Immigration has emerged as a divisive political issue in many countries in recent decades. One of the most controversial debates over immigration policy involves the integration of already-settled migrants and, in particular, their access to citizenship.Footnote 1 In the U.S., there are heated debates about restricting birthright citizenship for children of unauthorized immigrants.Footnote 2 Throughout Europe, right-wing parties use citizenship policies as a vehicle to mobilize voters against immigration with campaigns that emphasize the societal dangers of naturalizing increasing numbers of immigrants (Dancygier Reference Dancygier2010; Helbling Reference Helbling2008; Howard Reference Howard2009). These groups reject the integration of foreigners as citizens because they view immigrants as undeserving outsiders who poach jobs from native workers, unsettle local communities, and undermine traditional values; such outsiders should not be rewarded with equal access to the political and social rights of the host country (Brubaker Reference Brubaker1989; Givens Reference Givens2007; Koopmans et al. Reference Koopmans, Statham, Giugni and Passy2005). Intense debates over naturalization policies are likely to escalate further in the years ahead in light of increased migration flows; immigrants already account for about 10% of the population across advanced industrialized countries (Dumont, Spielvogel, and Widmaier Reference Dumont, Spielvogel and Widmaier2010).
Why do some natives oppose and others favor immigration and naturalization of immigrants? Do natives discriminate against particular types of immigrants, and if so, which immigrants are welcomed and which immigrants are rejected? A large body of literature has examined attitudes toward immigration in Europe, the U.S., and several other countries, but scholars still disagree about the prevalence and causes of anti-immigrant sentiment (Ceobanu and Escandell Reference Ceobanu and Escandell2010; Hainmueller and Hopkins Reference Hainmueller and Hopkins2012). One important limitation of existing research is the absence of detailed behavioral data on anti-immigrant sentiment. Most published studies to date are based on attitudinal measures from public opinion surveys, which have their merits, but also impose important limits on the inferences we can draw. For example, most existing surveys are fairly blunt instruments that ask respondents only to describe their attitudes toward immigration in general, although we expect that natives’ views vary in important ways across different types of immigrants (e.g., country of origin, skill level, etc.).Footnote 3 Another limitation is that many studies rely on cross-sectional surveys, which makes it difficult to assess how hostility toward immigrant groups varies over time.Footnote 4 Moreover, surveys put participants in an artificial research setting where responses have almost no real-life consequences, and therefore often boil down to “top of the head” reactions that may inaccurately capture true beliefs (Bertrand and Mullainathan Reference Bertrand and Mullainathan2001; Taylor and Fiske Reference Taylor and Fiske1978). Another well-recognized problem is that answers to sensitive questions about immigration and racial policies can be biased by social desirability effects, because respondents are unwilling to admit to discriminatory attitudes in times when discrimination has become illegal and socially unacceptable in many countries (Berinsky Reference Berinsky1999; Kuklinski, Cobb, and Gilens Reference Kuklinski, Cobb and Gilens1997).Footnote 5
In light of these problems, some scholars have turned to field experiments such as audit studies to behaviorally measure racial discrimination in the labor market and other economic domains (Adida, Laitin, and Valfort Reference Adida, Laitin and Valfort2010; Bertrand and Mullainathan Reference Bertrand and Mullainathan2004; Jowell and Prescott-Clarke Reference Jowell and Prescott-Clarke1970; List Reference List2004). We take a complementary approach and study discrimination in naturalization decisions, drawing upon a natural experiment from Switzerland that allows us to overcome some of the inferential challenges mentioned above and illuminates how discrimination against immigrants varies across different types of immigrants and over time.
In Switzerland, each municipality autonomously decides on the naturalization applications of its foreign residents who seek Swiss citizenship. We focus on the group of municipalities that until 2003 used referendumsFootnote 6 with closed ballots to decide on naturalization requests. A typical naturalization referendum involved two stages. Local voters first received official voting leaflets that explained the pending naturalization request with a detailed description of each immigrant applicant. Voters then cast a secret ballot on each individual request, and applicants with a majority of “yes” votes were granted Swiss citizenship. Drawing upon local municipality archives, we collected a new dataset that contains applicant characteristics and voting outcomes for the 2,400 recorded naturalization referendums held between 1970 and 2003 in the 44 Swiss municipalities that used secret ballot referendums with voting leaflets. We use these data to examine how applicant characteristics affect the outcome of naturalization referendums.
This research design has at least three advantages compared to previous work. First, naturalization referendums enable us to measure revealed preferences regarding immigrants based on actual behavior in a real-world setting. Social desirability bias is not a concern, because voters used secret ballots and needed to provide no justification for their votes. Voters also had to face the consequences of their voting behavior, as naturalized immigrants instantly acquired the same rights as existing members of the local citizenry (including the right to vote and permanently stay in the municipality). The data should therefore reveal a fairly accurate assessment of the immigration preferences of the local voting population. As Bell Jr. (Reference Bell1978, 14) put it, referendum voting “enables voters’ racial beliefs and fears to be recorded and tabulated in their pure form.” Second, much like a real-world vignette experiment, our data covers comparable application decisions regarding thousands of immigrants with radically different attributes and at different points in time, allowing us to pinpoint particular immigrant attributes that are valued or disliked by local voters. This data set also allows us to examine how discrimination against particular immigrant groups varies over time in response to immigration waves. Third, our design enables us to minimize potential omitted variable bias, since we measure and control for the same applicant information from the official voting leaflets that voters had at their disposal when they decided on the naturalization requests. We also show that our main results are insensitive to hidden bias that could arise from the fact that some knowledgeable voters decided based on private information about particular applicants that extended beyond the information provided in the leaflets.
When interpreting our results, it is important to emphasize the we capture only the effects of the applicant characteristics in the last hurdle of the application process, once a naturalization request was put to a popular vote. Since we do not capture additional forms of potential discrimination that may have deterred immigrants from applying for citizenship in the first place, our estimates are best understood as a lower bound on the overall prevalence of discrimination in naturalization outcomes in the sample municipalities.
We find that naturalization outcomes vary dramatically across and within municipalities. Country of origin is by far the most important determinant of naturalization success. The average proportion voting “no” in the naturalization referendums is about 13–15 percentage points higher for applicants from (the former) Yugoslavia and Turkey compared to observably similar applicants from richer northern and western European countries who apply in the same municipality at the same time; this corresponds to a 40% increase over the average proportion of “no” votes (or about a 120% increase over the average probability of being rejected). This massive origin disadvantage is very similar in smaller and larger municipalities and insensitive to unobserved confounders with Rosenbaum Gamma values between 9 and 11, indicating that it would take an enormous hidden bias to explain away the origin effects. We also find smaller origin penalties for applicants from other groups, including applicants from central and eastern European countries (6 percentage points), Asia (3 percentage points), and other non-European countries (7 percentage points), but no disadvantage for applicants from southern Europe. Taken together, origin alone accounts for about 40% of the within-municipality variation in the proportion of “no” votes.
Other immigrant characteristics also influence the naturalization success. Voters systematically prefer applicants with better economic credentials (as measured by occupational skill, education, and prior unemployment), applicants born in Switzerland, and applicants with longer residency, but the effects of these characteristics are very small compared to the origin effects. We also find that language skills have almost no effect on naturalization success. Even applicants who are described to voters as “perfectly fluent” in the Swiss-German dialect earn no significant advantage. Similarly, the assessed integration status of the applicant plays very little role overall (the exception is applicants who earn a slight advantage because they are described to voters as being “completely indistinguishable” from a Swiss native).
