Is Europe’s right wing turn due to immigration? Empirical evidence from Italy

Abstract Immigration has unquestionably gained importance in Italian public opinion over the past few years; but what has been its effect on politics? Building on a recent trend in literature, this paper appraises the effect of immigration on the success of far-right parties. We analyze the results of the elections from 2001, when immigration was still marginal, to the last elections held in 2018. Applying time-series-cross-section methods to an original dataset, we find that immigration has had a consistent effect on votes for the far-right. Interestingly, we also show that economic factors likely weigh more on far-right votes than immigration.


Is Europe's right wing turn due to immigration? Empirical evidence from Italy
Over the last few years, immigration has gained significant relevance in the European Union. Political instability in North Africa (with the Arab Spring) and in the Middle-East (with the ongoing war in Syria), along with enduring economic disparities between North and South, have spurred large numbers of people on the roads towards the Old Continent. The repeated tragedies in the Mediterranean, the lack of a collective response to the phenomenon on the part of EU member states and the xenophobic reactions in many of the destination countries have contributed to putting the "issue" of migration at the forefront of political concerns (Ortensi et al. 2017;Van Wolleghem 2018). The various elections throughout the EU in 2017 and 2018 have shown that a good deal of the political competition was played out on migration-related issues, with far-right parties scoring very well, even in countries which traditionally feature a certain openness to diversity (Ortensi et al. 2017). Nonetheless, how much of the success of far-right parties is actually due to immigration remains a thorny question; a question which this paper proposes to address. Considering Italy as a case study, it seeks to appraise the effect of immigration -understood as the actual settlement of third country nationals in Italian territory -on people's preference for far-right parties.
Whereas the study on the determinants of people's attitudes towards foreigners are plentiful (see inter alia Dustmann and Preston 2001;Mayda 2006;O'Rourke and Sinnott 2006;Card, Dustmann and Preston 2012), only recently has the focus turned to the effect of immigration on far-right parties' electoral outcomes (Dustmann, Vsiljeva and Piil Damm 2016;Becker and Fetzer 2016;Davis and Deole 2017). Most of these studies regard countries with a long-standing migration history. Conversely, little attention has been paid to countries that have become destination countries more recently. Yet, it is in these countries that the effect of sudden demographic changes is likely to produce the most dramatic political changes . Likewise, it is also in these countries that the effect is likely to be better appraised as the number of citizens with an immigration background 1 is likely smaller than in old destination countries. Building on the growing body of evidence on the topic, this paper proposes to investigate the impact of migration on election outcomes. More precisely, we look at the effect of the settlement of foreigners in Italian regions on the success of far right parties over the last 18 years, thus including the elections of March 2018, in which the far-right fared particularly well (see Pritoni and Vignati 2018, for an overview of the results). We find that immigration has a significant effect on the success of far right parties over the period considered, thus confirming the findings already advanced in literature. Such an effect, however, proves somehow less relevant when compared to the effect of unemployment. Empirical evidence is drawn from the application of time-series cross-section methods to an original dataset compiled for the purpose of this study. Different model specifications are tested to ensure the robustness of our results.
Acknowledging the specificities of Italy's recent history, the first section is dedicated to setting the boundaries of what we will call far-right, underlining the differences between the latter term and that of Fascism and Populism. The second section reviews the recent strand in international literature which aims at appraising the effect of migration on the vote for far-right parties. In so doing, we refine our research question and the scope of this paper. The third section defines our empirical strategy and explains our preference for random effects models over fixed effects and Instrumental Variables (IV). We also provide details on our data and estimation technique. The fourth section then presents and discusses our results. Finally, we conclude in the fifth section.

