Table 18 show the effect of the naturalization law in Flanders: from column (1) to (3) the outcome variable is the share of votes taken by the pro-immigrants party; in column (1) the naturalization growth rate is interacted with a dummy post reform, while in column (2) and (3) the treatment variable is interacted with a dummy for each election year. In the last column only election from 1988 to 2000 are considered. In columns (4) to (6) the same analysis is performed by using as outcome variable the probability of having a mayor from very anti immigration parties. In Table 19 the same results are shown for the region of Wallonia, where the outcome variable is the share of votes taken by the very pro immigrantion parties. The effect of the 2000 reform, as it is captured by the naturalization growth rate, is very small and never statistically significant from zero. It suggests that municipalities experiencing a higher naturalization growth rate do not experience post-2000 changes in elections outcomes different from municipalities with a lower naturalization growth rate. In Table 20 and 21 the effects of the two reforms are analyzed together, respectively for Flanders and Wallonia: while the effect of the naturalization law are always small and statistically not significant, the effect of enfranchising non Eu immigrants remains statistically significant. 6 Concluding Remarks This paper studies the effect of granting non citizens of non Eu origin with the right to vote in local elections on political and economic outcomes. The introduction of a reform in Belgium in 2004 is exploited as an exogenous variation. According to the refrom s rules, non citizens of non Eu orign legally residing in the country for at least five years can vote at local elections. It generates variation acorss municipalities in the share of non Eu immigrants who are potential voters. This cross sectional variation, interacted with the time variation, coming from the introduction of the reform, is exploited as an exogenous shock in a difference in difference framework. Results point to an opposite effect of the reform in the regions of Flanders and Wallonia: while in the former the main effect is a decrease in the votes for the pro-immigrants parties 12

and, consequently, a reduction in welfare expenditures and total revenues; in the latter the reform determines an increase of votes to the left and pro immigrants parties 7. The magnitude of the effects, if compared with the small group of new voters, suggests that the law generates a reaction in natives voting behavior. Results can be interpreted as causal, given that there is no evidence of a violation of the common trends assumption and that plecebo experiments show no significant diferences across municipalities with different treatment intensity in the pre reform elections. As discussed in section 2.1, Flanders and Wallonia pursue very different immigration policies and have an opposite attitude toward immigration. Moreover, the two regions are characterized by a very different economic and demographic context (Dandoy 2014): compared to Wallonia, Flanders are a more prosperous economy, with no need of immigrant workers. These differences might explain the results: the rich Flanders have stronger (negative) ethnic preferences and care less about public good, then natives vote less for the pro immigrants and pro public good parties. This is also consistent with the fact that the lost of political power of pro-immigration parties in Flanders is not driven by citizens being worried about the adverse effects of immigrants in the economy and the labor market, as shown in section 4.4. Most of the existing literature studying the political effect of migration in Europe focuses on the impact of the mere presence of immigrants on natives local behavior. The main contribution of this paper is to show that a big natives changes in voting behavior comes from the enfranchisement of an ethnically different group. The opposite results on the two regions suggest that enfranchisement per se has not necessarily positive effects for immigrants: if they are a small and non-integrated group, as in the case of Flanders, the reaction of natives, who support less the pro immigrants parties, result in a decrease in the demand public goods, that mostly hurts the low income population, including immigrants. 7 Results in the region of Brussels do not allow for a causal interpretation of the reform effect. 13