We use additional tests that consider statistical and taste-based discrimination theories to illuminate the mechanisms that may explain the varying treatment of applicants based on their country of origin. Theories of statistical discrimination (Arrow Reference Arrow1972; Phelps Reference Phelps1972) suggest that if the average integration level of immigrants varies by origin group, then voters, even if all they care about is that applicants are sufficiently well integrated, find it optimal to place some weight on an applicant's origin to make a better guess about her true integration level (which is measured with noise in the voting leaflets). This logic implies that voters will reward an applicant more strongly for additional observable credentials that are informative about her integration status (such as higher educational attainment) if the applicant belongs to an origin group that is believed to have a lower average integration level. Consistent with this theoretical expectation, better economic credentials strongly decrease the proportion of “no” votes among applicants from Turkey and (the former) Yugoslavia but have no effect among applicants from richer northern and western European countries.
We also find evidence for theories of taste-based discrimination (Allport Reference Allport1979; Becker Reference Becker1971) where a “taste for discrimination” directly enters the utility function of voters who are assumed to hold xenophobic animus against immigrants from particular origin groups. Consistent with this logic, we find that origin-based discrimination steeply increases with the xenophobic preferences of the local population (as measured by vote shares from federal anti-immigration referendums); the origin penalty for applicants from (the former) Yugoslavia and Turkey almost triples in the most, compared to the least, xenophobic municipalities. Overall, these results indicate that statistical discrimination may account for about 40% and taste-based discrimination for about 60% of the origin-based discrimination in our sample.
Consistent with group conflict theories that link the rise of xenophobic animus to a defensive reaction of natives who view immigrant out-groups as a threat to the natives’ dominant position (Blalock Reference Blalock1967; Blumer Reference Blumer1958; Quillian Reference Quillian1995), we also find that origin-based discrimination is dynamically correlated to changes in the relative sizes of the different origin groups. While discrimination against applicants from Turkey and (the former) Yugoslavia increases following the rapid growth of these groups during the 1990s, discrimination against immigrants from southern European origins abates as the relative size of this group decreases.
Our study contributes to several literatures. First, our findings have important implications for the research that studies the prevalence and drivers of anti-immigrant sentiment (e.g., Brader, Valentino, and Suhay Reference Brader, Valentino and Suhay2008; Dustmann and Preston Reference Dustmann and Preston2007; Hainmueller and Hiscox Reference Hainmueller and Hiscox2007; Reference Hainmueller and Hiscox2010; Sides and Citrin Reference Sides and Citrin2007; Sniderman, Hagendoorn, and Prior Reference Sniderman, Hagendoorn and Prior2004). Although most studies to date are based on survey data and consider only attitudes toward immigration in general, our results provide clear behavioral evidence that natives do not treat all immigrants equally but instead draw important distinctions between different types of immigrants and that these preferences are not static but vary over time. Our results also show that measurement matters: the immigrant preferences revealed in our behavioral data contrast sharply with immigration preferences as measured by comparable public opinion surveys, which are prone to social desirability bias. Second, our findings resonate with the literature on racial discrimination (e.g., Altonji and Blank Reference Altonji and Blank1999; Blank, Dabady, and Citro Reference Blank, Dabady and Citro2004; Pager and Shepherd Reference Pager and Shepherd2008). While most of this work has focused on race- and gender-based discrimination in the labor market, our results demonstrate that naturalization decisions for resident aliens can be subject to similar types of discrimination, and that similar theoretical mechanisms provide explanatory leverage to account for this discrimination. Third, by comparing microlevel data about individual naturalization decisions across several municipalities, our study adds to the small but growing literature in immigration studies that has begun to examine outcomes and policies at the local level, where more focused comparisons are less vulnerable to biases from unobserved heterogeneity that often plague cross-national research (e.g., Adida Reference Adida2011; Dancygier Reference Dancygier2010; Helbling Reference Helbling2008; Hopkins Reference Hopkins2010). Fourth, our findings add new empirical evidence to the extensive literature and policy debates about citizenship policies and the integration of already settled immigrants (e.g., Freeman Reference Freeman2004; Givens Reference Givens2007; Hochschild and Mollenkopf Reference Hochschild and Mollenkopf2009). Finally, our study contributes to the literature that studies the relationship between direct democracy and the protection of minorities’ civil rights (e.g., Frey and Goette Reference Frey and Goette1998; Gamble Reference Gamble1997; Hajnal, Gerber, and Louch Reference Hajnal, Gerber and Louch2002; Maskin and Tirole Reference Maskin and Tirole2004). Our results provide evidence that, in the absence of other institutional safeguards, referendum voting can result in systematic discrimination against particular minority groups who find themselves at the whim of the native majority. This finding informs ongoing policy debates about reforming the Swiss naturalization system.
CITIZENSHIP POLICY IN CONTEXT
Before we turn to the empirical analysis, it is helpful to put the Swiss naturalization system in a comparative perspective and discuss the potential benefits of obtaining Swiss citizenship.
The Swiss Naturalization System
Ever since classical Athens, states have used citizenship as a closure mechanism to define a select group of members who belong to a polity and enjoy special privileges denied to nonmembers. Historically, eligibility criteria for citizenship often included attributes such as class, wealth, ethnicity, race, and gender, and requirements frequently changed over time to accommodate developments in state capacity (Bellamy Reference Bellamy2008). In Switzerland, this link between citizenship and state building gave rise to a system of triple citizenship, which defines Swiss citizenship based on citizenship in a municipality, a canton, and the Confederation (Helbling Reference Helbling2008, 12–17). This three-tiered system is unique in that it delegates responsibility for naturalizing foreigners largely to the municipal level. Federal laws impose formal naturalization requirements, but an immigrant cannot obtain a Swiss passport without acquiring citizenship of a municipality, and municipalities enact the naturalization procedures and ultimately decide on the applications.Footnote 7 This contrasts with many other countries where naturalization procedures and criteria are typically defined at the federal level and implemented by federal ministries or agencies (as in the U.S., France, Canada, and Belgium). Other countries have hybrid regimes where rules are stipulated at the federal level, but applications are checked and decided at the regional level (as in Germany and Austria).Footnote 8
In general, Switzerland is often categorized as part of the group of countries with relatively restrictive citizenship regimes, such as Germany, Denmark, Austria, and Greece (Goodman Reference Goodman2010). In the other pool are countries with more liberal citizenship regimes such as France, the UK, Belgium, and the U.S. Although naturalization regimes are complex and generally difficult to compare across countries, the more restrictive countries are typically characterized by the fact that they rely on the jus sanguinis principle, which implies that citizenship is passed on from the citizenship of the parents, rather than granted based on the place of birth. More liberal countries also typically require around 5 years of permanent residence, while more restrictive regimes require up to 8 to 10 years before resident immigrants become eligible for citizenship.Footnote 9 More restrictive regimes also typically require renunciation of prior citizenship and do not allow naturalized immigrants to hold multiple passports. Switzerland introduced multiple citizenship for naturalized immigrants in 1992, in contrast to many of the traditionally more restrictive countries. Almost all Western countries, including Switzerland, also have additional naturalization criteria that require a clean criminal record, some evidence of financial self-sufficiency, and, in many cases, demonstrated mastery of the country's official language(s). Several countries, such as Switzerland, also require that applicants demonstrate knowledge of the country and meet a standard of integration.Footnote 10
About 2 in 100 foreigners were naturalized in Switzerland in 2000. Although southern European countries like Spain, Italy, and Portugal experienced even lower naturalization rates in the same year, several OECD countries had higher rates with 2.5% in Germany, 3% in the U.S., and 4% in the UK (OECD 2003, 93). Despite the relatively low naturalization rate, Switzerland has experienced a sizeable immigration inflow over the last three decades. By 2000, the share of the non-naturalized immigrant population reached 25% in Switzerland, by far the highest level among all other Western countries except Luxembourg; for comparison, the share of the foreign-born population in the same year was 15% in the United States, 13% in Germany, and 9% in the UK (Dumont, Spielvogel, and Widmaier Reference Dumont, Spielvogel and Widmaier2010).