The contours of the Italian far-right: unpacking the "center-right"
As a result of Italy's recent history and political scene over the last sixty years, delineating the contours of the country's far-right is a delicate undertaking. Even the term "right" is subtly avoided by its representatives, for that it resonates with Fascism in public perceptions (Ignazi 2005). Conservative parties thus self-locate themselves at the center-right of the spectrum whilst only the MSI (Movimento Sociale Italiano), which claimed a direct lineage with the fascist regime, affirmed its right-wing positioning (Ignazi 1998;. Analytically speaking though, the self-proclaimed centerright may (and should) be unpacked; and the parties that makes it up may be distributed on a larger array of the political spectrum.
Whilst terms like Fascism, far-right or else Populism are often used interchangeably in common parlance, and even though they may overlap in reality, they remain different concepts that ought to be disentangled for the purpose of this article. Firstly, not all far-right parties share the fascist ideology insofar as they do not all rely on the palingenetic myth of national rebirth through revolution (Griffin 1994;Hainsworth 2008). Instead, they provide answers to those demands and needs generated by post-industrial and individualistic societies, responding to the identity crisis produced by their atomization (Ignazi 2003). As Ignazi (2003, 2) puts it, these "are no longer neo-fascist parties (...) they are perceived as right-extremist because they unquestionably occupy the right-most position of the political spectrum". Secondly, not all populist parties are located at the far-right of the political spectrum, for Populism is more a communication and/ or discursive style than a consistent ideology or programmatic set of ideas (Rydgren 2005; see Gidron and Bonikowski 2013, for an interesting literature review on the topic).
So what is exactly the far-right? This question has mobilized a significant amount of scholarly attention without, however, producing a unanimous definition (Mudde 1995;Eatwell 2004;Hainsworth 2008), despite consensus has been found on the fact that there exists indeed such a family of parties located at the far-right of the spectrum (Camus and Lebourg 2017). The academic production has thus led to the generation of multiple definitions over the years, which, in some instances, differ to significant extents (see inter alia Mudde 1996;Druwe and Mantino 1996;Ignazi 2003;Hainsworth 2008;Griffin, Loh and Umland 2014). An interesting, and empirically useful, approach is proposed by Ivaldi (2004), which consists in defining the core characteristics of this heterogeneous family of parties. He defines four such characteristics: i) an anti-immigration and xenophobic stance; ii) an authoritarian discourse centered on security and strength of the state; iii) an economic program which combines elements of neo-liberalism with protectionist and social claims; and iv) anti-establishment discourse and focus on the public's demands. As Hainsworth (2008) holds, the anti-immigration characteristic appears to be the most important, to the point that some analysts prefer the term "anti-immigration parties" to that of far-right.
Applying these criteria to the Italian case may prove challenging. Whereas there is little doubt that AN (Alleanza Nazionale -National Alliance) or CasaPound may be regarded as far-right political formations, the case of the Lega (Nord -Formerly known as the Northern League, now simply Lega, the League) requires some further considerations. Born as a regional party in 1991, it may have appeared as somewhat "moderate" at the outset (Passarelli 2013, 55), when mostly concerned with a transformation of the state along federal lines (Ignazi 2005). However, from the early 2000s onwards, it has decidedly turned to the right. As the opening line of a book recently written by Passarelli and Tuorto reads: "The (Northern) League is a far-right party" 2 (Passarelli and Tuorto 2018, 9). Corroborating is the Chapel Hill Expert Survey (CHES, 2002-2017) --a pan-European expert survey on political parties and their positioning --, which places the then Northern League close to the AN at the right-most position of the political spectrum. 3 In order to classify Italian parties according to their position on the left-right spectrum, we consider CHES data from 2002 to 2017. Parties featuring an average score greater than 7.5 are considered as far-right parties. Those parties are AN, Lega Nord and Despite the high number of parties that, over the past 18 years, may be assimilated to the far-right, most of them are very small formations, often representing less than one percent of the total number of votes attributable to the far-right (and thus even less if we consider the total of votes expressed). As a matter of fact, AN, Northern League and Brothers of Italy represent together the largest shares of votes to far-right parties. Figure 1 and 2 display the distribution of votes between parties within the group of far-right parties delineated above, respectively for the Lower Chamber and the Senate.