References Adam, I. and D. Jacobs (2014). Divided on immigration, two models for integration. the multilevel governance of immigration and integration in belgium. In The Politics of Immigration in Multi-Level States, pp. 65 85. Springer. Barone, G., G. De Blasio, and P. Naticchioni (2014). Mr. rossi, mr. hu and politics: The role of immigration in shaping natives political preferences. Böheim, R. and K. Mayr (2005). Immigration and public spending. IZA Discussion Paper. Cascio, E. U. and E. Washington (2014). Valuing the vote: The redistribution of voting rights and state funds following the voting rights act of 1965*. Quarterly Journal of Economics 129(1). Dahlberg, M., K. Edmark, and H. Lundqvist (2012). Ethnic diversity and preferences for redistribution. Journal of Political Economy 120(1), 41 76. Dandoy, R. (2014). Regionalist parties and immigration in belgium. In The Politics of Immigration in Multi-Level States, pp. 200 222. Springer. Harmon, N. A. (2013). Immigration ethnic diversity and political outcomes. Husted, T. A. and L. W. Kenny (1997). The effect of the expansion of the voting franchise on the size of government. Journal of Political Economy, 54 82. Lott, J. R. (1999). How dramatically did women s suffrage change the size and scope of government? Journal of Political Economy 107(6 Part 1), 1163 1198. Mayda, A. M., G. Peri, and W. Steingress (2015). Immigration to the us: A problem for the republicans or the democrats? Razin, A., E. Sadka, and P. Swagel (2002). Tax burden and migration: a political economy theory and evidence. Journal of Public Economics 85(2), 167 190. 14

7 Figures and Tables Figure 1: Share of local parties labelled according to their national counterparts 0 20 40 60 80 1988 1994 2000 2006 2012 1988 1994 2000 2006 2012 1988 1994 2000 2006 2012 Flanders Brussels Wallonia Figure 2: Interacted coefficients from fully flexible estimates: share pro-immigration parties - Flanders -10-5 0 5 1994 2000 2004 2006 2012 15

Figure 3: Interacted coefficients from fully flexible estimates: very anti-immigration mayor - Flanders -.1 0.1.2.3.4 1994 2000 2004 2006 2012 Figure 4: Interacted coefficients from fully flexible estimates: share very pro-immigration parties - Wallonia -5 0 5 10 15 1994 2000 2004 2006 2012 16

Figure 5: Internal Migration 2 4 6 8 10 1990 1995 2000 2005 2010 2015 Exit - Brussels Entry - Wallonia Entry - Flanders Entry - Brussels Exit - Wallonia Exit - Flanders Table 1: List of parties according to their attitudes toward immigration Very pro-imm Committee for Another Policy (CAP) Communist Party (PCB) Egalité Gauche Parti Socialiste (PS) Socialist Party Differently (sp.a.) Workers Party of Belgium (PVDA, PTB) Anti-imm Centre démocrate humaniste (CDH) Christelijke Volkspartij (CVP) Christian Democratic and Flemish (CDV) Christian Social Party (PSC) Pro-imm Liberal Reformist Party (PRL) Lijst Dedecker Mouvement Rï 1 2formateur (MR) Open Flemish Liberals and Democrats (VLD) Parti des Pensionnes Vivant Very anti-imm Agir Front National (FN) New Flemish Alliance (N-VA) Rassemblement Wallon (RW) Vlaams Belang (VB) Table 2: Potential non citizens non Eu voters as a percentage of all other voters 2006 2012 Difference Belgium 0.6% 0.79% 32% Flanders 0.41% 0.62% 51% Wallonia 0.46% 0.59% 28% Brussels 5.42% 6.06% 11% Note: Source: data provided by the Belgian Ministry of Interior on potential non Eu non citizens voters during 2006 and 2012 elections. 17