With the increased immigration flows in recent decades, citizenship policies have become heavily politicized in many Western countries, as right-wing parties have discovered the topic is an effective springboard for mobilizing voters against immigration. In Switzerland, the Swiss People's Party has repeatedly emphasized the need to restrict access to citizenship and immigration inflows, campaigning against “mass naturalizations” with signs that portray brown, black, and white hands snatching Swiss passports. Anti-immigrant political movements such as the Freedom Party in Austria, the National Front in France, the DVU in Germany, and the Danish People's Party in Denmark have similarly mobilized voters against immigration by highlighting the societal dangers of liberal citizenship policies. Conflicts over naturalization policy are therefore an important part of the general phenomenon of anti-immigrant sentiment (Brubaker Reference Brubaker1989; Dancygier Reference Dancygier2010; Givens Reference Givens2007; Howard Reference Howard2009; Koopmans et al. Reference Koopmans, Statham, Giugni and Passy2005).Footnote 11
The Benefits of Swiss Citizenship
Why would immigrants seek Swiss citizenship? For the 1970–2003 period covered in our study, almost all eligible immigrants applying for ordinary naturalization have a settlement permit, and therefore share many of the same rights and benefits as Swiss citizens, such as the right to choose their employers, access to health benefits, the right to travel and return, and the responsibility to pay taxes. However, Swiss citizenship still carries an important symbolic value as a visible marker of full membership in the host country. Citizenship also comes with several tangible benefits. First, only citizens have the right to vote in the many local, cantonal, and federal referendums and elections, and the right to run for office or express their concerns at municipality assemblies.Footnote 12 Second, only children born to citizens automatically receive Swiss citizenship at birth; children born to foreign residents have to apply through naturalization procedures. Third, only citizens have the right to stay in Switzerland indefinitely, while the settlement permit can theoretically be challenged if immigrants return to their home country for more than 6 months (Wanner and Piguet Reference Wanner and Piguet2002, 919). Fourth, certain jobs formally require Swiss citizenship.Footnote 13 Fifth, citizenship may improve immigrants’ economic prospects. Correlational studies from several countries document a positive relationship between naturalization and higher wages and employment (see, for example, OECD 2011). Citizenship can signal to employers higher levels of human capital and lower risk of return migration. Citizenship can also give immigrants an edge in the hiring process when employers discriminate based on nationality. Fibbi, Kaya, and Piguet (Reference Fibbi, Kaya and Piguet2003) document strong discrimination against non-naturalized immigrants among Swiss employers. In sum, a Swiss passport provides more than just symbolic value for immigrants. Citizenship marks the difference between being a tolerated resident who may express her views and being “entitled to have them heard on an equal basis” (Bellamy Reference Bellamy2008, 12).Footnote 14 Naturalization therefore provides an important indicator for the level of societal integration and reciprocity between natives and immigrant populations.
EXPLAINING NATURALIZATION DECISIONS
Immigrants who seek Swiss citizenship have to apply via the ordinary naturalization procedure at three administrative levels: federal, cantonal, and municipal.Footnote 15 Although the federal and cantonal authorities check if an applicant fulfills the basic eligibility requirements such as the residency period and clean criminal record, each municipality evaluates the merits of its applicants and ultimately decides on naturalization requests. In our sample period, 1973–2003, municipalities used a wide variety of institutions to vote on local naturalization requests. Most municipalities used direct democratic arrangements in which citizens voted on applications in popular votes by hand-raising at regular meetings of the citizens’ assembly. Other municipalities delegated the naturalization decision to the elected municipality council where politicians voted on the applications (see Hainmueller and Hangartner Reference Hainmueller and Hangartner2012 for an overview of the various institutions).
In this study, we focus on the relatively small subsample of “ballot box” municipalities that used popular votes with secret ballots to decide on citizenship applications. This institutional arrangement provides perhaps the purest form of direct democracy and resonates with the political culture in Switzerland, which emphasizes local autonomy and direct democratic principles. This arrangement also has historical antecedents in the polis of Athens.Footnote 16 A typical naturalization referendum in our ballot box municipalities involved a two-step process in which citizens received an official voting leaflet with résumés that detailed information about each immigrant applicant (below we provide a list of reported applicant characteristics). Voters then cast a secret ballot to reject or approve each naturalization request, and applicants with a majority of “yes” votes were granted Swiss citizenship. Voting on citizenship requests was part of the typical Swiss direct democratic routine in which referendums were used at regular intervals to decide on a wide variety of municipal, cantonal, and federal matters. Typically, voters cast their ballots at the local polling place, and naturalization referendums appeared on the ballots alongside other referendums that took place on the same day.Footnote 17
Although, historically, relatively few municipalities used naturalization referendums to decide on citizenship applications, the practice recently sparked political debates following media reports about seemingly discriminatory rejection of applicants. One such case was brought before the Swiss Federal Court, which in July 2003 ruled that closed ballot voting for naturalization referendums violates the Swiss Constitution (BGE 129 I 232 and BGE 129 I 217). The Federal Court argued on two different levels. The key reason for ruling against naturalization referendums was that immigrants have the right to appeal rejected applications (BGE 129 I 217), and therefore the decision-making body is obligated to provide justification for the rejection.Footnote 18 Since the very nature of closed ballot referendums means that voters do not have to justify their decisions, the court reasoned that such procedures cannot be used for naturalization. Interestingly, the Federal Court also explicitly mentioned the danger that an applicant may be rejected simply because of her affiliation in a certain “ethnic-cultural group” (BGE 129 I 232: 241), which violates the antidiscrimination clause provided by the Swiss Constitution.Footnote 19 In response to the Federal Court rulings, ballot box municipalities changed their naturalization procedures, and most transferred the authority for naturalization decisions to the municipality council.Footnote 20
The court rulings triggered heated debates about the use of closed ballot naturalization referendums. The Swiss People's Party launched persistent campaigns arguing that “the people” should have full discretion over the naturalization process, and public support for this position remains strong.Footnote 21 In 2006, the Swiss People's Party successfully collected the 100,000 signatures necessary for a federal initiative that, by changing the Swiss Constitution, would grant municipalities full discretion over naturalization (including, of course, secret ballot voting) and remove the rejected applicant's right to appeal. Although this particular initiative “for democratic naturalization” was rejected by 64% of voters in 2008, related popular initiatives at the cantonal level are already being prepared and will ensure continuing politicization of the issue in the ongoing reform debates about naturalization policy.
Empirical Strategy
Closed ballots and voting leaflets are the two main features of the research design that allow us to overcome some of the inferential challenges typically associated with studying immigrant discrimination (Blank, Dabady, and Citro Reference Blank, Dabady and Citro2004). The first feature addresses the measurement problem: the anonymity of closed ballot voting guards against social desirability effects. Unlike surveys in which answers bear almost no real-life consequences for respondents, in our case, preferences are revealed by voters in a real-world setting where the respondents are not aware of the research context and must face the consequences of their voting behavior.
