1 Introduction

Hate crimes are growing, together with their socioeconomic and emotional costs. This makes hate a growing source of concern for institutions, thus stimulating research to understand its risk factors. In this paper, we assess the relevance of the geography of refugees as a trigger for hate, using Italy as a case study.

The link between hate and refugees intertwines with the social impact of refugee-hosting policies. This issue is currently in the spotlight, given the growing number of refugees fleeing from geopolitical hotspots, such as Ukraine, Afghanistan and several African countries affected by food shortage (also as consequence of the Ukrainian War) (UNHCR 2021).

The importance of studying the link between the geography of refugee reception and hate is supported by different contributions in the literature. In fact, scholars widely acknowledge that hate stems from local features capable of triggering a perception of threat from outside forces among members of a community (Baumeister & Vohs 2004; Gerstenfeld 2017; Glaeser 2005; Perry 2001). Refugees could belong to these outside forces, since they might be seen as a cultural threat for the receiving community (Denti 2022; Hainmueller & Hopkins 2014; Hopkins 2010).

Yet, evidence of the impact of local exposure to refugees on hate is scarce and largely descriptive (Dancygier et al. 2022; Liebe & Schwitter 2021; Piatkowska et al. 2016). Also, most quantitative research on hate and refugees measures hate through police records. This choice is largely due to the lack of different micro-regional data, but it implies non-negligible measurement biases due to the severe underreporting of hate crimes (European Union Agency for Fundamental Rights 2021; Pezzella et al. 2019). Police records do not represent a reliable proxy for hate (Piatkowska et al. 2016), given that 70% of hate victims in Europe do not report the crime to the police (European Union Agency for Fundamental Rights 2021). In this paper, we contribute to build this evidence by alleviating concern on measurement bias, since we exploit micro-regional data on hate from multiple sources including newspapers and NGOs. Also, using a novel panel database for Italian provinces (NUTS3) between 2009 and 2017,Footnote 1 we prove a causal link between hate and local exposure to refugees.

Our estimates show that, in Italy, places with larger refugee reception capacity have high incidence of hate events. Between 2009 and 2017, a 1% increase in the local refugee hosting capacity relates to 8% increase in hate. Using a Bartik-type instrumental variable estimation and controlling for the confounding effect of other factors, including the role of media, we show that this evidence might indeed be causal.

Our investigation considers Italy, for its high volume of hate events (with a 31% average annual growth between 2009 and 2017) (Lunaria 2019; OSCE-ODIHR 2019). Evidence also shows that in Italy hate is spatially heterogeneous (Denti et al. 2023) and influenced by local features (Denti & Faggian 2021). Italy also plays a key role in refugee migration, since its geographic location makes the country a major first-stop for refugees, either for permanent or provisional settlementFootnote 2 (UNHCR 2021).

The geography of Italian refugee reception between 2009 and 2017 included two types of hosting centers: (1) centers belonging to “The Protection System for Asylum Seekers and Refugees” (SPRAR) and (2) centers belonging to “The Extraordinary Reception Centers” (CAS). SPRAR is the long-standing Italian refugee reception policy, while CAS was introduced in 2014 to cover emergency hubs. Combining the hosting capacity of both SPRAR and CAS, it is possible to get a reliable measure of the local capacity for refugee reception.

Our results make several contributions. First, they contribute to the literature on the link between migration and resentment, by adding causal evidence that exposure to refugees relates to more hate. Also, our estimations rely on a measure of hate that alleviates concerns of measurement bias. Second, this paper complements existing evidence on behavioral responses to refugees, adding to research on refugees and voting preferences. This is relevant given that the radicalization of anti-immigrant behaviors has a limited transposition in voting preferences (Minkenberg 2013). Also, evidence on the refugees-voting nexus is mixed (Gessler et al. 2021; Rickardsson 2021; Steinmayr 2021), calling for more investigation. Third, our results contribute to the evidence supporting the prominent role of geography in explaining hate (Denti 2022; Denti & Faggian 2021; Medina et al. 2018). Forth, they have key policy implications concerning refugee hosting policy as well as hate prevention.

