The political effects of (mis)perceived immigration

Abstract

Several studies document that exposure to actual immigration affects political outcomes. This article examines, instead, the influence of expected immigration, using data from local elections in Italy. We develop an index of potential exposure to pre-electoral sea arrivals, which varies over time and space depending on immigrants’ nationality. We find that such potential exposure causes a decrease in turnout and an increase in protest votes, shifting valid votes toward extreme-right parties. Support for populist and anti-immigration parties increased in highly exposed municipalities, where voters believed that the new inflow of refugees would increase the local stock of immigrants. However, Twitter data show that these expectations do not reflect actual immigration trends; immigration salience rises mainly during the election period, while most arrivals occur months later. This suggests that, around elections, informal media can bias people’s expectations and, consequently, influence voting behavior.

1. Introduction

Recent national and European Parliament elections across European countries have shown increasing support for far-right and right-wing populist political parties, calling for a rise of nationalism in Europe (Guiso et al. 2017, 2019; Colantone and Stanig 2019; Daniele, Aassve, and Le Moglie 2023). This political change has been exacerbated by the refugee crisis that peaked in 2016, as suggested by a growing number of studies examining, more in general, the influence of migration on political outcomes (Alesina and Tabellini 2024). Most of these studies, however, focus chiefly on the political effects of actual immigration episodes, overlooking the role of voters’ beliefs about prospective immigration. This study fills this gap analyzing how expectations of refugee arrivals affect political preferences. This question is crucial for countries like Italy, which are exposed to frequent episodes of refugee immigration, and where the mismatch between perceived and actual immigration, jointly with the upsurge of populist parties, has been extraordinarily high.¹

Our main hypothesis is that the threat of (and anxiety about) refugee arrivals affected voting outcomes in Italian local elections. Voting preferences might be gradually shaped not much, or not only by past and present exposure to local immigrants (actual immigration) but also by expectations about future immigration (expected immigration). Arrival episodes usually gain importance in the media before the elections, thereby increasing the salience of immigration in political contests and citizens’ perceptions (Newman and Velez 2014). Exposed to slanted media reports presenting the refugee arrivals as an “immigration crisis,” voters may form biased expectations about prospective immigration, overestimating the magnitude of the phenomenon. Leaders of far-right political parties, in turn, fuel such expectations, showing that immigration can be a threat, aiming to influence voters’ attitudes toward immigration and their political preferences (Gentzkow, Shapiro, and Stone 2015; Allcott and Gentzkow 2017; Barrera et al. 2020; Stantcheva, Alesina, and Miano 2022; Couttenier et al. 2024). The combination of increased salience, biased expectations, and the heightened threat of immigration might lead unsatisfied supporters of traditional parties to either abstain from voting or express discontent through invalid/blank ballots or support for anti-establishment parties (Barone et al. 2016). Consequently, those parties proposing immigration restrictions can gain vote shares.

To assess the political impact of immigration expectations, we exploit variation in the time of refugee arrivals and the nationality composition of the boats landing at Italian ports, which, as we show, is exogenous to the local political cycle. We build an index of potential exposure to immigration that varies by municipality and over time, weighting the number of arriving nationalities by the share of incoming refugees’ co-nationals residing in each municipality. Because, after disembarking, refugees cannot freely and immediately reach their desired destinations, our index captures the expected threat of refugee arrivals before the elections. More specifically, as migrants tend to settle where they have a sizeable pre-existing network of co-nationals (Altonji and Card 1991; Barone et al. 2016), voters may form expectations about future immigration by combining information on pre-electoral refugee landings and the local stock of immigrants. Thus, immigration expectations can be higher (lower) where refugees are more (less) expected to go after landing, that is, in municipalities with a high (low) share of regular migrants that have the same nationality as that of the incoming refugees.

Relying on data from Italian municipal elections from 2010 to 2018, we perform first-differences regressions of political outcomes on the potential exposure to refugee arrivals, that is, our empirical measure for expected immigration. Controlling also for the local share of migrants and the presence of refugee centers, which capture actual exposure to migration (Bratti et al. 2020; Vertier, Viskanic, and Gamalerio 2020), our empirical strategy provides estimated impacts of expected immigration on changes in political preferences.

We present two key findings. First, the potential exposure to new arrivals predicts the decline in voter turnout in Italian municipalities during the studied period and significantly contributes to the increase in protest votes. This result may be driven by pro-multicultural voters, including left-wing citizens, who are concerned about the evolving immigration crisis and dissatisfied with the current administration. Consequently, these voters may prefer to abstain from voting or invalidate their ballots rather than support an ideologically distant political force, such as a far-right party (Barone et al. 2016). Second, support for populist, anti-immigration, and extreme-right parties increase in municipalities with high exposure to potential immigrant inflows. Voters in these areas, anticipating significant immigration, tend to favor these parties. All these results, however, show nonlinear trends, supporting the established idea (Steinmayr 2021; Gamalerio et al. 2023) that prolonged exposure to immigration may allow for the development of positive contact with immigrants, thereby fostering positive attitudes toward refugees upon their arrival.

Additional analyses shed light on the primary driving force behind our findings. We posit that the political effects of potential refugee exposure mainly stem from misperceptions and the perceived threat of future immigration. To support this hypothesis, we first document that immigration salience, proxied by immigration-related tweets, does not reflect the actual number of arrivals but rather follows the electoral cycle. Furthermore, while immigration salience and political preferences are influenced by refugee landings, actual local immigration levels are not. These findings suggest that heightened attention to immigration on social media may create false expectations about future immigration, which can consequently increase support for anti-immigration and populist parties. Additionally, we show that municipalities with better broadband coverage and provinces where immigration is a salient issue on social media drive the main findings, especially where sentiments of fear and anger dominate the discussion. Moreover, our results are particularly strong for municipalities closer to disembarkation ports, emphasizing the significant role the perceived threat of future arrivals plays in voting behavior. Finally, a comprehensive set of robustness and heterogeneity tests confirms our results, highlighting the main mechanisms at work and ruling out endogeneity concerns.