Immigration and vote for the far-right
Whereas the study of the determinants of people's attitudes towards foreigners are plentiful (see inter alia Dustmann and Preston 2001;Mayda 2006;O'Rourke and Sinnott, 2006;Card, Dustmann and Preston 2012), only recently has the focus turned to the effect of immigration on far-right parties' electoral outcomes. The wealth of this recent strand in literature lies in the different approaches employed and applied to different contexts. Mendez and Cutillas (2014) look into the effect of immigration on Spanish national elections over the years 1996-2011. Otto and Steinhardt (2014) estimate the effect of immigration on voting outcomes in the districts of the city of Hamburg, Germany, for national and federal elections in the 1987-2000 period. Harmon (2018) analyses the impact of immigration on votes for the far-right in Denmark. Similarly, Dustmann, Vsiljeva and Piil Damm (2016) study the effect of the refugees' random allocation policy in Denmark and its effect on the vote for far-right parties in parliamentary and municipal elections over the period 1989-1998.  appraise the relationship between immigration trends and vote for the FPÖ in the Austrian elections held in 1983, 1994, 2002. Becker and Fetzer (2016 analyse the effect of immigration on the success of the UKIP party. Finally, Barone et al. (2016) investigate the effect of immigration on Italy's center-right coalition in the national elections of 2001, 2006 and 2008. Interestingly, most these studies have shown a positive effect, more or less sizeable, of immigration on (far) right voting (see Davis and Deole 2017; a notable exception is Steinmayr 2016, who finds a negative association). Whilst they concur in producing a coherent body of evidence, the variety that characterizes them renders the cumulativeness of findings difficult. Most studies were conducted in different countries and consider elections at different levels of government. Likewise, empirical strategies differ significantly. Many have opted for instrumental variables (IV) based on foreigners' self-sorting to proxy their geographical distribution (Mendez and Cutillas 2014;Barone et al. 2016;. Harmon (2018) too opted for an IV, although he used the availability of living space as a predictor of immigrant settlement. Others have used fixed effects models and focused on the share of foreigners in a given location (Otto and Steinhardt 2014;4 Davis and Deole 2017) or have taken advantage of particular contingent features to implement quasi-experimental research designs (Dustmann, Vsiljeva and Piil Damm 2016;Becker and Fetzer 2016).
Most these studies were conducted in countries with a long-standing immigration background and in which many former migrants have acquired citizenship and, thus, the right to vote in the elections. This is problematic if one wishes to truly assess the effect of immigration on election outcomes (Otto and Steinhardt 2014). More recent destination countries have only recently seen their demographic composition change. Consequently, the effect of the phenomenon is likely to be greater and thus better analyzed, empirically speaking. Italy is a case in point. An emigration country, it has only recently (and rapidly) turned into a destination country; this virtually offers the possibility to assess the impact of the passage from very-little to quite-some level of foreign presence (an argument also proposed in Barone et al. 2016). Whereas the percentage of foreigners in 1998 barely amounted to 1.7 per cent of the total population, it nowadays accounts for 8.5 per cent of the population, a share close to that observed in France and Germany. The acceleration of the phenomenon and the recent success of the far-right in Italy thus begs the question: how much is the latter due to the former?

Empirical strategy, estimation technique and data
To answer our research question, we propose to assess the effect of the share of foreigners in each of the Italian regions on the share of ballots that goes to far-right parties out of the total of the votes cast in an election year. Opting for such an approach presents the interest of facilitating its replication in other EU countries, with similar data, thus contributing to increase the comparability of results. We also compare elections for the lower and higher chambers as they feature different electorates and different stakes. Whereas the electorate for the lower chamber comprises all nationals above 18 years old, voters must be 25 or more to vote at Senate elections.