Table 3: Summary statistics: Election data Mean Std. Dev. Min. Max. Obs Flanders Very pro-imm 13.2 11.6 0 60.2 1540 Pro-imm 32 17.2 0 78.5 1540 Anti-imm 15.4 13.5 0 57.9 1540 Very anti-imm 12.6 12.1 0 68.8 1540 Belgians turnout 94.4 3.3 0 100 1540 Non eu registration rate 11.7 11.9 0 100 616 Winning incumbent 0.8 0.4 0 1 1540 Winning share 16.2 13.3 0 81.8 1540 Wallonia Very pro-imm 23.6 23.1 0 100 1263 Pro-imm 7.1 13.2 0 69 1263 Anti-imm 9.7 14.7 0 68.8 1263 Very anti-imm 0.7 3.8 0 55.8 1263 Belgians turnout 93.1 2.8 79.2 100 1263 Non eu registration rate 24.5 19 0 100 500 Winning incumbent 0.8 0.4 0 1 1262 Winning share 21.5 15.7 0 82 1263 Brussels Very pro-imm 13.6 12 0 54.4 95 Pro-imm 6.6 6.5 0 23.5 95 Anti-imm 13.9 13 0 54.2 95 Very anti-imm 12.7 11.6 0 52.1 95 Belgians turnout 85.3 2.5 77.8 89.9 95 Non eu registration rate 16.1 3.7 7 23.8 38 Winning incumbent 0.7 0.4 0 1 95 Winning share 19.5 12.6 1.4 50 95 Note: The share of votes for all parties is computed as the average of five rounds of local elections. Belgians turnout is computed for all election, except 2000 elections. Non Eu registration rate is computed for 2006 and 2012 elections. 18

Table 4: Summary statistics: Demographic Data Mean Std. Dev. Min. Max. Obs Flanders Population 19545.6 31913.2 85 507911 1536 Density 5151.7 4484.8 491 33683.7 1536 Log of Population 9.5 0.8 4.4 13.1 1536 Share of foreigners 3.9 5.2 0 47.7 1536 in the population Unemployment rate 2.6 2.2 0 6.2 1536 Wallonia Population 13096.1 21090.9 1358 208021 1262 Density 3083.3 4390.1 139.3 35137 1262 Log of Population 9 0.9 7.2 12.2 1262 Share of foreigners 6.2 5.3 0.8 37.7 1262 in the population Unemployment rate 6 5.3 0 14.4 1262 Brussels Population 53393 34522.5 15974 168576 95 Density 90815.8 52193.4 18650.7 238178 95 Log of Population 10.7 0.6 9.7 12 95 Share of foreigners 26.8 10.7 9.3 57.1 95 in the population Unemployment rate 10 8.2 0 17.6 95 Note: Unemployment rate is measured at provincial level for 2000, 2006 and 2012 elections. Table 5: Average Pre/Post effect of immigrants enfranchisement: Flanders (1) (2) (3) (4) Veri pro-imm Pro-imm Anti-imm Very anti-imm Pot_voters*Post 0.067-2.039** 0.577 0.050 (1.322) (0.859) (1.151) (0.989) Observations 1,540 1,540 1,540 1,540 R-squared 0.777 0.770 0.689 0.674 Municipalities indicators YES YES YES YES Years of elections indicators YES YES YES YES District Time Trend NO NO NO NO Note. The dependent variable is the share of votes taken by each party. Standard errors clustered at the municipality level are in parentheses. Significance at the 10 % level is represented by *, at the 5% level by **, and at the 1% level by ***. 19