This article contributes to the previous literature in two ways. First, we assess how voters perceive and react to refugees’ (prospective) arrival on the Italian coasts instead of estimating the impact of actual immigration (often measured as stock or flow of regular immigrants) on voting behavior. In other words, we test the role of potential rather than actual contact with immigrants, as real intergroup interactions do not enter our measure of immigration exposure (though we also control for it). Thus, conditional on the local share of regular migrants, our estimates identify the role of expected immigration on top of the political effects of regular migration. Second, we focus on the role of mass media in political attitudes and outcomes. Both the frequency and the tone of coverage of immigrants in the media influence the dynamics of anti-immigration attitudes (Boomgaarden and Vliegenthart 2009). It has also been highlighted that the media coverage of migration boosts immigration worries (Benesch et al. 2019), and the spread of fake news may affect electoral outcomes (Barrera et al. 2020; Cantarella, Fraccaroli, and Volpe 2023). By analyzing whether immigration salience is higher in municipalities mostly exposed to the media, we also test whether the importance of immigration-related issues followed the actual trend of refugee arrivals or, instead, the electoral cycle.

The article is structured as follows: Section 2 discusses the institutional and political context. Section 3 outlines the variables and empirical strategy. Section 4 presents the baseline results, addresses endogeneity issues, and reports heterogeneity and robustness checks. Section 5 explores the main mechanisms. Section 6 concludes.

2. Background

2.1 Migrant landings in the media

Landing episodes were primarily discussed in the media before the elections. Google Trends statistics show that the frequency of searches for a migration-related topic in Italy follows the electoral cycle (Fig. 1). In the binscatter plot, we observe a positive correlation between the standardized number of Google searches for the word “Immigrants” and the distance to the elections. Notably, immigration-related Google searches have an upward trend until local election month, yet go back to the pre-electoral level thereafter. Moreover, considering the Italian words “Immigrati” (immigrants) or “Rifugiati” (refugees) contained in province-level tweets, the salience of immigration increased over time, reaching its peak in 2018.² While the frequency of tweets generally follows the actual sea arrivals of refugees, we observe a mismatch between actual arrivals and migration-related tweets from 2017 onward, when immigration-related tweets increase while boat landings decrease (Fig. 2).

These dynamics suggest that, while voters are aware of and concerned about arrivals, their perceptions might not align with actual immigration patterns. Additional descriptive evidence highlights that national news outlets prioritize coverage of immigrants following arrivals, thereby increasing the topic’s salience. We collect immigration-related tweets from prominent national news providers spanning from 2014 to 2018. Using an event study methodology, we compare the share of immigration-related tweets 3 months before and 4 months after a boat landing. Results indicate that media outlets’ coverage of refugee landings increases following an arrival episode. This broader coverage coincides with heightened public demand for information, as evidenced by a rise in Google searches for both boat landings and immigrants in the months following a boat landing (Supplementary Appendix Fig. A3). We interpret this additional evidence as indicative of a pattern wherein national news outlets intensify coverage of boat arrivals, thereby informing citizens and eliciting increased public interest in the topic.

2.2 Institutional background

We use data from Italian municipal elections, which involve different municipalities in different years. Several elections occurred in the years considered in this article, with only some of the municipalities voting in the same year and at the same time.³ The municipal level in Italy includes over 8,000 government authorities. Elections (local council and mayor) occur every 5 years, with direct election of the mayor on a single or dual ballot, depending on population size. Cities with more than 15,000 inhabitants have a runoff stage among the two most-voted candidates if none collects more than 50 per cent of the votes in the first stage. We use 2,877 municipalities regularly voting twice in 2010 and 2015, 2011 and 2016, 2012 and 2017, and 2013 and 2018 (91 out of 110 provinces). We drop from the sample municipalities that vote with a different schedule, for example, when mayors, or at least half of the councilors, resign before the end of the term, dissolution of suspected mafia presence in the council, merging with other municipalities, and other law violations.

We describe the Italian migrants’ reception system (as of the period considered in this study) in Section 2 of the Supplementary Appendix. Here, we highlight a critical institutional feature of our identification strategy: refugees cannot choose which protection centers for asylum seekers and refugees (SPRAR) to settle in. Instead, they are assigned to a center based on the beds’ availability. Additionally, refugees arriving at Italian ports immediately enter the reception system, which in the initial phase severely restricts their freedom of movement.

Refugees can stay in a SPRAR for up to 12 months. During and after this period, local administrations implement initiatives to promote individual autonomy, including enhanced language training, vocational guidance, access to essential public services, and knowledge of fundamental constitutional rights and duties. Only after a considerable period of time are refugees eventually free to move to their desired destination. Thus, there is often a lag as refugees may take considerable time to achieve economic independence and relocate. While they can move earlier, this may involve illegal actions such as escaping from assistance and reception centers while awaiting asylum procedures or from repatriation centers if they do not qualify as refugees.

3. Data and empirical strategy

The primary dataset is a combination of different sources of data. Information on refugees’ arrivals by boats at Italian ports from 2010 to 2018 comes from the Italian Minister of Interior. Data on electoral outcomes of all Italian municipalities that voted twice from 2010 to 2018 are also available on the Italian Ministry of Interior’s website (https://elezionistorico.interno.gov.it). We use municipality characteristics from the Italian Statistical Office (https://www.istat.it), the presence and capability of SPRAR centers from the Reception and Integration System (https://www.retesai.it), and the broadband availability in 2018 at the municipality level from the Authority for Communications Guarantees (https://www.agcom.it). We also rely on Twitter data to derive the frequency of messages containing immigration-related words from 2010 to 2018 at the province level. Further details on data sources are in Supplementary Appendix, Section 3. We show descriptive statistics for the main variables of interest, comparing the two electoral periods in Table 1. To complement this, Supplementary Appendix, Section 4 includes additional summary statistics (Supplementary Appendix Table A1A–D) and maps (Supplementary Appendix Fig. A14) that illustrate the temporal and cross-sectional variation of these variables. In the Supplementary appendix, we also provide further details on the construction of the main variables used in the empirical analysis.