2 Background: hate, refugees and the Italian refugee reception system

2.1 Hate and refugees

Criminologists define hate as violent oppressive behaviors targeting victims on the basis of the perpetrator’s prejudice against the actual or perceived status of the target (Craig 2002). Hate also affects the wider community acting as a “message” (Iganski 2008), i.e., creating an “in terrorem” effect that reaches all community members, prompting social tension and fear (OSCE-ODIHR 2009) and generating socioeconomic costs (Glaeser 2005). The role of context, rather than personality traits, in influencing hate is becoming increasingly clear (Anderson et al. 2020; Denti et al. 2023; Denti & Faggian 2021; Green et al. 2001). In the US, for example, no more than 5% of hate crimes are committed by members of hate groups, while most of them are perpetrated by ordinary people with no sadistic or extremely biased personalities (Gerstenfeld 2017; Hall 2013).

Hate is “not abnormal; rather it is a normal (albeit extreme) expression of the biases that are diffused throughout the culture and history in which it is embedded” (Perry 2001). Local factors escalate prejudices into hate (CPS 2018; Hall 2013), where prejudices are a structural part of the cultural norms and beliefs of each local community (Huggins & Thompson 2015, 2019).

An established body of work supports exposure to migrants as relevant for intolerant attitudes. Actual changes in the local social fabric due to sudden inflow of migrants might trigger the perception of threats to the established local identity (Hainmueller & Hopkins 2014; Newman 2013; Perry 2001). Evidence on the causal effect of exposure to migrants on hate is still scarce, mainly due to the lack of micro-regional data on hate. Recent descriptive findings show that sudden inflow of migrants relate to more hate (Dancygier et al. 2022; Liebe & Schwitter 2021; Piatkowska et al. 2016). Features such as mate competition (Dancygier et al. 2022), job competition (Liebe & Schwitter 2021; Piatkowska et al. 2016) and anti-immigrant narratives (Liebe & Schwitter 2021) are risk factors influencing hate against refugees. Evidence on the size of foreign population is mixed (Liebe & Schwitter 2021; Wagner et al. 2020).

Notably, this evidence is mainly correlational (Dancygier et al. 2022; Liebe & Schwitter 2021; Piatkowska et al. 2016; Wagner et al. 2020), with little information on causality patterns. The latter is needed since it matters for effective policy design (Angrist & Pischke 2014). Some causal evidence about migrants and hate is available for England, with a focus on racist school bullying (Denti 2022), and Germany (Entorf & Lange 2023). In both cases, there is support for the fact that sudden exposure to previously unknown ethnic groups causally increases hate. However, causal evidence on Germany uses police records as proxy for hate with possible measurement biases due to underreporting (Piatkowska et al. 2016).Footnote 3 Causal evidence for England exploits victimization data, but in relationship with immigrant flows overall, rather than refugees.

Close-but-different evidence explores the link between immigrants/refugees and intolerant attitudes which manifest through voting. Some contributions show that immigrants/refugees are perceived as threat that gets channeled in voting for parties with anti-immigrant platforms (Dustmann et al. 2019; Gessler et al. 2021; Halla et al. 2017; Roupakias & Chletsos 2020). Others claim that exposure to migrants is beneficial in reducing support for anti-immigrant parties (Gamalerio et al. 2022; Steinmayr 2021). There is also evidence of a non-significant influence of refugees on voting (Rickardsson 2021). This mixed evidence might suggest that there are country-specific patterns, and this calls for further research on the association between intolerant attitudes and refugees. Also, voting preferences do not completely reveal anti-immigrant attitudes, especially in terms of radicalization of these attitudes (Minkenberg 2013). Hence, it appears relevant to understand if the same exposure determines more radicalized actions, such as hate. This would contribute to a better understanding of the immigrant-resentment nexus.

Hence, it seems that there is room to contribute to the causal evidence about the refugees/hate nexus, by using comprehensive measures for hate. In this paper, we address this issue focusing on Italy, since evidence on this country is limited notwithstanding the high incidence of hate (OSCE-ODHIR 2019) and Italy being the major European first-stop for refugees (UNHCR 2022). Further, we can account for potential bias due to underreporting by exploiting a database for Italian hate events, which includes cross-checking from multiple sources.

2.2 Refugee reception in Italy between 2009 and 2017

The Italian reception system for refugees was shaped as a multistage reception system by several regulations between 2002 and 2008.Footnote 4 First-stage reception was centralized and provided by few large governmental facilitiesFootnote 5 whose tasks were first aid, identification and prevention of illegal entry. These activities occurred soon after disembarkation or rescue at sea in few sites with no dispersion of refugees across the country. Second-stage reception was decentralized and implied transferring refugees to the SPRAR system (“Sistema di Protezione per Richiedenti Asilo e Rifugiati,” i.e., protection system for asylum seekers and refugees) where they stayed until a final decision on their application was made (SPRAR 2014). The SPRAR system was made by decentralized small reception structures set up voluntarily by local authorities and refugees were placed in SPRAR all over the territory, depending on the availability of places.Footnote 6

In 2014, SPRAR was complemented with CAS (“Centri Accoglienza Straordinaria,” i.e., extraordinary reception centers),Footnote 7 facilities aimed at providing temporary and extraordinary accommodation to the growing number of refugees caused by the Syrian War. CAS was conceived to be opened in the event of substantial and sudden arrivals of refugees who cannot be accommodated through the ordinary SPRAR system. Like SPRAR, CAS system is made of decentralized reception structures.

