Empirical Strategy

Building on Beerli et al. (Reference Beerli, Ruffner, Siegenthaler and Peri2021), we exploit the reform’s staggered implementation to identify its effects using a DID design. This design compares outcomes before and after the border opening in municipalities immediately next to the border versus those slightly further away. While it is impossible to directly test the parallel-trends assumption—that is, that both types of municipalities would have developed in the same way if the border had not opened—we perform a series of placebo tests to validate it during the pre-treatment period.

The left panel of Figure 1 illustrates our identification strategy. At the start of the transition phase, we define treated municipalities as those within a 15-minute drive of a border crossing (dark green). The control group includes municipalities within $ 15 $ – $ 30 $ minutes of a crossing (light green). The panel on the right visualizes the increase in the number of “immigrant workers” (the sum of CBWs and resident immigrants, see below) between 1996 and 2016, scaled by the number of Swiss citizens in 1998.Footnote ⁴ This panel indicates that the number of immigrant workers increased substantially after the reform: increases of more than 10% were common, especially for municipalities within 15 minutes of the border.

Figure 1.

Visualization of the Empirical Strategy

Note: The left panel shows treated municipalities (dark green) within less than $ 15 $ minutes of the border and control municipalities (light green) $ 15 $ – $ 30 $ minutes from the border. The right panel depicts the increase in the share of immigrant workers between 1996 and 2016.

We focus the main analyses on municipalities within 30 minutes of the border to maximize the internal validity of our estimates. Restricting our comparison to municipalities 0–15 versus 15–30 minutes from the border allows us to consider sets of municipalities that are ex ante quite similar in terms of political cleavages, government spending, and the overall structure of the economy.Footnote ⁵ To the extent that municipalities in the 15–30-minute range also experienced an increase in immigrant workers due to the border opening (see Beerli et al. Reference Beerli, Ruffner, Siegenthaler and Peri2021), our estimates represent a lower bound of the overall treatment effect.

Data and Measures

Our treatment indicator for border-proximate municipalities is a binary indicator based on the travel time to a border crossing. We measure travel times from the center of a municipality to the border using the Open Source Routing Machine (http://project-osrm.org/). Border Proximity takes a value of 1 for municipalities less than 15 minutes from the border, which includes all municipalities with a border crossing and those in close proximity ( $ n=523 $ ). Since they are easier to access, we expect citizens of neighboring countries to be more likely to work or live in these places compared to municipalities in the control group that are 15–30 minutes away ( $ n=470 $ ). Depending on the data source, we use 2012 or 2019 classifications of Swiss municipalities.

We create two binary indicators to indicate the period of the reform: Transition for 2000–03 and Free Movement for 2004 onward. We quantify the impact of the liberalization process on immigration by measuring the Immigrant Worker-to-Swiss Ratio, which is the ratio of CBWs and resident immigrants to the number of Swiss citizens. We focus on this ratio to account for the different sizes of border municipalities (which include Zurich, Geneva, and Basel, Switzerland’s three largest cities, as well as municipalities with fewer than 1,000 residents) and the fact that the same absolute increase in immigration affects communities of different sizes differently. We calculate this ratio by combining information from two datasets compiled by the Swiss Federal Office of Statistics (SFOS): (1) data on the resident immigrant and Swiss populations between 1991 and 2018 and (2) the number of CBWs from 1996 to 2016, which we aggregate to the municipal level (Swiss Federal Office of Statistics 2017; 2019a). We use 1998 levels to fix the number of Swiss citizens in a municipality in this ratio to ensure that any variation in our measure is due solely to changes in the number of foreigners and not to changes in the population caused by factors such as Swiss citizens moving away due to an increase in immigrant workers.Footnote ⁶

We measure the electoral consequences of open borders with Anti-Immigrant Party Support in federal elections. This outcome comes from the SFOS, which covers elections from 1991 to 2019 (Swiss Federal Office of Statistics 2019b). While Switzerland has an extensive party system with several regional parties, right-wing populist parties that espouse nativist and anti-immigrant positions have been prevalent throughout the country during the study period. The most successful of these parties is the Swiss People’s Party (SVP; Schweizerische Volkspartei / Union Démocratique du Centre), which has existed under this name since the 1970s, but only experienced substantial growth in the 1990s and 2000s after a turn to anti-immigrant populism. In addition to the SVP, our measure includes support for the following anti-immigrant parties: the Ticino League, the Geneva/Romandy Citizens’ Movement, the Swiss Democrats, the Republicans, and the Freedom Party of Switzerland.