With respect to the method used, this study builds on previous research but opts for a different strategy. First of all, our study differs from that of Barone et al. (2016) on the Italian case in that we seek to estimate the effect of immigration on far-right parties rather than on the center-right coalition, presumably less concerned altogether with immigration. Symmetrically, voters more concerned with immigration issues are also more likely to lean towards the far-right rather than vote for the conservative party. Another difference, we propose to investigate the relationship over a longer timeframe (2001-2018) so as to render our estimates more robust. Note that the period under scrutiny starts in times in which foreign presence in the country was limited, thus reinforcing the validity of our results. 5 The downside of this strategy is that data at province level is scarce and we must therefore reason at region level. From another standpoint, our preferred estimation approach is that of random effects. Indeed, Instrumental Variables (IV) provide valuable insight and considerably reduce the risk of endogeneity posed by cross-section data; however, it is often hard to find a valid IV, able to effectively unveil causal mechanisms. 6 Random (and fixed) effects techniques, on the other hand, can prove particularly useful if used with sound data and if the results are submitted to rigorous robustness tests. 7 Unlike a number of studies in this field, instead of a fixed-effects approach, we prefer the use of random effects and a model specification that accounts for both within-and between-cluster effects; i.e. a model that estimates the effects of variables within the regions but also between them (see Bartels 2009). Because fixed-effects models compute a coefficient for each cluster (here the regions), they absorb the variation between them so that the effect of time-invariant variables (or variables with little change over time) cannot be consistently estimated or meaningfully interpreted, despite their potential substantive importance. Differently, random intercept models allow the estimation of between-clusters effects with, however, the downside of confounding the within-and between-clusters effects. In other words, random effects models assume that the effect within clusters is equal to the effect between clusters, which is oftentimes not the case. Another, unlikely to be true, assumption underlying the use of random effects is that of the absence of covariance between clusters and covariates. Bartels (2009) proposes a random intercept approach that accounts for both the above assumptions. Accordingly, we propose to specify a model based on random effects whilst duly accounting for the difference between within-and between-cluster effects. Note though that we provide different specifications to test the robustness of the results obtained. Our starting equation is therefore: Where ��������� i and � i are the cluster-specific means of our covariates and IMMIG w it and X w it represent the effect within clusters as being respectively IMMIG w it = IMMIGit -��������� i and Regarding the risk of endogeneity mentioned above, it is considerably decreased by the nature of the data at hand which repeats observations over time. In order to further reduce the risk of reverse causality, we also account for the existence of a vote for the far right before migration became a relevant phenomenon in Italy 8 through the addition of a lagged dependent variable. Regarding the estimation technique, we use maximum likelihood as is custom for time-series cross-section analyses.
The data is collected from different sources. Data on election outcomes is taken from the Italian Ministry of the Interior's archive. 9 In order to calculate the share of votes for the farright, all the lists and/or parties that received ballots from 2001 to 2018 have been carefully categorized and pooled together. 10 The data regarding foreigners' settlement is drawn from municipalities' official registers. Because these register people moving in and, less consistently, people moving out, the share of migrants within regions is likely overestimated. We consider that using such data is still more suitable than resorting to IV as it accounts for a trend in (1) Where i represents the regions, t measurement occasions, y our dependent variable, and IMMIG the percentage of foreigners on total population in region i, year t. X is a vector of the other covariates inserted in the model, u 0i accounts for unobserved heterogeneity across clusters, either with fixed or random intercepts, and e it is the level-1 error term. Our main coefficient of interest is therefore β 1 . Following Bartels (2009), we propose a refined specification which accounts for the difference between the effects of covariates within clusters and between them. The refined equation is: Where ��������� i and � i are the cluster-specific means of our covariates and IMMIG w it and X w it represent the effect within clusters as being respectively IMMIG w it = IMMIGit -��������� i and Regarding the risk of endogeneity mentioned above, it is considerably decreased by the nature of the data at hand which repeats observations over time. In order to further reduce the risk of reverse causality, we also account for the existence of a vote for the far right before migration became a relevant phenomenon in Italy 8 through the addition of a lagged dependent variable. Regarding the estimation technique, we use maximum likelihood as is custom for time-series cross-section analyses.