In the 2010s, the growing numbers of refugees reaching Italy implied an increase in: (1) the total number of refugee centers (SPRAR and CAS) (Fig. 1a), (2) their hosting capacity (Fig. 1a) and (3) the number of provinces having at least one refugee hosting center (Fig. 1b).

Fig. 1
figure 1

The geography of refugee hosting centers (SPRAR and CAS) in Italy (2009–2017)

Full size image

The refugee reception system has relevant socioeconomic effects in Italy. Evidence shows that it influences social capital of the receiving communities (Fratesi et al. 2019). Other works find significant effects of the Italian geography refugee on voting, although with mixed results (Bratti et al. 2020; Campo et al. 2021; Gamalerio et al. 2022). Hence, it seems interesting assessing its effect on the Italian geography of hate.

3 Data and descriptive statistics

Micro-regional data on hate come from a database designed and maintained by Lunaria, an Italian non-profit organization. This database contains information on hate events, which are property damage, threats, assault, murder motivated by prejudice against disempowered groups (Lunaria 2017; OSCE-ODHIR 2017). Lunaria database is referred to by international institutions, including OECD, OSCE and the European Commission (Siragusa et al. 2020), and national institutions, including the National Institute of Statistics and the National Anti-Discrimination Office.Footnote 8 Information is collected from Italian newspapers and NGOs and verified through cross-checking from 2009. By considering multiple sources, the Lunaria database alleviates concerns on measurement bias arising when hate is measured relying only on police records, which suffer from high underreporting. For each hate event, the database gives the date and the location, allowing to build the yearly geography of hate across Italian NUTS3. We choose NUTS3 as baseline geography for several reasons. First, a relevant share of hate events involves people from several municipalities.Footnote 9 Second, NUTS3 level evidence allows comparability with extant evidence on hate and refugees, which is mainly at this spatial level (Entorf & Lange 2023; Liebe & Schwitter 2021) or larger (Piatkowska et al. 2016). Third, annual data on features that we must acknowledge as relevant confounders are available at NUTS3 level or higher. These features emerged from previous studies as risk factors for hate and they include unemployment (Anderson et al. 2020; Piatkowska et al. 2016), literacy (Finseraas et al. 2018) and media exposure (Gilliam et al. 2002).Footnote 10

Figure 2 shows summary statistics. Between 2009 and 2017, hate events were rising (Fig. 2a), at the same time being spatially dispersed (Fig. 2b I–II). Figure 2a–c compare trends in hate and refugees’ inflow for Italy between 2009 and 2017, showing clear overlapping trends between 2009 and 2010 and from 2013 to 2017 (correlation coefficient = 0.52).

Fig. 2
figure 2

Trend of hate and refugees in Italy between 2009 and 2017

Full size image

Data on the local hosting capacity for refugees come from three sources. The geography of SPRAR hosting capacity is mapped using the Italian Ministry of Interior SPRAR database.Footnote 11From this database, we collected yearly information on the number and the capacity of SPRAR hosting centers in each Italian province (NUTS3) between 2009 and 2017. For CAS, we used two databases. The Openpolis CAS database gives the geography of CAS centers using information on the mandatory public bid issued by each provincial Government office (“Prefetture”) to open a CAS (Openpolis 2018). The OpenPolis-Centrid’Italia CAS database details the hosting capacity of each CAS (Centri d’Italia—Openpolis 2022). CAS data cover the period 2014–2017 as CAS was introduced in 2014.Footnote 12 Combining these sources, we design a yearly measure for the refugee hosting capacity at the NUTS3 level, which is given by the cumulative number of available beds from SPRAR and CAS. Figures show that the refugee hosting capacity is a good proxy for the actual presence of refugees between 2009 and 2017. SPRAR were fully occupied (SPRAR-CITTALIA 2017; SPRAR 2014) and also more than 80% of CAS capacity between 2014 and 2017 was fully occupied (Centri d’Italia–Openpolis 2022; Openpolis 2018).Footnote 13 These figures follow from the Italian hosting system, which introduced CAS to open new accommodations whenever refugees cannot be quartered in the existing capacity.