3.1 Exposure to refugee arrivals

Our primary source of exogenous variation relies on the random matching of nationalities on boats arriving at Italian ports before the elections (presumably exogenous to local political dynamics) with those residing in the voting municipalities. For each municipality, we constructed an index of exposure to refugees arrived at Italian ports, described in the following equation:

where $IMMj,i,y$ is the ratio between the number of immigrants of nationality $j$ residing in municipality $i$ and the total number of immigrants of nationality $j$ in Italy, at time $y$ (year); $REFj,k,t,y$ is the number of refugees of nationality $j$ in boat landing $k$ occurred $t$ days before the election day in year $y$ (year).

We leverage the temporal and spatial variations in exposure to arrivals, determined by the match between resident and landed nationalities of immigrants. We assume that refugees often settle in communities with existing clusters of co-nationals, as documented by numerous studies (Buggle et al. 2023) and tested in Section 4.4. Consequently, municipalities with a substantial (limited) pre-arrival proportion of regular immigrants, whose nationality coincidentally matches that of the incoming refugees, experience higher (lower) exposure—and thus greater sensitivity—to the new arrivals.

Hence, EXPIND measures the intensity of exposure to refugee immigration at the municipal level, weighting the number of entrant refugees by the local share of migrants in the municipality whenever their nationality coincides. The index is our proxy of the number of incoming immigrants expected to arrive in the municipality because of pre-electoral boat landings.⁴

First, we consider $t$ as the period between the 1st of January and the election day in year $y$⁠. Then, we also compute the index by changing the time window such that $t$ includes all landing episodes 30, 60, or 90 days before the election day.

Some concerns may arise from the potential endogeneity of the local share of immigrants (and of specific nationalities) to political preferences and electoral outcomes. Through a large set of additional estimates in Section 4.2, we offer reassuring evidence that these issues do not bias our results. In Section 4.4, we relax the implicit assumption that voters are perfectly informed about every nationality of the incoming refugees and of migrants residing in their municipality.⁵ Finally, we prove that pre-determined municipal characteristics do not predict changes in the exposure index.⁶

Our index of expected immigration relies on the assumption that voters form their immigration expectations based on information about arrivals and the share of immigrants in their municipalities. Proving this assumption empirically is challenging due to the limited data on immigration expectations at the municipal or provincial level. However, we identify an alternative data source: ITANES, a large sample survey conducted after parliamentary and other elections in Italy, which in 2013 included a question on perceived immigration reception capacity. We use this question to derive a province-level measure of immigration (mis)perception by comparing respondents’ estimated hosting capacity for immigrants with the actual number of immigrants residing in their province (relative to the regional median). After residualizing this (mis)perception measure to account for individual-specific characteristics (e.g. political leaning), we correlate it with our key measure of immigration expectations (exposure index) and find a positive correlation (see Section 5 in the Supplementary Appendix for further details). Although this evidence is just descriptive, it nonetheless supports our conjecture that the exposure index reflects individuals’ beliefs about immigration.

3.2 Electoral outcomes

Turnout and distribution of votes are our main outcome variables. Turnout is the ratio between the number of votes and individuals entitled to vote (rescaled to be 0–100 per cent). The distribution of votes allows us to observe the political preferences of citizens. We group votes into four non-mutually exclusive categories and compute their relative share of votes.

First, we consider the protest vote, which groups null and white votes.

Second, we use anti-immigration votes, defined as the sum of preferences favoring right and extreme-right parties. To categorize anti-immigration parties, we group those characterized by strong rhetoric against immigrants and ethnic minorities, publicly referring to migration flows as a national security concern, advocating for national border closures, and prioritizing the domestic population over foreign citizens. This group includes Lega, Forza Nuova, Casa Pound, Movimento Sociale Italiano, and Alleanza Nazionale.

Third, we define populist votes as the sum of votes for parties classified as populist based on the seminal work by Kessel (2015). According to this classification, populist parties are characterized by political ideas that especially emphasize: (1) the distinction between “the people,” viewed as the inherently good segment of society, and “the elite,” (2) the supremacy of “the people” over “the elite,” and (3) themes of national sovereignty. Populist parties included in this category are Forza Italia, Il Popolo delle Libertà, Lega, and Movimento 5 Stelle.

Finally, we consider the Northern League coalition as the sum of all the votes collected directly by “Lega” and its closely related parties. The Lega list contains votes cast for Lega, Lega Nord, and Lega Padana.

Further details on the classification of parties are in the Supplementary Appendix, Section 6.

3.3 The empirical model

We assess the political effects of potential exposure to refugee landings in all the voting municipalities by estimating the following equation:

where for each municipality $i$⁠, $ΔY$ is the difference in turnout, protest votes, or vote shares for anti-immigrant, populist, and Northern League parties between the two municipal elections (occurred in year $y$ and $y−5$⁠). $ΔEXPINDi,y,t$ is the change between the two municipal elections in the exposure to migration (our treatment variable), which is defined in Equation (1).