The data is collected from different sources. Data on election outcomes is taken from the Italian Ministry of the Interior's archive. 9 In order to calculate the share of votes for the farright, all the lists and/or parties that received ballots from 2001 to 2018 have been carefully categorized and pooled together. 10 The data regarding foreigners' settlement is drawn from municipalities' official registers. Because these register people moving in and, less consistently, people moving out, the share of migrants within regions is likely overestimated. We consider that using such data is still more suitable than resorting to IV as it accounts for a trend in (2) Where yit = i0 + β1IMMIGit + β2Xit + ui0 + eit (1) Where ��������� i and � i are the cluster-specific means of our covariates and IMMIG w it and X w it represent the effect within clusters as being respectively IMMIG w it = IMMIGit -��������� i and Regarding the risk of endogeneity mentioned above, it is considerably decreased by the nature of the data at hand which repeats observations over time. In order to further reduce the risk of reverse causality, we also account for the existence of a vote for the far right before migration became a relevant phenomenon in Italy 8 through the addition of a lagged dependent variable. Regarding the estimation technique, we use maximum likelihood as is custom for time-series cross-section analyses.
The data is collected from different sources. Data on election outcomes is taken from the Italian Ministry of the Interior's archive. 9 In order to calculate the share of votes for the farright, all the lists and/or parties that received ballots from 2001 to 2018 have been carefully categorized and pooled together. 10 The data regarding foreigners' settlement is drawn from municipalities' official registers. Because these register people moving in and, less consistently, people moving out, the share of migrants within regions is likely overestimated. We consider that using such data is still more suitable than resorting to IV as it accounts for a trend in and yit = i0 + β1IMMIGit + β2Xit + ui0 + eit (1) Where ��������� i and � i are the cluster-specific means of our covariates and IMMIG w it and X w it represent the effect within clusters as being respectively IMMIG w it = IMMIGit -��������� i and Regarding the risk of endogeneity mentioned above, it is considerably decreased by the nature of the data at hand which repeats observations over time. In order to further reduce the risk of reverse causality, we also account for the existence of a vote for the far right before migration became a relevant phenomenon in Italy 8 through the addition of a lagged dependent variable. Regarding the estimation technique, we use maximum likelihood as is custom for time-series cross-section analyses.
The data is collected from different sources. Data on election outcomes is taken from the Italian Ministry of the Interior's archive. 9 In order to calculate the share of votes for the farright, all the lists and/or parties that received ballots from 2001 to 2018 have been carefully categorized and pooled together. 10 The data regarding foreigners' settlement is drawn from municipalities' official registers. Because these register people moving in and, less consistently, people moving out, the share of migrants within regions is likely overestimated. We consider that using such data is still more suitable than resorting to IV as it accounts for a trend in are the cluster-specific means of our covariates and IMMIG w it and X w it represent the effect within clusters as being respectively IMMIG w it = IMMIG it -yit = i0 + β1IMMIGit + β2Xit + ui0 + eit Where ��������� i and � i are the cluster-specific means of our covariates and IMM represent the effect within clusters as being respectively IMMIG w it = IMMIGit - Regarding the risk of endogeneity mentioned above, it is considerably de nature of the data at hand which repeats observations over time. categorized and pooled together. 10 The data regarding foreigners' se municipalities' official registers. Because these register people moving people moving out, the share of migrants within regions is likely over that using such data is still more suitable than resorting to IV as it . Regarding the risk of endogeneity mentioned above, it is considerably decreased by the nature of the data at hand which repeats observations over time. In order to further reduce the risk of reverse causality, we also account for the existence of a vote for the far right before migration became a relevant phenomenon in Italy 8 through the addition of a lagged dependent variable. Regarding the estimation technique, we use maximum likelihood as is custom for timeseries cross-section analyses.