We control for local media exposure, given the acknowledged influence of media on people’s perception about migrants (Becker et al. 2017; Eberl et al. 2018). Our measures for media exposure rely on data on the number of newspapers sold at NUTS3 level between 2009 and 2017 using the ASDFootnote 14 database. ASD is a company funded by publishers to collect data about the press published in Italy. Then, referring to existing works on Italian media (Cusumano & Bell 2021; Treccani 2022; Urso 2018), we consider data on newspapers classified respectively as progressive and conservative to design a measure for exposure to progressive media and a measure for exposure to conservative media.Footnote 15

Other controls are introduced to account for previous research on the determinants of anti-immigrant attitudes. We consider foreign resident population (Barone et al. 2016), unemployment (Anderson et al. 2020), literacy (Denti & Faggian 2021; Finseraas et al. 2018), crime (Phillips & Bowling 2020), population size. Controlling for undocumented migrants (i.e., migrants that are neither legally registered nor inside the asylum seeker process) is not possible due to data unavailability, however existing work shows that the Italian geographies of regular and irregular migrants are strongly related (Bianchi et al. 2012). Detailed information on data sources and descriptive statistics are in the Appendix (Tables 5, 6, 7, 8).

4 Empirical strategy

We assess the association between hate and exposure to refugees through a simple least-squares model

$${\text{HATE}}_{{it}} = \alpha + \beta {\text{Refugees}}_{{it}} + X_{{it}} \gamma ^{\prime } + \lambda _{i} + \tau _{t} + \mu _{{it}} + \varepsilon _{{it}}$$
(1)

where \({\mathrm{HATE}}_{it}\) is the incidence of hate (number of hate events per 100,000 inhabitants in logs) in province (NUTS3) \(i\) at time \(t\). \({\mathrm{Refugees}}_{it}\) is the log of the provincial capacity of refugee hosting centers, measured combining reception capacity from SPRAR and CAS in per capita terms. \({X}_{it}\) is a set of control variables. \({\lambda }_{i}\) is a province fixed effect, which should capture most part of remaining unobserved spatial heterogeneity. \({\tau }_{t}\) is a year fixed effect and \({\mu }_{it}\) a linear region-specific time trend to control for exogenous and regional factors. Control variables in \({X}_{it}\) are the size of NUTS3 population, the share of foreign resident population, unemployment, adult literacy, crime rate, exposure to conservative (progressive) media.

Our analysis faces an identification challenge since the location of SPRAR refugee centers might not be random, given that their opening depends on a voluntary choice by the local authority. Unobservables could jointly affect the decision to open a SPRAR center and hate events. Reverse causality could deter places with many hate events from increasing refugee reception. We address this challenge using an instrumental variable strategy (IV) in which we predict the current local capacity of hosting refugees using extant information on local residential support for disempowered and vulnerable groups. Before 2009 refugee, inflow was negligible and local residential support for disempowered and vulnerable groups was entirely dedicated to other groups: homeless people, orphans, people with addictions, people with disabilities, victims of violence, elderly.Footnote 16

Operationally, our IV is a Bartik shift-share using as “shares” the NUTS3 pre-existing local residential services for vulnerable and disempowered groups and as “shift” the national growth in the provision of these residential services. The shift-share IV is summarized by Eq. (2) (Goldsmith-Pinkham et al. 2020; Roupakias & Chletsos 2020)

$${z}_{it}={{s}_{i}g}_{t}^{\mathrm{IT}}$$
(2)

where \({s}_{i}\) is the share of pre-existing residential support per capita in NUTS3 i and \({g}_{t}^{\mathrm{IT}}\) measures the national annual growth in residential support at time \(t\), with \(t\in [2009, 2017]\). Pre-existing residential support, \({s}_{i}\), combines the stock of buildings devoted to residential support with the local public budget allocated to the support services in those buildings.