The second feature addresses the causal identification problem: Official voting leaflets summarizing the applicant characteristics were sent to all citizens usually about two to six weeks before each naturalization referendum. Since we retrieved the voting leaflets from the municipal archives, we measure the same applicant information from the leaflets that the citizens observed when they voted on the citizenship applications. Since most voters simply draw on the leaflets to decide on the applicants, this design enables us to greatly minimize potential omitted variable bias and attribute differences in naturalization outcomes to the effects of differences in measured applicant characteristics. For example, imagine that voters are faced with observably similar applicants who only differ in their country of origin, say Italy and (the former) Yugoslavia. If voters reject the Yugoslavian applicants at a higher rate than Italians, then we can attribute the difference in naturalization success to the difference in the country of origin, based on the assumption that a typical voter has no private information about the applicants that goes beyond the information provided in the leaflets and can be used to systematically discriminate between the applicants. Notice that this assumption may be violated for a subset of knowledgeable voters who cast their ballot based on private information about an applicant, for example, because they are friends with the applicant or have been exposed to gossip about the applicant that goes beyond the information provided in the leaflet. We discuss this possibility in separate sections below and show that our main results are robust to hidden bias that may arise from private information.
Before we proceed to the estimation, we remind readers that the scope of our analysis is limited to estimating the effects of applicant characteristics, conditional on the fact that the application was put to a popular vote. The internal validity of our estimates is therefore not compromised by the selection of immigrants into applying for citizenship in the first place. Once an application was put to the vote, we observe and control for the same applicant characteristics that voters learned about from the voting leaflets. That said, the fact that we condition on the voting stage implies that we can detect potential discrimination only among the group of fairly well-integrated immigrants who have completed the eligibility criteria and chosen to apply for citizenship. Our study is not designed to capture additional forms of discrimination that may deter eligible immigrants from applying for citizenship in the first place because they are discouraged by municipality officials, anticipate a discriminatory voting outcome, or deem the costs of applying to be higher than the expected benefits (Steiner and Wicker Reference Steiner and Wicker2004). Our results are therefore best interpreted as a lower bound for the overall prevalence of discrimination against immigrants in the naturalization process of ballot box municipalities.Footnote 22
Data and Sample
To construct our data, we first identified ballot box municipalities that used referendum voting with secret ballots to decide on naturalization requests before the court ruling in 2003. Since data on municipal decision making was unavailable, we compiled a list of all municipality offices and fielded a survey to the Gemeindeschreiber (head secretaries) to collect information about the history of the local naturalization process. This survey yielded an overall response rate of 60%; the coverage was 74% for larger municipalities that had at least 10 naturalizations in 2000.Footnote 23 To complement the data, we also contacted canton officials and searched newspaper archives and municipality websites for additional reports about municipalities with naturalization referendums.Footnote 24 Overall, we identified 44 ballot box municipalities, which are defined as municipalities that (1) used secret ballot naturalization referendums at some point before 2003 and (2) sent voters leaflets with information about applicants. To the best of our knowledge, this constitutes a complete list.Footnote 25 Members of the research team then visited each municipality and extracted the official voting leaflets with applicant information and the vote counts for all ordinary naturalization requests documented in the municipality archive for the period from 1970 to 2003.
Table 1 displays basic information about the sample. Ballot box municipalities were located in seven different cantons, and all were in the German-speaking region. The average municipality had 4,029 registered voters (in 2003), although the size varied considerably from 563 registered voters in Oberiberg to 22,441 voters in Chur. Overall, the sample includes 2,429 naturalization referendums. The period coverage varies somewhat due to differences in data availability, but for most municipalities, we collected data on all naturalization referendums going back to the 1970s and 1980s. The average municipality had about 55 naturalization referendums in our sample period, and the number of referendums was strongly proportional to the municipality size (a bivariate regression indicates that a 1% increase in the number of voters is associated with about a 1.1% increase in the number of applications (t value > 8.6)).
TABLE 1. Ballot Box Municipalities with Naturalization Referendums
Note: Teufen voted on naturalization requests at the polling place. Cantons are Appenzell Outer-Rhodes (AR), Grisons (GR), Lucerne (LU), Nidwalden (NW), Schwyz (SZ), St. Gall (SG), and Uri (UR). Voting location refers to the place where voters submit their secret ballots. Number of voters is measured in 2003. Referendums refers to the total number of collected naturalization referendums that were voted on in a given municipality over the period. See text for details.
Although all municipalities in our sample used popular votes with secret ballots to decide on naturalization requests, the details of the voting process varied somewhat. In 70% of the municipalities, voters cast their secret ballots for the naturalization referendums at the local polling place and usually with their votes on other contemporaneous municipal, cantonal, or federal referendums.Footnote 26 Turnout for such naturalization referendums was about 40% on average, and was mainly driven by the turnout for referendums on the other municipal, cantonal, and federal matters that were voted on the same day.Footnote 27 In a small number of our municipalities, voting on local matters—including citizenship requests—took place at the citizens’ assembly. Secret ballots were also used in these cases, but the average turnout was lower (around 18%) since fewer citizens typically attended the assembly meetings. Two municipalities, Chur and Davos, further restricted the voting on naturalization requests to the Burghers, a select group of about 20% of families who have lived in the municipality for a long time.
Before we turn to the empirical analysis, we consider how the sample of ballot box municipalities compares with other Swiss municipalities. Overall, our sample covers about 4% of all municipalities and about 4% of the Swiss population in our time period. One concern for external validity is that the municipal naturalization procedure is endogenous to the local community's immigration preferences. If, for example, more xenophobic municipalities opted for popular votes because this institution makes it easier to discriminate against immigrants, then our results may be best interpreted as an upper bound for the overall level of discrimination (compared to other Swiss municipalities). Although systematic data on this issue has not been collected, the available anecdotal evidence suggests that municipalities commonly chose their naturalization procedures many decades before our sample period when immigration became politicized and presumably did not radically alter their institutions until forced to do so by the 2003 Federal Court decision (Argast Reference Argast2006).Footnote 28
In Table 2, we compare ballot box municipalities with other Swiss municipalities on various characteristics in 1990 (roughly the middle of our study period). We find that ballot box municipalities are fairly similar to the rest of Switzerland regarding the share of the foreign-born population, the naturalization rate, the proportion of elderly, highly educated, and highly skilled, the female labor force participation, and the unemployment rate. The average vote share of the Swiss People's Party in the 1991 federal elections and support for a federal anti-immigration referendum advocating restrictive immigration laws in 1988 are lower in ballot box municipalities than in the rest of Switzerland, indicating that the former are on average perhaps slightly less xenophobic. The only considerable difference is that ballot box municipalities are somewhat larger on average. Most municipalities in Switzerland are fairly small, while our sample includes a relatively high share of larger towns such as Chur, Emmen, and Schwyz. In sum, the comparison reveals that ballot box municipalities are not very different for a range of relevant characteristics from other Swiss municipalities, in particular municipalities in the German-speaking region.
TABLE 2. Ballot Box Municipalities in Comparison
Note: Covariates are measured in the year 1990 except for SVP vote share which is measured in 1991, the anti-immigration vote share which is measured in 1988, and the age distribution, the proportion of citizens with high education (qualification required for university entrance), and the female labor force participation which are all based on the 2000 census. All means are weighted by the size of the native population except for municipality size.
Outcome and Explanatory Variables
The goal of our analysis is to examine how immigrant attributes affect the outcome of naturalization referendums, conditional on applying. For our main dependent variable, we focus on the proportion of “no” votes, which for each applicant is defined as the fraction of “no” votes to total valid votes. Since referendums were decided by simple majority rule, a naturalization request was rejected if the proportion of “no” votes exceeded 50%. We also replicate the regressions using a binary dependent variable coded as 1 for rejected and 0 for accepted applications; the results from this alternative measure are very similar. We prefer the proportion of “no” votes as our main measure, because it captures information about the intensity with which voters rejected an application.