$ΔΓi,y$ comprises two variables that capture how the presence of refugees or regular immigrants affects voting behavior. The first variable, $Δ$  Share of Migrants, measures the change in the population share of regular immigrants (excluding those with Italian citizenship) living in the municipality. This variable allows us to control for the exposure to actual immigration.⁷ The second variable, $Δ$  Total SPRAR beds pc, is the total number of available beds in SPRAR centers per 1,000 inhabitants at the province level. This variable is used as a proxy for the presence and size of refugee centers to control for (present and past) contact with refugees and asylum seekers through refugee allocation on voting behavior.⁸

$Δϒi,y$ is a vector of municipality characteristics including, as first differences: the number of individuals entitled to vote at the municipality level, Electorate, which takes into account the changes in the size of the electorate due, for instance, to the historical variation in the dimension of the cohorts where voters enter for the first time⁹; the number of mayor candidates at the elections at municipality level, Number of Mayors, which allows controlling for political competition (higher values imply higher competition); the share of citizens with annual personal taxable income greater than 120,000€, Share of taxable income $≥$  120,000 euro, since immigration policies and political outcomes may be sensitive to top-income voters¹⁰; the ratio between the share of elderly individuals over 65 years old and the share of children between 0 and 14 years old, Ageing index, to capture demographic dynamics. All these controls are included for each municipality $i$ and year $y$⁠.¹¹

We also include time-fixed effects $δy$ to control for common factors specific to each year, such as, for instance, the business cycle. Note that all municipality time-invariant characteristics are net out by the first-difference estimator.

In all specifications, standard errors are clustered at the province level to account for within-province error correlation that could derive, for instance, from geographical spillovers (e.g. voters’ reactions to perceived immigration to neighboring cities). $ϵi,y$ is the classic error term. Equation (2) is estimated considering exposure to migration in the period $t$ ranging from 1^st January and the election day in year $y$⁠; in alternative estimates, we compute exposure to migration for $t$ = 30, 60, or 90 days before the election day in year $y$⁠.

The main parameter of interest in Equation (2) is $β1$⁠, which identifies the effect of the change in the potential exposure to migration across municipalities on changes in the electoral outcome.¹²

4. Results

4.1 Expected immigration, participation, and preferences

Table 2 reports the estimates of the effect of the potential migration exposure considering all the arrivals occurring from the beginning of the year to the election day, as well as for 1, 2, or 3 months before the election day.¹³ The shorter period in the first check further excludes the possibility that refugees legally or illegally reach the municipality.

Results highlight that the increase in potential exposure causes a decrease in turnout (Table 2, Column 1), although this effect is marginally statistically significant. This result suggests that the recent trends in immigration may have mildly contributed to a surge of disaffection toward political participation. As suggested by Barone et al. (2016), some of the center and left-wing voters, who are ideologically more in favor of a multi-ethnic society but unhappy about the immigration trends and policy regulations, have decided not to vote instead of voting for the ideologically distant center-right coalition (Dustmann, Vasiljeva, and Damm 2019; Steinmayr 2021). This result is also confirmed by Edo et al. (2019), who find that higher immigration increases abstention rates.

Furthermore, if citizens are unsatisfied with existing political parties and immigration policies, we should expect increased protest votes, too. Indeed, we find that exposure to arrivals has a positive effect on the share of blank/invalid votes (Table 2, Column 2), which is consistent with the idea that the prospect of incoming refugees contributed to an increase in dissatisfaction with how mainstream parties addressed the issue (see Barone et al. 2016 for a similar result).

Results show a positive effect of expected immigration on votes for center-right coalitions, which have a political platform less favorable to immigrants, suggesting that exposure to migration increases support for anti-immigration and populist parties, independently from the time window for refugee arrivals (Table 2, Columns 3 and 4). Finally, exposure to arrivals increases support for the Northern League party (Table 2, Column 5). In this case, the sample is restricted to municipalities in the North macro-area, where the party traditionally enjoys higher consensus.

Regarding effect size, we estimate that a one standard deviation increase in the rescaled exposure index reduces turnout by 0.02 standard deviations. This implies that an increase in the rescaled exposure index by 0.05 units (equivalent to the average change of 5 units in the original scale) leads to a decrease in turnout of 0.032 per cent with respect to the mean outcome. Similarly, a one standard deviation increase in the rescaled exposure index is estimated to increase protest votes by 0.137 standard deviations, indicating that a 0.05-unit increase leads to a 0.045 percentage-point rise in protest votes, equivalent to a 1.15 per cent increase relative to the mean protest vote share. Furthermore, we find that a one standard deviation increase in the rescaled exposure index raises the share of votes for anti-immigration parties, populist parties, and the Northern League by 0.089, 0.169, and 0.041 standard deviations, respectively. Translating these into changes for a 0.05-unit increase, the vote shares for these parties increase by 0.106, 0.237, and 0.032 percentage points, which correspond to relative increases of 2.46 per cent, 3.71 per cent, and 0.79 per cent compared to their mean vote shares of 4.3 per cent, 6.4 per cent, and 4.1 per cent, respectively.¹⁴

Since our identification strategy relies on a continuous treatment (exposure index), establishing a clear benchmark can aid in interpreting our findings. Therefore, we divide our municipalities into two groups represented by a dummy variable: those with an exposure index below the 75th percentile at baseline (in the first year of observation) and those above this level. We then interact this dummy with our continuous and time-varying exposure index. This allows us to compare the effect of a marginal increase in exposure index on political outcomes between municipalities with a lower versus high exposure index (at baseline). Municipalities with a lower exposure index should be the ones driving our results since voters residing in those municipalities are generally less exposed to immigrants and, hence, more likely to be “shocked” by the migrant waves. We present this result in Supplementary Appendix Table A4, which provides empirical support for this hypothesis.¹⁵

Finally, we check for potential nonlinearities in the political effects of perceived immigration and re-estimate Equation (2) adding the quadratic term for the exposure index (Supplementary Appendix Table A5).¹⁶ We find an inverse U-shape effect of perceived immigration on voting for anti-immigration and populist parties (including voting for Lega). Consistent with previous studies (Steinmayr 2021; Gamalerio et al. 2023), this nonlinearity can be explained through the lens of the contact and habituation hypotheses: municipalities with a large share of migrants of given nationalities are less politically responsive to the potential arrival of further migrants of those same nationalities since they might have developed a history of positive contact and everyday experience with those immigrants over time. Conversely, voters who are less experienced with migrants from given countries might perceive them as a threat upon arrival.