The pertinence of pre-existing residential support for disempowered and vulnerable groups as instrument for current refugee reception is detailed by Steinmayer (2021). Local authorities mainly host refugees in pre-existing buildings that can accommodate large groups. Most of these buildings were built in the past to provide residential support to other disempowered and vulnerable groups: homeless people, orphans, people with addictions, people with disabilities, victims of violence, elderly.Footnote 17 Differently from Steinmayer (2021), we weight this building stock with the extant budget spent for the services provided in those buildings. In this way, we can account for the pre-existing local endowment in skills and knowledge about supporting vulnerable groups. This aspect is relevant for the Italian refugee reception system because this system requires a local bundle of housing and integration services.Footnote 18

By freezing the geography of residential support before the 2010s refugee waves, we alleviate sorting concerns and reverse causality (Boustan et al. 2013). In the robustness check, we will perform falsification tests to alleviate concerns on unobservables.

An alternative identification strategy could have exploited the different timeline of CAS refugee center openings, since CAS should result from a dispersal policy that assign refugees to province based on resident population (around 2.5 reception places per 1000 residents according to the Refugees National Allocation Plan). Actual figures show a different pattern. For instance, in 2015 refugee hosting capacity in the province (NUTS3) of Agrigento and Macerata amounted to 3.47 places per 1000 residents. So, both provinces were allocated more refugees than the Refugees National Allocation Plan cap. In the same year, hosting capacity in other provinces was way lower: 0.08 per 1000 residents in Como, 0.32 in Imperia, 0.13 in Campobasso. Hence, alongside provinces that were hosting more than required, others were hosting less. The same large variation across provinces (NUTS3) can be detected in 2014, 2016 and 2017. These figures threaten the exogeneity of the CAS dispersal policy, suggesting that existing local uncodified views about refugees/foreigners, and lobbying between different governmental levels (municipal, regional, national) are unobservables threats to exogeneity also for CAS.

Having identified our IV, we use Eqs. (1) and (2) to estimate whether there is a causal relationship between refugee hosting centers and hate using a Two-Stage Least Square with Instrumental Variable (2SLS-IV) model for panel data with time and space-fixed effects.

5 Results

5.1 Baseline results

We estimate Eq. (1) via a Two-Way Fixed Effect model with 104 NUTS3Footnote 19 yearly observations between 2009 and 2017. The results in Table 1 show that an increase in refugee hosting capacity relates to more hate, and this holds in all model specifications (for detailed control variables estimates see Appendix, Table 8).

Table 1 Two-way fixed effect panel model estimates and sensitivity tests for the effect of hosting refugees on hate events in the Italian NUTS3 areas between 2009 and 2017
Full size table

The size of the estimated effect is non-negligible. Starting from the baseline specification (column 1), if the local capacity for hosting refugee increases by 1%, hate events increase by more than 8.5%. These findings hold to the inclusion of potential confounders as summarized by columns 2–5 in Table 1. Estimates in column 2 show that hosting refugees increases hate when we introduce economic controls. Column 3 corroborates this finding by showing that the effect of hosting refugees does not change when we include controls for media and column 4 further adds crime among controls. Finally, column 5 adds a region time trend and still confirms results.Footnote 20 Hence, estimates from the Two-Way Fixed Effect model support a relevant effect of hosting refugees on hate.

As for our control variables, we check whether the ethnic composition of foreign residents matters in shaping the geography of hate, given two opposing views on this respect. For the “contact theory”, proximity to ethnic groups that are perceived as diverse might reduce the likelihood of intolerance to ethnic diversity (Allport 1988; Gilliam et al. 2002). Conversely, the “group threat hypothesis” speculates that proximity to diverse ethnic groups might fuel intolerance built on stereotypes (Gilliam et al. 2002; Hopkins 2010). Operationally, we test these views through different specifications for the Two-Way Fixed Effect model, which differ in the measure for resident ethnic groups (African, Asian, North American, European Non-EU, Central and South American). Estimates show that the effect of hosting refugees on hate does not change. Aside East Asians, no ethnic group has significant association with hate (for detailed estimates see Appendix, Table 9). This evidence seems to support that the racial proximity theory is not relevant to understand the Italian geography of hate.

5.2 Endogeneity of refugee hosting capacity

We address identification by estimating a Two-Stage Least Square (2SLS-IV) model with the Bartik shift-share regressor summarized by Eq. (2) as IV. Table 2 outlines the results, which corroborate that the effect of hosting refugees on hate goes beyond correlation.