We measure an array of personal characteristics from the voting leaflets that described each applicant to the voters. We manually code the covariates mentioned in the applicant descriptions, using a consistent set of coding rules. Descriptive statistics for the variables are reported in Appendix A. These variables fall into four categories. The first category involves sociodemographic information such as the applicant's gender, age, marital status, number of children, and our research assistants’ binary coding of the picture (attractive or average).Footnote 29 The second category captures the applicant's integration status. Prospective applicants were interviewed by municipality officials who evaluated whether the applicant was sufficiently well integrated. Officials tested the applicant's language skills and assessed whether the applicant was sufficiently familiar with Swiss habits, customs, and traditions. The results of these assessments were reported in the voter leaflets. We classify applicants according to whether their assessed language skills were described to voters as “excellent,” “good,” “sufficient,” or “insufficient” command of Swiss-German. We also code several variables that measure the applicant's integration status, differentiating between candidates who were described to voters as “assimilated” (1 if assimilated, 2 if highly assimilated), “integrated” (1 if integrated, 2 if highly integrated), “adjusted,” “indistinguishable from a Swiss citizen,” and “familiar with Swiss customs and traditions.” These variables capture the definition of the language and integration requirements in the federal citizenship law.Footnote 30
The third set of characteristics measures the applicant's immigration history. This includes an indicator for whether the applicant was born in Switzerland, a variable that measures the number of years the applicant has lived in Switzerland before the application, an indicator for whether the applicant entered Switzerland as a refugee, and a variable that captures the number of applications (a few immigrants re-applied). We also record the country of origin and classify applicants into the following groups: applicants from richer northern and western European countries (Germany, the UK, Netherlands, Austria, Scandinavia, etc.); southern European countries (Italy, Spain, Portugal); (the former) Yugoslavia;Footnote 31 Turkey; other central and eastern European countries (Hungary, Poland, the Czech Republic, Russia, etc.); Asian countries (Vietnam, Tibet, the Philippines, etc.), and a residual category of other non-European countries (Latin American countries, African countries, the Middle East, etc.). This grouping is based on immigration history and roughly follows the frequency distribution of the origin countries. Applicants from Turkey and (the former) Yugoslavia are coded separately because they constitute the largest recent immigrant groups and are often at the center of the current Swiss immigration debate. Applicants from Italy, Spain, and Portugal are grouped together because these were historically the sending countries of the early waves of immigrants in the 1970s. Overall, about 21% of the applicants in our sample are from richer northern and western European countries, 18% are from southern European countries, 31% are from (the former) Yugoslavia, 15% are from Turkey, 6% are from other central and eastern European countries, 7% are from Asia, and 2% are from other non-European countries.
The fourth set of characteristics measures the applicant's economic and social credentials. We measure the number of years of schooling and code three categories: low education (9 years or less of schooling), medium education (between 10 and 14 years, corresponding to a vocational school degree), and high education (more than 14 years, corresponding to a university degree). We also record the applicant's skill level, which is coded from the first digit of the ISCO-88 occupational classification code. We combine the first two categories (managers and professionals) as highly skilled, groups three to five (technicians and associate professionals, clerical support workers, service and sales workers) as medium skilled, and the rest (craft workers, assemblers, elementary occupations) as low skilled. We also include a binary indicator for whether the applicant's description mentioned any period of unemployment.Footnote 32
RESULTS
Opposition to Naturalization Requests
The box plots in Figure 1 visualize for each municipality the distribution of the proportion of “no” votes in the naturalization referendums over the entire study period. We find a striking variation across and within municipalities (the between variation is about as large as the within variation). The average proportion of “no” votes is 37%, but some municipalities such as Wangen (the average proportion of “no” votes is 58%) are much more opposed to naturalizing immigrants than other municipalities such as Stansstad (the average proportion of “no” votes is 12%). This is remarkable given that these two municipalities are less than 80 kilometers apart, are very similar in size, and have almost the exact same number of naturalization requests (cf. Table 1). This between variation may reflect systematic differences in the pool of applicants, the preferences of the local electorate, and/or residual variation in the institutional process. One apparent difference is that municipalities where voting in the naturalization referendums took place at the citizens’ assembly have somewhat weaker opposition to naturalization compared to the larger group of municipalities where ballots were cast at the local polling place (recall that all municipalities used voting leaflets and closed ballots). This difference is partly driven by the fact that the former group of municipalities has a lower share of applicants from (the former) Yugoslavia and Turkey who—as we show below—have the highest proportion of “no” votes.Footnote 33 Apart from the variation between municipalities, the most striking feature in Figure 1 is that opposition to naturalization requests also varies dramatically within each municipality. The interquartile range of the proportion of “no” votes is about 20 percentage points on average, which suggests that aside from the general differences between municipalities, local voters reveal very heterogenous preferences for different types of immigrants who apply in the same municipality.
FIGURE 1. Opposition to Naturalization Requests
Note: Distribution of proportion of “no” votes across naturalization referendums in each ballot box municipality. Width of box measures the interquartile range, the vertical line decodes the median.
Effects of Immigrant Attributes
Why do natives regard some immigrant applicants as worthy of earning Swiss citizenship while rejecting others? To explore the revealed immigration preferences, we regress the proportion of “no” votes on the applicant characteristics and a full set of municipality fixed effects so that the effects of the applicant characteristics are identified purely based on the within-municipality variation. We also include a full set of decade fixed effects to account for common temporal trends in the naturalization rate and cluster standard errors at the municipality level. Model 1 in Table 3 presents the regression results. To facilitate the interpretation, Figure 2 plots the marginal effects with 95% confidence intervals.
TABLE 3. Effect of Applicant Characteristics on Opposition to Naturalization Request
Note: Point estimates and parenthesized standard errors (clustered by municipality) shown from OLS regressions with municipality fixed effects. The dependent variable is the proportion of “no” votes in Model 1 and a binary indicator coded as 1 for rejected and 0 for accepted applications in Model 2. The reference categories for the various contrasts are as follows: an indicator for the years 1970–1979, applicants with age <20 years, with low education, in low skilled jobs, with “sufficient” command of one of the Swiss languages, who are familiar with “Swiss traditions and customs,” and originating from a richer northern and western European country. P values shown from an F test for joint significance of the batch of variables for sociodemographics, immigration history, economic credentials, language skills, integration status, and country of origin, respectively.
FIGURE 2. Effect of Applicant Characteristics on Opposition to Naturalization Requests
Note: Marginal effect estimates with robust 95% confidence intervals based on OLS regression of proportion of “no” votes on applicant characteristics and municipality and decade fixed effects (Model 1 in Table 3). The reference categories are as follows: applicants with age <20 years, low education, in low skilled jobs, “sufficient” command of one of the Swiss languages, who are familiar with “Swiss traditions and customs,” and originating from a richer northern and western European country. The average proportion of “no” votes is 37%.
The applicant's country of origin has the most important impact on the outcome of naturalization referendums. Holding other characteristics constant, the proportion of “no” votes for applicants from (the former) Yugoslavia is about 15 percentage points higher (t value > 14.5) than for observably similar applicants from richer northern and western European countries (the reference category). This constitutes an increase of about 40% over the average proportion of “no” votes. Turkish applicants fare just as poorly with an average proportion of “no” votes that is 13 percentage points higher (t value > 10.7), a 35% increase over the average level of opposition. We find similar penalties of about 6 percentage points (t value > 5.2) for applicants from other central or eastern European countries, about 3.5 percentage points (t value > 2.3) for applicants from Asian countries, and about 7 percentage points (t value > 4.7) for applicants from other non-European countries. Applicants from southern European countries are the only ones who earn a lower proportion of “no” votes compared to observably similar applicants from richer northern and western European countries, although the premium is small at about 1.4 percentage points and falls short of conventional significance levels. Taken together, country of origin accounts for about 40% of the within-municipality variation in the levels of opposition to naturalization requests.