The data is collected from different sources. Data on election outcomes is taken from the Italian Ministry of the Interior's archive. 9 In order to calculate the share of votes for the far-right, all the lists and/or parties that received ballots from 2001 to 2018 have been carefully categorized and pooled together. 10 The data regarding foreigners' settlement is drawn from municipalities' official registers. Because these register people moving in and, less consistently, people moving out, the share of migrants within regions is likely overestimated. We consider that using such data is still more suitable than resorting to IV as it accounts for a trend in foreigner settlement that is confirmed by occasional census data. 11 In addition to that, the data register the movement of foreigners in a given municipality at a given time, which is likely to affect behaviors. Beside our main variables of interest, we also control for alternative explanations such as the economic and unemployment trends, demographic changes and region size 12 . We also provide estimates of time fixed effects to control the effect of external shocks, notably the absorption of AN into Berlusconi's party and the scandal over the misuse of Lega Nord's funds by ex-party leader Umberto Bossi (Passarelli and Tuorto 2018).  Table 2 below reports the results of a series of regressions following equation (1). The first model (M1) is a random intercept model; the second one (M2) is a random intercept model, too, but it includes time fixed effects. The third and fourth models are cluster fixed effects models, with (M3) and without time fixed effects (M4). Time effects are included in M2 and M4 to test the robustness of our variable of interest to external shocks: i) in 2008, the political party Alleanza Nazionale was to merge with Forza Italia (then Berlusconi's party) into the People of Freedom (Popolo della Libertà), thus considerably affecting our dependent variable; ii) in 2013, the leader of Lega Nord, Umberto Bossi, was accused of misuse of public funds, a scandal that has weighted on the party's success in the 2013 national elections (see Passarelli and Tuorto 2018, for more). In addition to that, time effects also account for the economic downturn which ensued the Global Economic Crisis. Model 1 is our preferred specification and will be used below for a refined analysis following equation (2). 13 The results of the regressions show an effect of the foreign population on the percentage of ballots for the far right that is consistent across models and statistically significant, despite some changes in the coefficient's magnitude (table  2). On average, a one percentage point (p.p.) increase of foreigners on the total regional population translates into a 1.2 p.p. increase in the percent of votes that goes to far right parties (with the estimates ranging from 0.9 to 1.4 across models). For our preferred model (M1), an increase of 1 p.p. in the foreign population corresponds to an increase in 1.4 p.p. of the dependent variable.
Note that the chamber 14 for which the ballot is cast matters a great deal: elections for the lower chamber count an average of 2.4 p.p. more for far right parties than votes at the Senate. The effect of unemployment, too, appears to be a relevant factor of the vote for the far right, as it is statistically significant in M1 and M3. It is not statistically significant in M2 and M4 due to model specification: its effects, consistent across regions, are absorbed by year fixed effects. Interestingly, voter turnout 15 proves to be significant, both statistically and substantively, so that, all things being equal, the more people go to vote, the more the percentage of votes for the far right increase.
We propose to further investigate the effect of migration onto votes for the far right by breaking down its effect at regional level and within the regions themselves (see equation (2) for more detail). Table 3 reports the results of the regressions. Considering the effect of foreigners, it appears to be mostly a within-cluster effect for that most of the effect estimated in M1 occurs within the regions and is statistically significant (this is corroborated by M3 and M4). Conversely, the between-region effect is of little magnitude and is not statistically significant, meaning that, on average 16 , there is no significant, region-specific, effect on the outcome. This strongly supports our hypothesis of a direct association between increase of foreign population and increase of the share of votes taken by far right parties. Interestingly, by disentangling the effects of unemployment between and within regions, there appears to be a strong correlation between unemployment and vote for the far right. The coefficient of the difference between within-and between-cluster effects (-2.012; fourth column on the right-hand side) is significant, indicating some level of cluster confounding between the two coefficients and suggesting the necessity to estimate them separately. The between-region effect compares average regional unemployment rates and their association with the dependent variable so that there appears to be an effect of overall unemployment levels in a given region. The sign of the coefficient is an interesting result: it suggests that regions with lesser average unemployment rates --typically northern regions --tend to vote more for the far right than regions characterized by more overall unemployment --typically southern regions --, this in spite of the overall levels of foreign presence (since the coefficient, 0.324, is of little magnitude and is not statistically significant). Turning to the within coefficient, it shows that variation of unemployment in time affects far right parties' success: as unemployment increases over time, so does votes for the far right. The magnitude of the coefficient of within-region unemployment is somewhat higher than that of the percentage of foreigners 17 , thus suggesting a greater effect on votes for the far right.