Table 6: Fully Flexible Estimates: Flanders (1) (2) (3) (4) (5) Pro-imm Pro-imm Pro-imm Pro-imm Pro-imm Pot_voters*1994-0.986-1.039-0.928-0.928-0.661 (1.477) (1.501) (2.035) (2.036) (1.941) Pot_voters*2000-1.341-1.435-1.474-1.478-0.701 (1.328) (1.491) (1.754) (1.755) (1.694) Pot_voters*2006-2.790* -3.163* -4.398** -4.401** -3.609* (1.662) (1.882) (2.123) (2.119) (2.040) Pot_voters*2012-4.345** -5.099** -5.945** -5.947** -4.959** (2.118) (2.472) (2.574) (2.570) (2.436) Log of Population -3.311-1.998-1.983-2.462 (19.178) (18.995) (19.034) (17.982) Population Density 0.001 0.001 0.001 0.001 (0.001) (0.001) (0.001) (0.001) Share of Foreigners -0.439-0.439-0.345 (0.378) (0.380) (0.349) Share of Foreigners*1994 0.001 0.001-0.026 (0.260) (0.260) (0.246) Share of Foreigners*2000-0.039-0.038-0.079 (0.239) (0.242) (0.231) Share of Foreigners*2006 0.213 0.214 0.177 (0.256) (0.256) (0.260) Share of Foreigners*2012 0.151 0.151 0.094 (0.272) (0.272) (0.250) Unemployment Rate -0.024 0.136 (0.870) (0.850) Being Incumbent 3.632*** (1.046) Winning Incumbent 2.074*** (0.772) Turnout -0.018 (0.092) Observations 1,540 1,540 1,540 1,540 1,540 R-squared 0.804 0.804 0.805 0.805 0.809 Municipalities indicators YES YES YES YES YES Years of elections indicators YES YES YES YES YES District Time Trend YES YES YES YES YES Note. The dependent variable is the share of votes taken by the pro-immigration parties. Standard errors clustered at the municipality level are in parentheses. Significance at the 10 % level is represented by *, at the 5% level by **, and at the 1% level by ***. 20

Table 7: Placebo Experiments: Flanders (1) (2) (3) Pre=1988 Pre=1988,1994 Pre=1994 Post=1994,2000 Post=2000 Post=2000 Pot_voters*Post 0.709-0.052 0.474 (2.329) (2.033) (3.294) Log of Population -19.934-21.784 6.270 (31.990) (30.894) (83.466) Population Density 0.005* 0.006* 0.006 (0.003) (0.003) (0.009) Share of Foreigners -0.081-0.115-0.274 (0.790) (0.837) (1.829) Share of Foreigners*1994-0.157-0.114 (0.324) (0.273) Share of Foreigners*2000-0.253-0.212-0.174 (0.354) (0.338) (0.418) Unemployment Rate 0.420 0.432 3.948 (1.684) (1.695) (4.612) Being Incumbent -1.006-1.013-1.094 (1.957) (1.967) (3.798) Winning Incumbent 2.743** 2.707** 3.435 (1.357) (1.352) (2.326) Turnout 1.102* 1.070* 0.314 (0.656) (0.636) (2.208) Observations 924 924 616 R-squared 0.877 0.877 0.937 Municipalities indicators YES YES YES Years of elections indicators YES YES YES District Time Trend YES YES YES Note. The dependent variable is the share of votes taken by the pro-immigration parties. Standard errors clustered at the municipality level are in parentheses. Significance at the 10 % level is represented by *, at the 5% level by **, and at the 1% level by ***. 21

Table 8: Expenditures and revenues: Flanders (1) (2) (3) (4) Per capita Per capita Log-Expenditures Log-Revenues Treatment Post -0.032** -0.030** (0.013) (0.015) Treatment I round after the law -0.031** -0.028* (0.013) (0.015) Treatment II round after the law -0.011 0.001 (0.022) (0.028) Observations 3,076 3,076 3,076 3,076 R-squared 0.773 0.773 0.674 0.675 Municipalities indicators YES YES YES YES Years of elections indicators YES YES YES YES District Time Trend YES YES YES YES Share of foreigners year YES YES YES YES District Time Trend YES YES YES YES Note. The dependent variable is the log of per capita expenditures and revenues from 2003 to 2013. Standard errors clustered at the municipality level are in parentheses. Significance at the 10 % level is represented by *, at the 5% level by **, and at the 1% level by ***. Table 9: Average Pre/Post effect of immigrants enfranchisement: Wallonia (1) (2) (3) (4) Veri pro-imm Pro-imm Anti-imm Very anti-imm Pot_voters*Post 3.315*** -0.274 0.346 0.178 (1.014) (0.540) (0.872) (0.384) Observations 1,263 1,263 1,263 1,263 R-squared 0.850 0.709 0.714 0.256 Municipalities indicators YES YES YES YES Years of elections indicators YES YES YES YES District Time Trend NO NO NO NO Note The dependent variable is the share of votes taken by each party. Standard errors clustered at the municipality level are in parentheses. Significance at the 10 % level is represented by *, at the 5% level by **, and at the 1% level by ***. 22