4.2 Addressing endogeneity

A possible threat to our identification strategy is the spatial sorting of immigrants into municipalities, which could be endogenous to the political process. Local political attitudes or unobserved characteristics might induce immigrants from a specific country of origin to settle in or move away from more or less favorable places (Fratesi, Percoco, and Proietti 2019). Location decisions might bias our index as the latter leverages the local distribution of immigrants by nationality. In other words, the estimated effect of expected immigration could reflect the endogenous sorting in nationality composition of local immigrants, which could stem from the political process or other unobserved factors affecting immigration decisions and voting behavior.

In theory, the endogenous sorting of immigrants should not represent a serious concern in our framework. In each landing episode preceding the election date—which has been exogenously determined—it is unlikely that the composition of the incoming nationalities is affected by the local political process. For this type of sorting to be a problem, refugees should be able to schedule the day and choose the destination city in response to the political process of that city. We can exclude this possibility because, at the departure, refugees do not enjoy the freedom of choice regarding the day of leaving and the day and place of arrival (see Section 2). Migrants could not exactly know when they would travel, when they will land, and whether and when they will eventually reach the municipality they intend to go to.

We nonetheless empirically address this concern in five ways. First, we re-compute our exposure index by fixing either in the first election year or in 2004 the local share of immigrants whose nationality matches that of the incoming refugees. This allows us to exclude migrants’ relocation decisions that may be affected by electoral outcomes that are favorable or unfavorable to them. We match the time-varying nationalities in the boat landing with the corresponding nationalities in each voting municipality measured in the first election wave. Therefore, the matching nationalities living in the Italian municipalities are treated as time-invariant. Results, reported in Supplementary Appendix Table A6A–B, are consistent with our baseline estimates.

Second, we implement an instrumental variable strategy similar to Card (2001) and instrument our index with the historical settlement patterns of immigrants. The latter would reasonably affect recent political outcomes only through the current spatial distribution of migrants; it is also expected to affect their present-time spatial distribution since immigrants tend to move to areas where their co-nationals have already settled in the past. We construct a historically lagged version of the exposure index, which we use as an instrument in the baseline estimates. More specifically, we replace the nationalities residing in each municipality in the two election years (matched with the landing nationalities) with those computed in 2004, the oldest official data we can get from the National Statistical Office website.¹⁷ In an alternative computation of the index, we use the local share of immigrant nationalities in 1991, which we have constructed through the data and methods used by Barone et al. (2016). The identifying assumption is that immigrants’ location decisions that occurred in the past (1991 or 2004) are uncorrelated with the recent changes in political preferences (2010–2018).¹⁸ Note that this IV approach also mitigates measurement error issues, which could be induced by the presence of illegal immigrants not entering the official tracking or by misreporting or under-reporting of migrants’ movements. The first and second-stage results are reported in Supplementary Appendix Table A7A–B. As expected, the historical version of the index positively predicts the index calculated at time $t$⁠, and second-stage estimates are consistent with our main results.

Third, we also perform a falsification test by looking at municipalities for which electoral outcomes cannot be affected by the sea arrivals of refugees. We apply the same identification strategy as in Equation (2) using, for each municipality, the electoral outcomes obtained 10 years earlier (e.g. 2000–2008). The exposure effect in these “placebo” estimates should not be significantly different from zero: political outcomes are unlikely to be affected by the boat arrivals that occurred 10 years later. Results are reported in Supplementary Appendix Table A8 and confirm that the effect of exposure is not significantly different from zero in 95 per cent of the cases.

Fourth, we conduct an event study to check for significant pre- and post-electoral trends in the spatial distribution of immigrants by nationality of origin (further details are in Supplementary Appendix, Section 7). Results (available on request) suggest that only for ten out of ninety-two nationalities, some pre-trends or post-trends are statistically significant at a 10 per cent level. For only two, pre- or post-electoral trends are statistically significant at a 5 per cent level. Therefore pre- or post-electoral trends of immigrant nationalities in response to the local political shocks do not represent a severe threat to our identification strategy.

Fifth, we compute an additional falsification exercise based on a “placebo” index and perform a permutation test to check the validity of our results (Wing and Marier 2014; Carrieri, Madio, and Principe 2019). We first compute a “fake” exposure index using a random distribution of resident immigrants across Italian municipalities, which ranges from the minimum and maximum observed values each year and for each nationality. Supplementary Appendix Table A9 shows that the effect of this fake exposure index is never statistically significant. We then implement a permutation test using this index. This test is based on a Monte Carlo simulation, which allows us to simulate the effect of unreal immigration exposure (Robert and Casella 2004). We performed 1,000 simulations of all the models in Table 2 using the artificial exposure index and stored the estimated coefficients. Since the distribution of resident immigrants is randomly generated, the estimated effect of fake exposure should be, on average, equal to zero. Supplementary Appendix Figure A12 shows the plot of the estimated coefficients for each outcome (and time window) of interest. In all cases, as expected, the distribution is centered around zero. At the same time, the coefficient of the “true” exposure index always lies outside or at the very extreme tail of the distribution.