What else affects the likelihood of being rejected for citizenship? We find that sociodemographics such as age, gender, marital status, attractiveness, or children have little effect overall. If anything, middle-aged applicants (21–60 years) earn a penalty of about 2 percentage points compared to the reference category of younger applicants (≤20 years), but there is no such penalty for older applicants (≥60 years).
The applicant's immigration history matters somewhat for the naturalization success. Applicants who are born in Switzerland earn a premium of about 2 percentage points (t value > 2.7) compared to foreign-born applicants. This effect may seem surprisingly small from a comparative perspective given that place of birth is the main requirement for granting citizenship in jus soli countries such as the United States. Natives also reward applicants for longer residency in Switzerland, but the rewards are rather paltry. A 10-year increase in residency decreases the proportion of “no” votes by merely 1.7 percentage points on average (t value > 4.4); given this modest effect size, even an implausibly large increase from the shortest to the longest residency in our sample (75 years) would not be sufficient to compensate for the disadvantage that applicants from (the former) Yugoslavia and Turkey face due to the country of origin effects. History as a refugee and the number of prior applications have no consistent effects on the level of opposition.
We find that voters systematically reward applicants for better economic credentials. Applicants in the highest occupational skill category (managers, senior officials, professionals, etc.) earn a premium of about 2.6 percentage points compared to applicants in the lowest skill category (elementary occupations, assemblers, etc.). Similarly, immigrants with high educational attainment earn a premium of about 1.2 percentage points compared to observably similar applicants with low education levels, again a very modest gain compared to the magnitude of the origin disadvantages.Footnote 34 Applicants who experienced a period of unemployment face a penalty of about 6 percentage points, suggesting that voters have a rather strong distaste for such applicants. This effect is fairly robust but not very precisely estimated since there are few applicants with unemployment experience in our sample.
Assessed language skills have almost no effect on naturalization success. There are no significant differences in the proportion of “no” votes for applicants who are described to voters as having sufficient or a good command of German. Even applicants who speak the local Swiss-German dialect perfectly (“perfekt Mundart”) earn no significant advantage. The only exceptions are applicants for whom officials describe their language skills as insufficient. The proportion of “no” votes for this group is 20 percentage points higher (t value > 2) compared to applicants with sufficient language skills (the reference category). We need to be careful in drawing strong conclusions from this finding, since we have only six applicants with insufficient skills in our sample; thus the estimate is very imprecise with a 95% confidence interval of [1.2; 39.5].Footnote 35 Moreover, we cannot reject the null that all the language dummies are jointly insignificant at conventional levels, which adds further evidence that language skills are inconsequential for naturalization success once other applicant characteristics are controlled for.
We find some evidence that the assessed immigration status affects opposition to naturalization requests, but again the impact of this factor is fairly small. We find no significant differences in the proportion of “no” votes for applicants who are described as “integrated,” “adjusted,” or “well familiar with Swiss traditions and customs” (the reference category), but immigrants who are described to voters as completely “indistinguishable” from a Swiss native receive a premium of about 3 percentage points.
Model 2 in Table 3 replicates the same specification with a linear probability model that replaces the proportion of “no” votes with a binary dependent variable that measures whether the naturalization application was rejected or accepted. The overall results from this model are very similar. The main country of origin effects are, if anything, slightly stronger: applicants from Turkey and (the former) Yugoslavia have about a 30-percentage-point higher probability of being rejected for citizenship compared to observably similar applicants from richer northern and western European countries—a whopping 120% increase over the average probability of being rejected. The only difference in the origin effects is that the penalties for the small groups of applicants from Asian and other non-European countries are no longer significant at conventional levels, which suggests that for these groups the findings are not robust.
Overall, the results from both measures yield a clear picture. Naturalization referendums are mostly decided based on the applicant's country of origin. The origin effects are more important than other applicant characteristics for which we find significant effects, such as economic credentials and immigration history. Once these factors are controlled for, language skills and integration status play only a minor role.
Robustness: Private Information
How robust are the country of origin effects? One potential concern is that some voters may act upon private information that they possess about certain applicants that goes beyond the information provided in the leaflet. The effect of private information is theoretically ambiguous and depends on whether it leads voters to be more or less inclined to vote for a particular applicant. For example, an immigrant whose application is being voted on could try to mobilize her Swiss friends to vote for her. Many of the differences in such mobilization efforts are presumably captured by various covariates (e.g., the years since arrival, born in Switzerland, education level, etc.). However, residual differences in unobserved mobilization efforts may still bias our country of origin effects if, for example, immigrants from Turkey and (the former) Yugoslavia are more or less effective in activating their network of friends to vote for them compared to immigrants from richer northern and western European countries.
Another form of private information that could affect the naturalization outcome is gossip. If an applicant has a personal enemy in town who spreads negative gossip against her that extends beyond the information on the leaflet, then this could sway some voters to reject the application. Such gossip effects do not necessarily constitute an ordinary omitted variable bias. In fact, to the extent that negative gossip is systematically targeted toward applicants from particular origin groups, such as immigrants from (the former) Yugoslavia and Turkey, it should be considered part of the origin effect that the regression intends to capture since in such cases applicants face a penalty because of their origin.
We can exploit variation in municipality size to test how sensitive our results are to potential bias induced by the presence of private information. Assume that voters in each municipality can be partitioned into two types: “knowledgeable” voters who are exposed to private information beyond the applicant characteristics reported in the leaflet, and “ignorant” voters who are exposed only to the information in the leaflet.Footnote 36 Since the reach of private information is limited, we can expect that the proportion of knowledgeable voters declines as the size of the electorate increases. For example, an applicant may be able to mobilize a fixed number of friends to vote for her, but the impact of these votes on the outcome of the referendum will be increasingly negligible in larger municipalities. Similarly, gossip against an applicant may reach a considerable fraction of voters in small municipalities, but is unlikely to affect a significant fraction of voters in large municipalities such as Chur, Emmen, and Schwyz.
In Table 4, we replicate our benchmark model for three subsamples: large municipalities with more than 5,000 voters, smaller municipalities with less than 5,000 voters, and municipalities where votes were cast at the polling place. If private information is an important confounder, then we would expect that the origin effects vary strongly between small and large municipalities. Instead, we find that the origin effects are almost identical across subsamples for the proportion of “no” votes (Models 1–3) and the binary rejection indicator (Models 4–6). If anything, the origin disadvantage for immigrants from (the former) Yugoslavia and Turkey is slightly larger in the larger municipalities. This strongly suggests that our results are not driven by omitted variable bias that may arise from private information.
TABLE 4. Effect of Applicant Characteristics on Opposition to Naturalization Request in Large and Small Municipalities
Note: Point estimates and parenthesized standard errors (clustered by municipality) shown from OLS regressions with municipality fixed effects. The dependent variable is the proportion of “no” votes in Models 1–3 and a binary indicator coded as 1 for rejected and 0 for accepted applications in Models 4–6. Models 1 and 4 are based on large municipalities with more than 5000 voters in 2003, Models 2 and 5 are based on small municipalities with less than 5000 voters in 2003, and Models 3 and 6 are only based on municipalities where the ballots were cast at the polling place. The reference categories for the various contrasts are as follows: an indicator for the years 1970–1979, applicants with age <20 years, with low education, in low skilled jobs, with “sufficient” command of one of the Swiss languages, who are familiar with “Swiss traditions and customs,” and originating from a richer northern and western European country. For all models, the standard errors are virtually identical if the block (by municipality) bootstrap is used, and about 50% smaller if no clustering is used.