To study the attitudinal effects of border liberalization, we use data from the Swiss Household Panel (SHP) and VOX surveys (GfS-Forschungsinstitut. 2017; Tillmann et al. Reference Tillmann, Voorpostel, Antal, Dasoki, Klaas, Kuhn and Lebert2022). The SHP is an annual panel survey that began in 1999 (so we have only 1 year of observations in the pre-treatment phase). The VOX survey has been conducted after every popular vote, including referendums and popular initiatives under Switzerland’s direct democracy system. These votes typically occur four to six times each year, and the survey has been consistently administered since 1996, providing a repeated cross section of the population. All survey outcomes are standardized to have zero mean and unit standard deviations. We exclude missing and inapplicable responses from the analyses.

To measure the potential displacement effects on Swiss citizens’ labor market outcomes, we use the biannual Swiss Earnings Structure Surveys (SESS) from 1994 to 2010 (Swiss Federal Office of Statistics 2016).Footnote ⁷ Conducted by the SFOS, the SESS provides a stratified random sample of workers employed in private and public firms. Each iteration covers 16.6% (1996) to 50% (2010) of the country’s total employment. We use the SESS data to construct measures of Employment and Hourly Wages for Swiss workers.Footnote ⁸

To measure traffic congestion, a prominent indicator of density stress, we use data from road traffic counting stations operated by the Swiss Federal Roads Office. These stations counted the number of cars passing every hour of the day from 1997 to 2015 (Swiss Federal Roads Office 2022). We calculate Traffic as the average number of cars passing on both sides of the road per hour. We assign each counting station to the 0–15 or 15–30 minute region by either directly using the coordinates given in the data description (for about 70% of the counting stations) or linking the station’s name to the closest municipality.

Finally, we compiled a dataset of bills introduced between 1992 and 2021 in the cantonal parliament of Ticino (Ticino Parliament 2022). After scraping all the bills from the Ticino parliament’s website, we identified the authors and sponsors of each bill and matched them with a dataset containing the names and municipalities that the authors represent. We retained the bills sponsored by legislators representing either treated or control municipalities. We then used keywords to search the titles to identify bills related to immigration to proxy for elite anti-immigrant rhetoric.Footnote ⁹ We manually reviewed this list to determine whether each bill’s title was actually related to immigration. This produced a list of 267 unique bills related to immigration between 1992 and 2021.

Statistical Models

We estimate the effect of exposure to open borders using linear panel regressions with municipality and year fixed effects. Our specification includes two predictors: Border Proximity × Transition and Border Proximity × Free Movement. The first estimates the effect ( $ \rho $ ) of the partially open border of the transition period. The second is the main quantity of interest ( $ \gamma $ ), the effect of open borders, comparing municipalities in Border Proximity (< 15 minutes) to control municipalities (15–30 minutes). By controlling for municipal ( $ {\delta}_i $ ) and year fixed effects ( $ {\lambda}_t $ ), our specifications account for all confounding variables that are constant within municipalities over time as well as all time-varying confounders that are constant across municipalities. Unless otherwise specified, we cluster standard errors at the municipality level to account for disturbances correlated within municipalities over time and include population weights. We run the following model:

⁽¹⁾ $$ {y}_{it}={\displaystyle \begin{array}{l}{\lambda}_t+{\delta}_i+\rho {\mathrm{Border}\ \mathrm{Proximity}}_i\times {\mathrm{Transition}}_t\\ {}+\hskip2px \gamma {\mathrm{Border}\ \mathrm{Proximity}}_i\times {\mathrm{Free}\ \mathrm{Movement}}_t+{\epsilon}_{it}.\end{array}} $$

This model estimates the average treatment effect for border-proximate municipalities separately for the transition and free-movement periods. We use this model to estimate how liberalization affects our various outcome measures, including the share of immigrant workers in the population and anti-immigrant parties’ vote share. We complement this with a second specification that estimates year- or election-specific effects by replacing our two predictors with interactions between Border Proximity and binary indicators for all years/elections except the last one before the transition period, which serves as our baseline. This specification allows us to compare outcomes in border versus control municipalities against the baseline year. While these specifications are somewhat more demanding, they provide two significant benefits. First, they allow us to empirically assess an implication of the parallel-trends assumption: we can test for significant differences in the outcome trajectories of treated and control municipalities before the borders opened, which would imply the presence of unmeasured time-varying confounders. Second, these specifications allow us to estimate how long it takes for any effects to materialize and their persistence using the following model:

(2) $$ {y}_{it}={\displaystyle \begin{array}{l}{\lambda}_t+{\delta}_i+\sum_{t=1996,t\ne 1999}^{2016}{\gamma}_tI\left(\mathrm{year}=t\right)\\ {}\times {\mathrm{Border}\ \mathrm{Proximity}}_i+{\epsilon}_{it}.\end{array}} $$

The Role of Political Elites

In light of these null findings for the standard explanations of anti-immigrant voting in this context, how can we explain the effects of border liberalization on voting for anti-immigrant parties? Research on public opinion formation emphasizes two processes to describe the relationship between political elites and the public. Early work in this area focused on a top-down process in which political elites influence public attitudes in a variety of ways, including through cues, persuasion, and issue framing (DellaVigna and Gentzkow Reference DellaVigna and Gentzkow2010; Friedman Reference Friedman2012; Gerber et al. Reference Gerber, Gimpel, Green and Shaw2011; Kinder Reference Kinder1998; Zaller Reference Zaller1992). Later research also highlighted a bottom-up process, in which the public influences elites. According to this view, the public has consistent opinions on specific issues, to which politicians respond by strategically adjusting their policy positions (Shapiro and Page Reference Shapiro and Page1988). Recent scholarship has argued that both processes can simultaneously co-exist (Kertzer and Zeitzoff Reference Kertzer and Zeitzoff2017; Steenbergen, Edwards, and De Vries Reference Steenbergen, Edwards and De Vries2007). Together, these strands of the literature suggest that elites can both influence and be influenced by public opinion.

The extensive research on elites’ impact on various political outcomes—including political attitudes, far-right voting, hate crimes, online prejudice, and the erosion of democratic norms (Alrababah et al. Reference Alrababah, Marble, Mousa and Siegel2021b; Clayton et al. Reference Clayton, Davis, Nyhan, Porter, Ryan and Wood2021; De Vries and Hobolt Reference De Vries and Hobolt2020; Zaller Reference Zaller1992)—highlights a significant gap in studies of anti-immigrant attitudes, particularly regarding the role of elites in shaping them.Footnote ¹⁷ In this section, we propose an explanation for why political elites can be effective at changing attitudes toward immigration. We argue that elites can be particularly effective at shaping public attitudes when they introduce novel narratives about the threats posed by immigrants, especially in areas that receive a large and relatively sudden influx of immigration.

Our starting point in this argument is Hopkins’s “politicized places” hypothesis, which maintains that anti-immigrant sentiment increases when communities experience sudden influxes of immigrants and when national rhetoric emphasizing the threat of immigration is salient. This suggests that a sudden rise in immigration gives anti-immigrant elites the opportunity to increase the salience of rhetoric against the flow of migrants. We argue that where traditional threats from immigration are insufficient, political elites could formulate and propagate new narratives to shift public opinion via public statements, legislative initiatives, political campaigns—and, in Switzerland, popular initiatives for the public to vote on.

When traditional threats are minimal, anti-immigrant rhetoric becomes compelling by highlighting novel threats, making it difficult for the general public and pro-immigrant advocates to respond effectively. We do not argue that elites will stop exploiting traditional fears, but that they can combine them with novel threats to mobilize voters against immigration. We expect elite narratives to change the opinions of certain subgroups. Drawing on Zaller’s model of public opinion formation, we also suggest that individuals with moderate levels of political awareness are especially susceptible to elite rhetoric. This group is more likely to receive messages from political elites and is less likely to resist them. Furthermore, in line with Hopkins (Reference Hopkins2010), we expect elite narratives to be most effective in areas that receive large immigration inflows.

Our argument may seem at odds with at least two recent studies on attitudes toward immigrants. For instance, Donnelly, Islam, and Savoie (Reference Donnelly, Islam and Savoie2020) use survey experiments informing respondents that some politicians, business leaders, or unions bring in skills or contribute to local culture and find null or small effects. Yet their findings are likely to differ from ours for three reasons. First, they use elites (politicians) as a control group, as their goal is not to isolate the effects of elites per se. Second, their experiment seeks to improve attitudes toward immigration, which might be more challenging than lowering them. Finally, their treatment consists of a single sentence, so its effect is unlikely to be comparable to the influx of large numbers of immigrants over multiple years.