Another endogeneity concern arises from the positive correlation between the exposure index and a municipality’s capacity to receive migrants from specific countries.¹⁹ This capacity may be influenced by a municipality’s historical political leaning, as left-leaning municipalities are more likely to devote resources to migrant reception and exhibit higher exposure indices. At the same time, left-leaning voters in these municipalities, while generally supportive of immigration, may express dissatisfaction with current immigration policies through higher abstention rates or shifts away from traditional left-wing parties, which could amplify treatment effects in these areas. However, the relationship between political leaning and the treatment effect is nuanced: while left-wing municipalities may attract more migrants, their voters might respond less strongly to migration shocks in terms of populist and anti-immigration vote increases, potentially leading to a downward bias in our estimates. To examine whether migrants of specific nationalities tend to cluster in municipalities with a greater capacity to attract them, we regress the share of immigrants for each nationality on the main political leaning of a municipality. Results show that being a municipality characterized by a specific political leaning (“leftwing” in our analysis) significantly predicts the presence of only seven out of ninety-one nationalities landed on Italian coasts.²⁰ Re-running the main regressions excluding those nationalities from the index computation produces similar results, thereby excluding this potential source of endogeneity (Supplementary Appendix Table A10).²¹

4.3 Heterogeneity tests

4.3.1 Socio-demographic

We perform a series of socio-demographic heterogeneity tests to provide further evidence on the main channels at work. To sum up, we find that the effect of exposure to arrivals is higher in municipalities: (1) with high taxable income per capita (perhaps because individuals respond to immigration in light of their economic concerns; Section 8.1, Supplementary Appendix Table A11) and (2) with a higher share of children, that is, where competition for local welfare might be higher (Section 8.1, Supplementary Appendix Table A12).

4.3.2 Spatial

The literature on immigration, political attitudes, and electoral outcomes suggests that municipality size matters and that the interaction between natives and immigrants may differ in small versus large cities. We employ the Department for Economic Development and Cohesion classification of the municipalities based on a combination of the presence of essential services such as education, health, and mobility that first identifies a network of municipalities or their aggregations as “service supply centers” and, subsequently, around them, the areas characterized by different levels of spatial peripherality. The first group of municipalities is defined as poles if they offer at least one upper secondary school, one technical and vocational institute, one hospital, and a railway station. The other remaining municipalities are divided into four categories—peri-urban areas, intermediate areas, peripheral areas, and ultra-peripheral areas—based on travel distance from the poles (20, between 21 and 40, between 41 and 75 and more than 75 min away from the nearest pole, respectively). We use poles as a proxy of metropolitan municipalities and aggregate the other categories as peripheral areas. Figure 3a summarizes the results showing that the immigration inflows increase the votes obtained by far-right parties, especially in non-metropolitan municipalities while leaving metropolitan cities mostly unaffected.²² This result is in line with Barone et al. (2016), Dustmann, Vasiljeva, and Damm (2019), who suggest that individuals residing in cities tend to have more positive attitudes toward refugees and immigrants compared to those in rural areas, as contact serves as a mechanism for reducing prejudice.

The literature has also stressed that the native-immigrant contest for jobs should be tougher for unskilled native workers (Borjas 2003; Mayda 2006).²³ Relying on census data, we use the ratio of migrants to natives who are active in the labor force at the municipal level as a proxy for employment competition. We, therefore, split the sample below and above the median value of the aforementioned ratio and run a heterogeneity analysis based on the distribution of the respective variable observed in 2011. In line with the previous literature (Barone et al., 2016), Fig. 3b suggests that our main findings are indeed driven by cities with higher employment competition.

As an additional spatial heterogeneity analysis, we compare the political effects of perceived migration exposure in cities hosting a migrant reception center (SPRAR) with those that do not have such a center. The results, detailed in Fig. 3c, indicate that our main findings are predominantly driven by places without reception centers. This further reinforces the habituation/contact hypothesis, suggesting that hosting refugees and asylum seekers through a reception system managed by local governments may foster positive interactions between natives and immigrants. These interactions, in turn, could undermine the electoral performance of far-right and anti-immigrant parties (Steinmayr 2021; Gamalerio et al., 2023).

Finally, our main results may be influenced by two common factors: the expectation that new arrivals represent a threat and the increased salience of local foreign-born communities. Unfortunately, our data do not allow us to distinguish whether the effect of exposure to arrivals primarily stems from the perceived threat of new refugee inflows or the salience of resident immigrants. To partially disentangle these effects, we examine whether the impact of exposure decreases with distance from the main immigration ports, which could help us understand whether the ‘threat-of-new-immigrants’ mechanism is at play. For each boat landing to which a municipality is exposed, we calculate the distance in kilometers of that municipality from the ports of arrival. Then, we average this measure over all the landings experienced by the municipality and compute the median distance. Subsequently, we rerun our baseline estimates, splitting the sample into municipalities below and above the median distances from the ports. The results, summarized in Fig. 3d, indicate that the effect is more pronounced in municipalities closer to the ports.²⁴ Assuming that the salience of local immigrants with the same nationality as the incoming refugees does not vary by distance, these findings suggest, again, that the threat of future arrivals is likely the main mechanism behind our results.

4.4 Robustness checks and actual migration patterns

We run additional tests for the robustness of our results. We discuss them in Supplementary Appendix, Section 8. Results are robust to different definitions of populist parties (Section 8.2, Supplementary Appendix Table A13) and different aggregations of refugees’ nationality to deal with the possibility that citizens do not have precise information on the nationality of the incoming refugees (Section 8.3, Supplementary Appendix Table A14). We aggregated both landed migrants’ and resident migrants’ nationalities over twelve world regions, following a modified version of the classification proposed by the United Nations Statistical Division (UNSD).²⁵ While the UNSD lists twenty-three world regions, adjustments were made to improve representativity and align with our dataset. Specifically, we aggregated the Caribbean region into Central America, combined all European regions into a single region, and attributed Southeast Asian countries to either the South or East Asia regions. Furthermore, as our dataset contained no migrants declaring origins in Oceanian regions, Polynesia, or Northern America, these regions were excluded from our list. Finally, we did not use the “Not Classified” option in the UNSD classification, leaving us with a total of twelve regions.²⁶ We then reconstruct our exposure index using world region origins instead of nationalities to match landed and resident immigrants. Note that this robustness check also allows for relaxing the implicit assumption of “super-rational” voters, that is, voters who perfectly understand that (1) the local shares of migrants from a given country can increase when refugees from that country arrive at Italian ports and (2) the size of this increase depends on the number of local migrants from that country relative to migrants from other countries.