The limited effects of private information are perhaps not surprising for several reasons. First, during our fieldwork we did not find any systematic evidence that applicants invested significant effort to campaign for their naturalization requests. Second, we may expect that large effects from gossip are not systematic, but rather limited to a few extreme cases. Third, for voters in our ballot box municipalities, voting in naturalization referendums was part of their usual direct democratic routine in which they were frequently asked to decide on municipal, cantonal, and federal referendums dealing with all sorts of issues, and turnout for local naturalization referendums was commonly driven by turnout for other, potentially more important, referendums voted on the same day. Presumably then, voters rarely had an incentive to treat naturalization referendums as a special event and seek out additional information about applicants beyond what was provided in the leaflets.
Robustness: General Hidden Bias
An alternative method for examining how robust the results are to hidden bias from unobserved confounders is to conduct a formal sensitivity analysis following the methods developed in Rosenbaum (Reference Rosenbaum2002). We compare applicants from (the former) Yugoslavia and Turkey with applicants from richer northern and western European countries and first nonparametrically adjust for the observed applicant characteristics from the benchmark model using Genetic Matching (Diamond and Sekhon Reference Diamond and Sekhon2006; Sekhon Reference Sekhon2011). The origin effect estimates from the (bias-adjusted) matching are shown in Table 5. They confirm that applicants from Turkey and (the former) Yugoslavia receive a much higher proportion of “no” votes and are much more likely to be rejected compared to observably similar applicants from richer northern and western European countries (the effect sizes are very similar compared to those from the regressions in Table 3). Next, we conduct Rosenbaum sensitivity tests to pinpoint the degree of hidden bias due to an unobserved confounder that would be needed to explain away these origin effects.
TABLE 5. Effect of Applicant Characteristics on Opposition to Naturalization Request: Matching Results and Sensitivity Analysis
Note: The origin penalty refers to the estimated difference in the proportion of “no” votes (upper panel) or the probability of being rejected (lower panel) between applicants from (the former) Yugoslavia or Turkey and observably similar applicants from richer northern and western European countries. Estimated differences are based on average treatment effect (ATE) estimates from 1:1 bias-adjusted genetic matching (with replacement) with Abadie-Imbens standard errors. For all models, only applicants originating from richer northern and western European countries or the former Yugoslavia and Turkey are used. The matching and bias adjustments include all covariates from the benchmark model. Below the ATE estimates and standard errors the table also displays results for the Gamma estimates from Rosenbaum sensitivity tests. The Rosenbaum Gamma (Γ) measures the degree of departure from a study that is free of bias; it is equivalent to the size of the log of the coefficient on an unobserved confounder. For the proportion of “no” measure, these tests refer to the lowest Rosenbaum Gamma (Γ) at which the upper bound of the p value from the Wilcoxon Sign Rank Test turns insignificant (p > 0.05). For the binary rejection measure the results refer to the lowest Gamma at which the upper bound of the McNemar's test turns insignificant (p > 0.05).
Borrowing from the causal inference literature, we can define applicants who report (the former) Yugoslavia or Turkey as their origin in the leaflet as the “treatment group,” and applicants who report a richer northern and western European country as their origin in the leaflet as the “control group” (the “treatment” therefore consists of manipulating the reported origin that voters observe). The degree of hidden bias is measured with the Rosenbaum Gamma, Γ. This parameter is defined as the upper bound on the degree to which observably similar applicants from the two different origin groups may nonetheless differ in their a priori odds of receiving the treatment due to differences in a powerful omitted variable that is a near-perfect predictor of naturalization success.Footnote 37 Notice that if Γ = 1, the comparison between observably similar applicants from the two origin groups is assumed to be free of hidden bias as the odds of treatment assignment is identical for the groups (as in a randomized experiment). A value of Γ = 2 allows for considerable hidden bias; of two observably similar applicants, one could be up to twice as likely to receive the treatment due to differences on the omitted variable.
How sensitive are the matching estimates to hidden bias? The rows below the effect estimates in Table 5 report the lowest Γ at which the effect estimates turn insignificant at the 95% confidence level (Keele Reference Keele2010). We find that the results are very robust to hidden bias with Γ values ranging between 9 and 11. This implies that only an enormous hidden bias could explain away the origin effects. Net of the observed covariates, an unmeasured confounder would need to be a near-perfect predictor of naturalization success and produce a 9- to 11-fold increase in the odds of treatment assignment. This level of insensitivity to hidden bias far exceeds those typically found for social science studies where Γ values are commonly in the range of 1–2 (DiPrete and Gangl Reference DiPrete and Gangl2004; Keele Reference Keele2010; Rosenbaum Reference Rosenbaum2002; Reference Rosenbaum2005).Footnote 38
Robustness: Further Issues
We conducted various additional checks that further support the robustness of our main findings (results are reported in Appendix B). In particular, Tables B.1 and B.2 replicate the benchmark models for the different municipality samples (all, large, and polling place) using several other control strategies, including year fixed effects, quadratic time trends, and linear and quadratic municipality specific time trends to absorb smooth trends in unobserved confounders that vary at the municipality level. The results are remarkably stable across the specifications for all subsamples. In Table B.3, we control for the lagged share of applicants from (the former) Yugoslavia and Turkey in each municipality (three-year moving average) and the number of applicants who appear on the ballotsFootnote 39 and the results are unaffected. Next, we estimate a series of models where in each municipality we regress the proportion of “no” votes on the country of origin indicators and a streamlined set of covariates (gender, marital status, age, born in Switzerland, years of residency, and years of schooling) to accommodate the smaller sample sizes (we exclude municipalities with fewer than 15 applications). In Figure B.1, we plot the municipality-specific origin effects with 95% robust confidence intervals for applicants from Turkey and (the former) Yugoslavia. The effect estimates are positive, sizable, and significant, revealing that Turkish and Yugoslavian applicants face a considerable penalty compared to observably similar applicants from richer northern and western European countries in almost all of the ballot box municipalities; the estimated increase in the proportion of “no” votes is between 12 and 24 percentage points in most municipalities (about a 32–64% increase over the average proportion of “no” votes). Finally, we investigate the model dependency of the origin effects in the full sample across a wide range of possible specifications. In particular, we fit 15,000 regressions of the proportion voting “no” on the origin indicators and a subset of the control variables that we randomly draw from the set of all possible subsets of the covariates from the benchmark model and all their first-order interactions (and squared terms for continuous variables). Figure B.2 shows the distribution of the origin effect estimates. The results from this extreme bound analysis suggest that the origin effects are remarkably robust with effect sizes clustering closely around the magnitudes estimated in the main regression.
Comparison with Attitudinal Data
We briefly contrast our findings with comparable public opinion data. In 2002, the European Social Survey, a widely used survey in the comparative literature on immigration attitudes (Ceobanu and Escandell Reference Ceobanu and Escandell2010), measured the views of Swiss voters about immigrants from different source countries. Seventy-four percent of respondents said that they are in favor of allowing more immigrants from poorer European countries to come and settle in Switzerland, while only 69% of respondents supported more immigration from richer European countries. Sixty-nine percent were in favor of more immigrants from poorer non-European countries, and only 64% in favor of immigrants from richer non-European countries. Clearly, our behavioral data suggest different preferences, as applicants from Turkey, (the former) Yugoslavia, and other central or eastern European countries are rejected at much higher rates than applicants from richer northern and western European countries. These differences may stem from social desirability bias if survey respondents are unwilling to admit that they actually prefer immigrants from richer origins. More recent survey data point to a similar contrast between public opinion and behavioral data in voter preferences. In a 2008 poll, 88% of Swiss voters agreed that naturalization decisions should be free of discrimination based on country of origin.Footnote 40 This sharply contrasts with our finding that country of origin is the single most important determinant of naturalization success.