Second, Kustov, Laaker, and Reller find that immigration attitudes tend to be stable over time. They provide convincing evidence from panel studies in several countries that the immigration attitudes of roughly 30%–80% of citizens remained stable during the study period. This could suggest that elites (or shocks) may not significantly affect attitudes toward immigration. However, we do not argue that the attitudes of the average citizen tend to be unstable. Instead, we suggest that, in line with Hopkins (Reference Hopkins2010), the effects are strongest among the citizens most exposed to immigration. In our case, border liberalization increased anti-immigrant voting by 4 (CI 1–7) to 6 (CI 2–11) percentage points. The findings in Kustov, Laaker, and Reller (Reference Kustov, Laaker and Reller2021) do not rule out the possibility that exposure to migration affects relevant subgroups—particularly those most exposed to immigrant arrivals.

The limited significance of traditional explanations in our context gives us an opportunity to unravel the top-down process. Since the 1970s, political entrepreneurs in Switzerland have launched multiple initiatives to reduce immigration. They have typically mobilized around two related narratives: “Überfremdung” (over-foreignization) and its sanitized cousin, “Übervölkerung” (overpopulation). The former communicates traditional fears associated with immigration, including a loss of cultural identity, social cohesion, and public safety. The latter echoes Malthusian theories of exponential population growth and is ostensibly less xenophobic as it implies that the country is getting too crowded. However, political actors in Switzerland who speak about overpopulation are clear about who they believe is responsible for population growth (foreigners) and their proposed solution (reducing immigration) (see Odenwald Reference Odenwald2021). James Schwarzenbach, the Swiss pioneer of far-right populism who was responsible for launching several early initiatives against over-foreignization and overpopulation, was transparent about his role in fueling anti-immigrant votes. Following the qualified success of his first initiative in 1970, he even styled himself after Goethe’s Sorcerer’s Apprentice in an op-ed: “The spirits that I called, I want to keep and multiply” (Schwarzenbach Reference Schwarzenbach1971).

With the border opening, and likely because of limited cultural and security concerns about immigration from neighboring countries, the overpopulation narrative took hold. In the early 2000s, this culminated in right-wing parties advancing a term borrowed from biology—“density stress” (Dichtestress)—that describes a collapse in behavior resulting from overcrowding due to unfettered population growth.Footnote ¹⁸ Populist right-wing politicians used this catch-all term to complain about issues as diverse as overcrowded trains, congested roads, urbanization of the countryside, and the loss of cultivated land (Odenwald Reference Odenwald2021). An SVP pamphlet even referenced crowded malls, queues at cinemas and shopping malls, and the lack of empty parking spaces as examples of density stress.Footnote ¹⁹

As seen previously with the term overpopulation, those employing the concept of density stress attributed overcrowding to foreigners and advocated immigration restrictions as the solution. The SVP initiative “Against Mass Immigration,” launched in 2014, sought to impose strict quotas on EU immigration. During the campaign leading up to the vote, supporters highlighted density stress as a primary reason why the free movement agreement with the EU has harmed Switzerland (Odenwald Reference Odenwald2021).Footnote ²⁰ Of course, this does not mean that anti-immigrant elites have stopped talking about traditional (e.g., economic, cultural, and security) threats from immigration. We include these anecdotes to highlight that anti-immigrant elites in Switzerland also discussed narratives related to density stress and overpopulation. Dataverse Appendix Figure B10 shows that news coverage of these narratives in articles that also mention anti-immigrant parties, significantly increased in the free movement period, starting in the late 2000s and peaking in the mid-2010s. While it decreased afterward, this rhetoric has increased again in recent years. The figure also depicts narratives associated with economic threats of migrants that mention these parties, suggesting that political parties may not have abandoned narratives of traditional threats but instead supplemented them with new discussions of density stress.

To determine whether this narrative of density stress was pure rhetoric or at least partly reflected a measurable change in the crowdedness of border neighborhoods, we examine density stress’ arguably most salient indicator: traffic congestion. Using our DID strategy and detailed data from road traffic counting stations from the Federal Roads Office for the years 1997–2015, we test whether municipalities within 0–15 minutes of border crossings were exposed to more traffic after liberalization than those within 15–30 minutes.Footnote ²¹ Figure 8 presents the results. Panel A indicates an overall increase in traffic during the transition, and especially the free movement phases. However, this increase is similar (and if anything, larger) for control municipalities than for treated municipalities, a pattern confirmed by both the event-study estimates (Panel B) and the DID estimates (Panel C). We also conduct robustness tests using two related outcomes: (1) traffic excluding counting stations on highways, which are more likely to contain longer-distance traffic, and (2) an index measuring traffic congestion, which we define as the absolute difference in traffic between the directions of a counting station in an hour divided by the mean traffic per day of the counting station. The results remain similar, as shown in Dataverse Appendix Figures B1–B3. Together, the data provide no evidence that traffic disproportionately increased in the border region following liberalization. This finding suggests that the narrative regarding density stress advanced by anti-immigrant parties did not reflect an actual change in a prominent indicator of density stress related to life quality in border municipalities.