Table 10: Fully Flexible Estimates: Wallonia (1) (2) (3) (4) (5) Very Very Very Very Very pro-imm pro-imm pro-imm pro-imm pro-imm Pot_voters*1994-0.391-0.329-0.084-0.107-0.207 (1.070) (1.063) (1.835) (1.836) (1.836) Pot_voters*2000 1.890 2.139 2.088 2.134 1.810 (1.368) (1.430) (2.414) (2.425) (2.255) Pot_voters*2006 3.099* 3.315* 6.400** 6.407** 5.986** (1.817) (1.846) (2.604) (2.621) (2.451) Pot_voters*2012 5.116** 5.170** 6.618** 6.630** 5.803** (2.102) (2.127) (2.951) (2.960) (2.812) Log of Population 1.325-2.801-2.964-3.978 (16.989) (17.251) (17.230) (16.958) Population Density 0.001 0.001 0.001 0.001 (0.003) (0.003) (0.003) (0.003) Share of Foreigners 0.122 0.114 0.117 (0.485) (0.486) (0.464) Share of Foreigners*1994-0.072-0.062-0.095 (0.230) (0.231) (0.233) Share of Foreigners*2000-0.076-0.100-0.106 (0.313) (0.312) (0.294) Share of Foreigners*2006-0.832** -0.850** -0.894** (0.417) (0.417) (0.420) Share of Foreigners*2012-0.558-0.565-0.535 (0.474) (0.477) (0.440) Unemployment Rate 0.306 0.271 (0.314) (0.321) Being Incumbent 5.916*** (2.065) Winning Incumbent -0.818 (1.103) Turnout -0.068 (0.317) Observations 1,263 1,263 1,263 1,263 1,263 R-squared 0.881 0.881 0.883 0.883 0.887 Municipalities indicators YES YES YES YES YES Years of elections indicators YES YES YES YES YES District Time Trend YES YES YES YES YES Note. The dependent variable is the share of votes taken by the very pro-immigration parties. Standard errors clustered at the municipality level are in parentheses. Significance at the 10 % level is represented by *, at the 5% level by **, and at the 1% level by ***. 23

Table 11: Placebo Experiments: Wallonia (1) (2) (3) Pre=1988 Pre=1988,1994 Pre=1994 Post=1994,2000 Post=2000 Post=2000 Pot_voters*Post 0.182 2.962 4.003 (2.247) (1.900) (3.738) Log of Population 6.522 7.532 8.325 (22.895) (23.089) (57.629) Population Density -0.001-0.001 0.001 (0.005) (0.004) (0.010) Share of Foreigners 0.320 0.727 1.312 (0.623) (0.679) (1.812) Share of Foreigners*1994-0.067 0.009 (0.317) (0.233) Share of Foreigners*2000 0.060-0.028-0.154 (0.464) (0.399) (0.420) Unemployment Rate 0.617 0.607 0.477 (0.584) (0.589) (1.293) Being Incumbent 5.261** 5.254** 3.602 (2.451) (2.471) (5.646) Winning Incumbent -0.603-0.454 4.395 (1.814) (1.812) (3.879) Turnout 0.128 0.154 2.812* (0.501) (0.521) (1.559) Observations 756 756 504 R-squared 0.939 0.939 0.970 Municipalities indicators YES YES YES Years of elections indicators YES YES YES District Time Trend YES YES YES Controls YES YES YES Note. The dependent variable is the share of votes taken by the very pro-immigration parties. Standard errors clustered at the municipality level are in parentheses. Significance at the 10 % level is represented by *, at the 5% level by **, and at the 1% level by ***. 24