Table 2 Two-stage Least Square with Instrumental Variable. Panel model estimates and sensitivity tests for the effect of hosting refugees on hate events in the Italian NUTS3 areas between 2009 and 2017
Full size table

The effect of refugee hosting capacity on hate is robust in the IV specification (Table 2, column a.1). A 1% increase in local capacity of hosting refugees increases hate by 7.018%. This result holds to the inclusion of controls as detailed in columns a.2–a.5Footnote 21 (for detailed results on control variables, see Appendix, Table 10). Pre-existing residential support is a meaningful predictor for current refugee hosting capacity in all model specifications as outlined by the significance of the coefficients in Table 2b, columns 1–5. Also, pre-existing residential support is a strong instrument for current refugee hosting capacity, as highlighted by the Kleibergen-Paap Wald (KPW) rk F statistic of the instrument in the first stage, which is always above 12, as reported in Table 2.

5.3 Robustness checks

Our estimates are robust to the inclusion of other potential confounders, such as social capital, spatial spillover, voting preferences. Also, the 2SLS-IV estimation exploits an instrument that appears not weak. At the same time, further checks on the validity of the identification strategy are needed.

Our identification strategy assumes that extant residential support (our IV) influences hate only through current refugee hosting capacity. We check the validity of this exclusion restriction through a falsification test that compares the direct effect of the IV on hate in 2009–2017 with the same effect in 2009–2010 (when refugees were scarce, as highlighted by Fig. 2). Table 3 details this comparison, which supports the validity of the exclusion restriction for our IV.

Table 3 Internal validity of results. Falsification test
Full size table

Column 1 reports estimate from the reduced-form equation for our IV model specification, where the IV directly impacts the outcome variable for the 2009–2017 period. The IV has a significant positive correlation with hate. Column 2 reports estimate from the reduced-form equation for the 2009–2010 period. In this second case, the IV does not correlate with hate. This evidence suggests that our IV, designed on extant residential support, correlates with hate only when this extant residential support could be used to support refugees.

Then, we consider only the hosting capacity of SPRAR centers, given that SPRAR is the structural and longstanding refugee reception policy and that SPRAR is speculated to stimulate contact and cooperation between refugees and hosting community (Gamalerio et al. 2022). Estimates are reported in Table 4, column 1 and they confirm our main findings.

Table 4 2SLS-IVsensitivity tests: (1) considering only SPRAR hosting centers and (2) dropping Sicilian provinces from sample
Full size table

This result is also corroborated by anecdotal evidence from the press about many hate events against SPRAR centers (Lunaria 2017, 2019). Finally, we estimate the 2SLS-IV removing Sicilian provinces, since existing evidence suggests that refugee hosting in the region could be influenced by the strong presence of organized crime (Luca & Proietti 2022). Results do not change, as summarized by Table 4, column 2.

6 Conclusions

Using micro-regional data on hate and refugee reception in Italy between 2009 and 2017, this paper show that the local presence of refugee hosting centers is a trigger for hate. This effect is robust to several tests, which support that it may indeed be causal.

Our evidence suggests that sudden exposure to migrants is perceived as a threat by receiving communities, who polarize this unrest into hate. Importantly, this result corroborates the relevance of investigating the link between refugee reception and social unrest going beyond voting preferences. The latter might fail to account for existing patterns of polarization of discontent toward violent behaviors, which are rising hence calling for an ad-hoc analysis and policy interventions.

Further, the paper provides evidence of a causal link by relying on a database on hate events that combines (and cross-checks) information from multiple sources, alleviating possible measurement biases from high underreporting (as it is the case for police records). By doing so, this paper complements existing evidence that is largely descriptive and/or based on biased hate measures.

Notably, our results on the Italian case align with recent evidence for Germany (Dancygier et al. 2022; Entorf & Lange 2023) and Greece (Hangartner et al. 2018), showing that refugee reception policies in these countries have a negative effect on inclusivity and tolerance. Alignment of evidence from different countries appears to corroborate the validity of the link between refugee reception and hate.

Comparing this to the current mixed evidence on the refugee reception and voting preferences supports the importance of investigating refugees and resentment through the hate lens. Voting preferences might depend on more elements than refugees alone, such as available political supply and the economic and the welfare agenda of each political party. This complexity might confound voters’ choice, making them choose political parties based on composite rankings of issues, ranging from resentment to endorsement of specific political initiatives on employment and welfare. On the contrary, hate events are distilled manifestation of violent resentment.

The evidence presented in the paper opens a series of questions worth investigating in the future. First and foremost, it is necessary to address the relationship between hosting structures for refugees and hate in other contexts, to contribute to a more detailed picture on how different reception policies perform.

Then, future research should bridge the existing mixed evidence on the refugee reception/voting nexus with the more harmonized evidence on the refugee reception/hate nexus. This could allow to detect patterns of polarization of discontent.