Discussion
The results presented above confirm those reached elsewhere and reported in specialized literature (see literature review above). They thus enrich and develop a larger corpus of findings. However, such a confirmation must not preclude prudence in interpreting the results. Despite all the precautions taken and the inclusion of a lagged variable that accounts for a period of time in which there was little immigration and there was vote for the far right, there still exists a risk of reverse causality; i.e. the risk that vote for the far right affects the distribution of immigrants in Italian territory (see Bracco et al. 2017, for an interesting example). In our study, and taking into account previous research, we conclude to the absence of reverse causality; if there were, our coefficient for the percentage of foreigners would most likely have turned negative, as suggested by Bracco et al.'s findings (2017). In addition, we have tested out various specifications, be it in terms of modelling or in terms of controls. None of our tests has drastically changed our results. A last caveat is in order. Our analysis considers regions instead of smaller constituencies such as towns or provinces. Yet, investigating smaller constituencies would make a lot of sense as there are disparities within regions and between towns when it comes to the share of foreigners. That said, regions offer a good geographical and political reality to study the phenomenon. As bigger entities, focusing our analysis on regions reduces the risks of "voting with one's feet" --so to speak --; that is, for migrants or natives to move from one town to another if vote for the far right increases for the former (Bracco et al. 2017) or if immigration increases for the latter . Likewise, considering regions also increases the quality of administrative data for the presence of migrants for the same reason. In addition, data scarcity at local level is a major issue when it comes to considering such a long time span. Reasoning at regional level ensures we use adequate data to test our hypotheses.

Conclusion
Whilst a great deal of specialized scholarship has looked into people's attitudes towards immigration, a recent trend in literature seeks to investigate the effect of immigration on election outcomes. Most studies, though, have concentrated on old destination countries, with the risk of having their estimates biased by a population of voters already diverse and, presumably, more opened to the phenomenon. This paper aims at contributing to this developing literature by adding the case of a country less accustomed to immigration: Italy. For long an emigration country, Italy has only recently, and rapidly, become a destination country. The sudden change in the composition of its population, but not so much of its electorate, likely brings political changes about. Accordingly, this paper investigates the relationship between immigration and vote for the far right in Italy. We focus our analysis on the Italian national elections for the Camera and the Senate that took place during the years 2001-2018, a period in which immigration passed from a marginal phenomenon to being one of the main subjects of concerns for Italian citizens (Ortensi et al. 2017). Our analysis is based on the collection and assemblage of administrative data at regional level. We resort to time-series cross-section methods to appraise the effect of immigration on votes for the far right over time. Different model specifications underpin the existence of an effect of the increase in the share of migrants in Italian regions on the share of ballots cast for far right parties (between 0.9 and 1.4 percentage point with every percentage point increase in the share of foreigners). We find that such effect is consistent across regions. At the same time, we find that the economic situation plays a significant role, yet not exactly that expected. Our results show that regions with generally higher unemployment rates are less prone to vote for far right parties. That said, increasing unemployment rates over the years within regions consistently translates into greater shares of votes for the far right. Our estimates reveal an almost symmetrical effect, with about one percentage point increase for every one percentage point increase in unemployment. With all due precautions taken, such an effect may be deemed higher than that of immigration, thus confirming and contrasting other studies conducted in European and North American countries.