Regarding the role of mass media, we posit that the latter may have influenced expectations about immigration by representing refugee landings as a crisis. This portrayal might have led individuals to believe that new landings would eventually increase the local immigrant population. These expectations could be stronger in municipalities with a larger share of immigrants from the same nationality as the prospective incomers, due to the attractive power of local networks. Yet, refugees are not allowed to move to the municipalities where citizens mostly expect them to go upon arrival. Here, we offer suggestive evidence showing that immigrants tend to cluster in cities where they can rely on a pre-existing network. However, an increase in the local immigrant population is not associated with the arrival of new refugees (see Section 8.4 in Supplementary Appendix). This evidence partially supports our main hypothesis: the political effects of immigration exposure stem from misperceptions about future immigration (i.e. incorrect beliefs that new refugee inflows will increase the local immigrant population) rather than actual future immigration. More formal tests for this hypothesis are provided in the next section.

5. Mechanisms: the role of media

5.1 Misperceptions, salience, and perceived threat

Figure 2 describes the correlation between the occurrence of immigration-related tweets and refugee landings, particularly noticeable during local election periods. Yet, trends diverged during the 2018 elections, suggesting that the perceived importance of immigration often does not align with actual inflows. In the following analysis, we examine econometrically whether the salience of immigration increases in proximity to elections and whether the expectation of refugee arrivals further amplifies the topic’s relevance. More specifically, we carry out an event study to test for significant electoral trends in the frequency of tweets containing the words “refugee/s” or “immigrant/s” (and similar words referring to the topic).²⁷ We estimate the following model:

where $Tp,m=ImmTweetsp,mTweetsp,m$ is the number of immigration-related tweets in province $p$ and month $m$ over the total number of tweets posted in province $p$ and month $m$⁠; $αp$ is a time-invariant province-specific effect capturing the socio-economic environment; $δy$ is a set of year dummies capturing common macro-level trends; $ϵp,m$ is an error term; the vector of additional control variables $Xp,y$ includes the yearly share of individuals aged 15–64 years over the total population at the province level and the share of immigrants residing in the voting municipalities nested into our sample provinces; $θEp,mτ$ is a set of dummies for each month $m$ equal to one if a local election is scheduled in province $p$ at month $m$⁠, with $τ=−5,5$⁠, that is, 5 months before and after the elections (the period before election is the omitted category; i.e. $t−1$ in the plot, $τ=−1$  Repetto 2018).²⁸ We are interested in $θd$⁠, which captures significant pre- or post-election trends of immigration salience, that is, whether the importance of immigration increases or decreases $τ$ periods before and after the elections. We estimate Equation (3) through OLS fixed-effects regression, clustering standard errors at the province level.

To understand whether changes in immigration salience mirror the actual arrivals of refugees, we carry out an event study similar to the previous one, but considering refugees landed in time $m$ as the dependent variable. In other terms, we estimate the following model:

where the dependent variable is now $Lm=ihs(TotLandingsm)$⁠, where TotLandings is the inverse hyperbolic sine transformation of the total number of refugees landed in month $m$⁠. Since landings are at the national level and vary only by month, we rely only on time variation; hence, we now estimate Equation (4) through OLS clustering by year and including season-of-the-year fixed effects to net out seasonality effects of migrants’ arrivals (⁠$δs$⁠).

Results show the share of tweets about immigrants increases significantly during the month of the election and remains consistently higher for the three subsequent months compared to the month before the election (Fig. 4a). This increase seems to fade away after 5 months. Conversely, elections seem not to impact boat landings. Indeed, the number of boat landings does not change before and after the local elections, highlighting that boat landings do not follow the electoral dynamics at the municipal level (Fig. 4b).

These results, jointly considered, suggest that the increased salience of immigration driven by the local elections does not mirror the actual dynamic of boat landings, further highlighting the mismatch between voters’ perceptions and real statistics on immigration outlined above.

To assess whether the electoral increase in the salience of immigration is more significant in provinces where citizens are mostly exposed to the potential arrival of refugees, we compare pre- and post-electoral trends of immigration-related tweets by high versus low values of the exposure index. We aggregate the exposure index by province and year and create a dummy variable equal to one if the exposure index value of a given province lies above the median exposure index of the relevant macro-area, and zero otherwise. We re-run the first event study and estimate Equation (3) interacting the trend dummies $Ep,m+d$ with an indicator equal to one if province $p$⁠, each year, is above the median.²⁹

Figure 4c reports the results and shows that provinces with a high exposure index produce a higher volume of tweets on immigration during the election month and immediately after, yet the effect fades away after 5 months. Overall, immigration salience in the electoral period is mainly driven by those provinces where the expected inflow of immigrants is higher.³⁰

Since the exposure index increases for municipalities with a larger share of immigrants having the same nationalities as the incoming refugees, the estimated effects may capture actual future immigration (or a plausible guess given the available information) and (mis)perceived immigration. In other words, our results could suggest that voters are forward-looking or have biased perceptions of immigration.³¹ Here, we run a test that could help to disentangle these two mechanisms, interacting the Twitter data on immigration salience with the exposure index. If the political effects of exposure were driven by provinces where most people (emotionally) reacted to it by tweeting more before the election, the perceived threat would most likely be the channel. In that case, keeping expectations of future arrivals (exposure) constant across municipalities, political outcomes should be most affected in municipalities where citizens were more emotionally sensitive to the arrivals (salience).

We re-estimate the baseline model of Table 2 by running a heterogeneity analysis to investigate whether provinces with a higher frequency of tweets about immigrants drive our results. We do so by interacting our exposure variable with a dummy variable, taking value one if, in a given province–year, the share of immigration-related tweets is higher than the national median. Figure 5a reports the results and highlights heterogeneous effects of exposure by immigration salience, with statistically significant effects only where immigration-related tweets are most frequent. Notwithstanding the Twitter data’s caveats, this result suggests that, in the face of a similar (potential) inflow of refugees, support for populist and anti-immigration parties increased mainly where arrivals were perceived as a critical issue.