Figure 8.

Effects of Opening Borders on Traffic

Note: This figure displays the impact of border liberalization on the annual traffic in treated and control municipalities. Panel A depicts annual traffic means per operative hour of car counting stations on Swiss roads located in both types of municipalities. The lines are loess-smoother estimates of the regional time trend, and the shaded areas represent 95% CIs. Panel B displays the estimates (and 95% CIs) from an event study akin to Equation 2 with year and counting station fixed effects and a linear regional trend. It shows the differential change in traffic between treated ( $ \hskip-2px <15 $ minutes to the border) and untreated (15–30 minutes to the border) municipalities. Panel C presents estimates (and 90% and 95% CIs) from a DID regression, as in Equation 1, including the linear trend interacted with the treated region indicator. The regression results are shown in models 1 and 3 of SM Table A23.

If not because of an increase in traffic congestion, why did the narrative of density stress resonate more with voters in the border region? We discuss two complementary explanations. First, we might expect elites to go beyond national rhetoric and strategically intensify their position taking in places where it is likely to resonate with voters. To explore this channel, we focus on parliamentary bills in the canton of Ticino. In line with previous research, we use parliamentary bills to measure legislators’ position taking (Mayhew Reference Mayhew2004), which can also shape voter attitudes (Broockman and Butler Reference Broockman and Butler2017). This measure has the advantages of being direct (i.e., not mediated via the media), behavioral, and geographically fine-grained. We focus on Ticino for two reasons. First, we were able to obtain data on parliamentary bills for this canton from 1992—before the bilateral agreement with the EU. Second, opening the border had a particularly sizeable effect on immigration in Ticino: Panel B of Figure 2 establishes that the DID estimates for the border regions in the Italian- and Romansh-language cantons were larger than in French-language and particularly in German-language cantons. This suggests that the border municipalities in Ticino are a prime target for anti-immigrant rhetoric.Footnote ²²

Our manual review of all bills identified 267 unique bills related to immigration between 1992 and 2021. Such bills often emphasize issues related to the density stress concept, such as the alleged overcrowding and pressure on Ticino’s infrastructure. We coded whether any Member of Parliament (MP) from treated or control municipalities sponsored these immigration-related bills. For each region, our outcome measure equals 1 if the bill is related to immigration and any of its sponsors represents a municipality in that region and 0 otherwise. Our unit of analysis is the bill-region level. We regress this outcome on our treatment indicators and adjust for year fixed effects. We cluster standard errors by bills.Footnote ²³

Figure 9 shows that the share of immigration-related bills was fairly constant in control municipalities during the pre-reform and transition phases. In border municipalities, the share declined during the pre-reform phase but then increased toward the end of the transition phase. This divergent trend gained further traction during the free movement phase. The DID estimates, which indicate a differential increase in the share of immigration-related bills of more than 1 percentage point (CI 0.4–1.8) during the free movement period, further support this finding.Footnote ²⁴ In Dataverse Appendix Figure B5, we subset the analysis by the party affiliation of the bill’s sponsor using binary coding for far-right and other parties. This analysis shows that the effect on immigration-related bills depicted in Figure 9 is driven by far-right parties.

Figure 9.

Effects of Opening Borders on Immigration-Related Parliamentary Bills in Ticino

Note: Panel A: Points represent yearly means for treated ( $ <15 $ minutes to the border) and control (15–30 minutes to the border) municipalities within Ticino. Lines are a loess-smoother estimate of the over-time trend; shaded areas denote 95% CIs. Panel B: Regression estimates of the differential change in immigration-related bills between treated and control municipalities. The 1999 election is left out as the baseline. Points are estimates with 95% CI bars. Panel C: Regression estimates of the differential change in the share of immigration-related bills in treated and control municipalities. Points are estimates with 95% CI bars. The regression results are shown in models 1 and 2 of SM Table A24.

In addition to elites targeting border municipalities, we might expect density stress to resonate more strongly with voters who experience higher exposure to foreigners. This expectation aligns with the “politicized places” hypothesis, which postulates that “political reactions to neighboring immigrants are most likely when communities undergo sudden influxes of immigrants and when salient national rhetoric reinforces the threat” (Hopkins Reference Hopkins2010, 40). In several dimensions, the border opening in Ticino is a textbook case of the politicized places hypothesis: a sudden and substantial increase in the presence of immigrant workers interacted with a highly salient rhetoric that construed immigration as a threat to the quality of life. The only caveat is that, in our context, political rhetoric takes place at the local and national levels, a point we revisit in the conclusion. While it is difficult to directly test the politicized places hypothesis without an exogenous shift in political rhetoric, we can still provide indirect evidence by investigating whether the effects of the density stress rhetoric are more pronounced among voters who are more susceptible to being influenced by this rhetoric. To do so, we marry the politicized places hypothesis with Zaller’s (Reference Zaller1992) RAS model of opinion change.

Building on McGuire (Reference McGuire1968), Zaller (Reference Zaller1992) suggests that persuasion partly depends on two factors: (1) the likelihood of being exposed to information (reception) and (2) the possibility that the message is persuasive (acceptance). In this model, political awareness, defined as “the extent to which an individual pays attention to politics and understands what he or she has encountered,” plays a central role (Zaller Reference Zaller1992, 21). According to the RAS model, people with low levels of political awareness are less likely to receive information, making them less susceptible to persuasion. People with high levels of political awareness are more likely to receive information. Yet, since they are also likely to have more solidified political predispositions, they are less persuadable by new information. Zaller thus suggests that people with intermediate levels of political awareness are most likely to change their attitudes. Combining the politicized places hypothesis with the RAS model, we derive the prediction that citizens with intermediate levels of political awareness who live in border municipalities in Ticino should react most strongly to the interaction of increasing immigration and the local and national rhetoric about density stress.Footnote ²⁵ We expect citizens in the treated region with low and high levels of political awareness and all citizens in the control region to be less persuaded by the density stress rhetoric. Using the SHP again, we measure political awareness using the question: Generally, how interested are you in politics, if 0 means ‘not at all interested’ and 10 ‘very interested’? We divide respondents into three roughly equal-sized groups of low, intermediate, and high levels of awareness. To measure attitudes toward immigrants, we use the only available question: Are you in favor of Switzerland offering foreigners the same opportunities as those offered to Swiss citizens, or in favor of Switzerland offering Swiss citizens better opportunities?

Figure 10 plots the results.Footnote ²⁶ We find no significant effects during the transition and free movement phases among respondents with low and high levels of awareness on their likelihood of opposing equal opportunities for foreigners. Citizens with intermediate levels of political awareness in the treated region were significantly more likely to oppose equal opportunities for foreigners during these phases. Based on six subgroup-specific estimates, this pattern aligns with the predictions based on the combined politicized places hypothesis and RAS model. While consistent with our argument, the differences between the political awareness subgroups are not significant and this evidence is only suggestive. We hope future research examines the causal effects of elites on anti-immigrant attitudes.

Figure 10.

Effects of Opening Borders on Favoring More Opportunities for the Swiss Compared to Foreigners

Note: This figure shows the DID estimates of the effect of border liberalization on attitudes toward providing foreigners the same opportunities as Swiss citizens by the level of self-reported political awareness. Higher scores on the outcome indicate favoring more opportunities for Swiss than foreigners. Points are estimates with cluster-robust 90% (thick) and 95% (thin) CIs. The regression results are reported in models 1, 3, and 5 of SM Table A27.

An alternative explanation is that immigrants from neighboring countries may have brought anti-immigrant discourse from their places of origin.Footnote ²⁷ If elites in neighboring countries engaged in anti-immigrant rhetoric focusing on overcrowding, then immigration into Switzerland may have caused these ideas to spread. To examine this possibility, we compare coverage in major Swiss newspapers by language region to newspapers in neighboring countries. We use Factiva and search in major German, French, and Italian newspapers, as well as German- and French-speaking newspapers in Switzerland.Footnote ²⁸ We report the results in Dataverse Appendix Figure B9. Coverage of terms related to overcrowding constituted a higher portion of the total coverage of immigration in Swiss newspapers, both German speaking and French speaking, than in German or French newspapers. Coverage of these issues in Italian newspapers was quite low during the study period, which suggests that migrants were unlikely to have imported this discourse from their countries of origin.