5.2 Twitter sentiment toward immigrants

In addition to the analysis of immigration salience, we conduct a more thorough examination of the sentiment expressed in our tweets by employing two distinct sentiment analysis techniques.

First, we classify tweets into positive, negative, or neutral categories. Second, we delve deeper into the specific emotions conveyed in the tweets. To this purpose, we use the “FEEL-IT: Emotion and Sentiment Classification for the Italian Language” model developed by Bianchi et al. (2021). This model assigns to each tweet one of the following emotions: anger, fear, joy, and sadness. By leveraging this dataset, we conduct both emotion and sentiment analyses on our corpus of tweets. We run a heterogeneity analysis comparing provinces that exhibit an above-average frequency of negative tweets. Finally, we disaggregate negative sentiment by emotion to discern whether our findings are influenced by provinces with a higher prevalence of anger, sadness, or fear in their tweets.

Figure 5b and c summarizes the results and suggests that our main findings are primarily driven by those places that exhibit more negative sentiments toward immigrants (specifically, anger and fear). In these provinces, our exposure index increases the vote shares for anti-immigrant, populist, and Northern League parties more than in provinces where immigration is not discussed negatively. Conversely, we also observe a decrease in voter turnout in these areas, supporting our intuition that left-wing individuals “punish” left-wing parties by reducing their propensity to vote.

5.3 Broadband access

To further explore the role of mass media in the relationship between expected immigration and electoral outcomes, we look at broadband diffusion, which is less subject to citizens’ selection into social media. Suppose voters have broad access to information on landings (and the nationality distribution of incoming refugees) through social media. In that case, broadband availability should amplify the exposure effect.

To test this hypothesis, we use data on the share of households without broadband connection and with a fast Internet speed (higher than 2 Mbps/second) as provided by AGICOM. We leverage the cross-sectional variation in broadband diffusion, focusing on data available only for the year 2018.

Results in Fig. 6 indicate that the political effects of exposure are primarily driven by municipalities with better access to the Internet. Specifically, these effects are more pronounced in municipalities where the share of households with broadband access (Fig. 6a) and fast (Fig. 6b) internet falls above the macro-area median. In contrast, for municipalities where broadband access is more limited, the effects are either not statistically significant or weakly so, with smaller magnitude.

All these findings bolster our hypothesized mechanism. The arrival of refugees impacted voter behavior via media coverage, which intensified the salience of immigration issues. Consequently, voters formed biased judgments regarding the severity of immigration and developed negative sentiments, which, regardless of actual interactions with immigrants, influenced their voting behavior.

6. Conclusions

This article explores the political effects of expected immigration in Italy. Our identification strategy rests on the exogenous variation in the nationality of refugees approaching the Italian ports from 2010 to 2018. We construct an index of exposure to arrivals that, each year, varies at the intensive margin across municipalities, with more (less) exposed cities having a larger (lower) cluster of regular immigrants with the same nationality of refugees approaching the Italian coasts before the elections. Since we also control for the local share of regular immigrants and the number of refugees in reception centers, our estimates capture the additional role that the arrival episodes, widely discussed in the media before the elections, played on voting behavior.

Our results show that potential exposure to arrivals mildly decreases turnout, while significantly increasing protest votes and support for extreme-right, populist, and anti-immigration parties. These findings are consistent with previous empirical evidence (Barone et al., 2016) showing that voters dissatisfied with mainstream party handling of immigration issues may abstain from voting or express discontent through protest or by supporting anti-establishment parties advocating immigration restrictions. We also find that the salience of immigration does not strictly mirror actual refugee arrivals but follows the electoral cycle. Finally, since the political effects of arrivals are larger where immigration salience is high, the upsurge of populist and anti-immigration parties can most likely be driven by voters perceiving future arrivals as a critical policy issue.

Further tests provide additional insights into the role played by the media in fuelling misperceptions. We find that the impact of expected immigration is driven by voters with access to fast Internet connections and by provinces where immigration is perceived as a key issue and discussed with a negative tone.

Moreover, the effects of exposure decrease with distance from the main immigration ports, suggesting that our findings could be driven by the threat of new arrivals. Finally, while immigration salience and voting behavior are affected by refugee landings, local immigration is not. Thus, biased beliefs—rather than rational expectations—about future immigration may facilitate the upsurge of anti-immigration and populist parties.

Our findings suggest that as immigration becomes central in electoral disputes, expected immigration starts playing a key role in voting behavior, alongside heightened perceptions of insecurity and socio-economic costs associated with hosting refugees. Representation of immigration as a permanent crisis in the media, even though this is not always the case, influences beliefs about new inflows of immigrants and hence boosts voters’ disappointment toward mainstream parties. By losing trust in the latter, citizens reduce political participation and increase protest or populist votes (Barone et al. 2016; Guiso et al. 2017, 2019; Algan et al. 2018). Hence, anti-immigration campaigns based on the severity of refugee arrivals and alternative policies against the alleged refugee crisis might effectively raise political consensus for far-right, populist parties, especially in the cities where refugees are more expected to arrive.

Footnotes

Supplementary data

Supplementary data is available at Journal of Economic Geography online.

Acknowledgements

We thank G. Barone, S. Bertoli, E. Bracco, V. Bove, V. Carrieri, G. De Blasio, J. Dennison, M. De Paola, A. D’Ignazio, M. Gamalerio, G. Hanson, J.S. Goerlach, M. Le Moglie, E. Levi, A. Martinangeli, J. Morales, M. Piovesan, F. Revelli, A. Venturini, and L. Windsteiger for valuable comments and suggestions. The Statistical Office of the Italian Ministry of Interior, Agicom, and Infratel are gratefully acknowledged for data provision. We also thank V. Perri for his invaluable assistance in collecting Twitter data.

Conflict of interest statement: The authors declare that they have no conflicts of interest to disclose.

Data availability

Programs and data are available online as supplementary materials on the same webpage where this article is published.

References