INTERPRETATION
The results raise two important questions for the interpretation. First, does the consistently higher level of opposition for applicants from certain origins imply that native voters discriminate against these applicants? If so, how do our findings relate to theories of discrimination?
Country of Origin Effects as Discrimination
Our results imply that applicants who differ only in their country of origin and are otherwise similar on all measured characteristics face much different probabilities of being rejected when applying in the same municipality, under the same institutional rules, and in the same time period. Are these findings consistent with a nondiscriminatory selection rule that is blind to the origin of the applicant? As detailed above, the naturalization law implies that candidates for naturalization have to meet a list of requirements. Federal authorities always check that applicants meet the residency and criminality requirements so the differential treatment cannot be accounted for by these factors (voters are aware that applicants have cleared these hurdles). Moreover, language requirements and integration status and familiarity with Swiss values and traditions are controlled for in our analysis using assessments by the municipality officials who interview the applicants and inform voters about the assessments. A selection rule that is neutral to country of origin would therefore imply that, conditional on these observed characteristics, the probability of being rejected should be similar across applicants of different origins. Instead, we find that otherwise observably similar applicants face dramatically different rejection rates depending on their country of origin. In fact, the applicants’ country of origin is by far the most important predictor of the naturalization outcome compared to all other observed characteristics, which implies that voters distinguish between applicants in part based on their origin. This matches the legal definition of discrimination as stated in the antidiscrimination clause in the Swiss Constitution.
Causes of Discrimination
As Pager and Shepherd (Reference Pager and Shepherd2008, 12) put it in their recent review, “Measuring the prevalence of discrimination is difficult; identifying its causes is far more so.” Nonetheless, it is worth asking which mechanisms may underlie the discrimination that we find. Below, we try to distinguish between the main theories of discrimination by testing additional observable implications.
The literature commonly distinguishes two main theories: statistical discrimination and taste-based discrimination.Footnote 41 Theories of statistical discrimination (Arrow Reference Arrow1972; Phelps Reference Phelps1972) emphasize how discrimination can arise due to problems of limited information. Assume that voters have a utility function such that they hold no animus against particular origin groups but care only about naturalizing applicants with a sufficiently high level of integration (which may encompass sufficient language skills and familiarity with Swiss values and traditions as stipulated by the naturalization law). Voters also take it for an empirical fact that the average integration levels vary by origin, for example, voters presume true that the integration level among immigrants from Turkey and (the former) Yugoslavia is lower on average compared to immigrants from richer northern and western European countries.Footnote 42 Since integration levels are inherently difficult to measure, these voters face a signal extraction problem when deciding how to vote in the naturalization referendums. The voters would like to evaluate an applicant based purely on her true level of integration, but this true level is unobserved and needs to be inferred from the noisy signal of the assessed integration level reported in the voter leaflet. To solve this signal extraction problem, rational voters place weight on the noisy signal and prior information about the origin group to which an applicant belongs. In other words, it is optimal for voters to use their knowledge about an origin group generally to evaluate whether a particular applicant from that group is likely to be sufficiently integrated since the most accurate estimate of an applicant's actual integration level is a weighted average of his assessed integration level from the leaflet and the average integration level of his origin group. This may explain why, even when faced with applicants who report the same assessed integration level, voters still estimate the actual integration level of individual Turkish and (former) Yugoslavian applicants as relatively lower, because voters evaluate such applicants more like the average Turkish or Yugoslavian immigrant.Footnote 43
If the differences in naturalization success stem from statistical discrimination, then we would expect that Turkish and Yugoslavian applicants earn relatively higher returns for other observable credentials such as education that can be informative about the applicant's actual integration status. From the perspective of voters who are solving the signal extraction problem, learning that a particular applicant is highly educated or highly skilled should lead to a larger increase in the estimated integration level for applicants from (the former) Yugoslavia and Turkey compared to applicants from richer northern and western European countries, if voters presume that the former origin group has a lower average integration level. To test this implication, we re-estimate our benchmark model while allowing the effects of economic credentials to vary across the applicant's origin. For tractability, we restrict the sample to contrast applicants from richer northern and western European countries and applicants from (the former) Yugoslavia and Turkey, and we combine the latter into a single origin group.
Table 6 displays the results. Consistent with statistical discrimination, the interaction terms between origin and high occupational skills or high educational attainment are negative and significant, which implies that (former) Yugoslavian and Turkish applicants earn relatively higher returns for economic credentials. For example, Model 1 reveals that among immigrants from (the former) Yugoslavia and Turkey the average proportion of “no” votes for highly skilled applicants is about 5.3 percentage points lower compared to low-skilled applicants (t value > 2.8). In contrast, skill level has no impact on the proportion of “no” votes among immigrants from richer northern and western European countries (the effect estimate on the lower-order term is close to zero). We find similar differential returns to education (Model 2), and the results are similar when we restrict the sample to polling place municipalities (Models 3 and 4). Although these results are consistent with statistical discrimination, they also reveal that statistical discrimination can only partly account for origin-based discrimination. Even taking the differential returns to economic credentials into account, highly skilled applicants from (the former) Yugoslavia and Turkey still face a penalty that is about 7–8 percentage points higher compared to observably similar applicants from richer northern and western European countries. This observation suggests that statistical discrimination may account for at least about 40% of the overall origin effect.
TABLE 6. Statistical Discrimination: Interaction Effect of Economic Credentials and Country of Origin on Opposition to Naturalization Request
Note: Point estimates and parenthesized standard errors (clustered by municipality) shown from OLS regressions with municipality fixed effects. For all models, only applicants originating from richer northern and western European countries or (the former) Yugoslavia and Turkey are used. Models 1 and 2 are based on the full sample of ballot box municipalities, Models 3 and 4 are based on municipalities where the ballots were cast at the polling place.
What else besides statistical discrimination may explain the discriminatory outcomes that we find? The other leading alternative are taste-based theories of discrimination (Becker Reference Becker1971; Quillian Reference Quillian2006) that operate on the assumption that natives have real prejudice and animus against particular origin groups. In these models, discrimination arises not from uncertainty, but from presumed xenophobic tastes built into the voters’ utility function. Such xenophobic tastes may include a wide range of feelings or beliefs that certain immigrant groups, such as applicants from (the former) Yugoslavia and Turkey, are culturally inferior and socially undesirable, and perceptions that such groups threaten the prevailing way of life or even reject Swiss culture and refuse to assimilate (Allport Reference Allport1979; Blumer Reference Blumer1958; Dovidio et al. Reference Dovidio, Kawakami, Johnson, Johnson and Howard1997; Kinder and Sears Reference Kinder and Sears1981; Tajfel Reference Tajfel1982). Testing the model of taste-based discrimination requires a measure of voters’ xenophobic tastes. We cannot measure the tastes of individual voters, but we can exploit the variation in xenophobic tastes across municipalities using municipality-level vote shares from a 1982 federal anti-immigration referendum that called for removing restrictions on immigration. The share of “no” votes from this referendum, which occurred close to the beginning of our sample period, is a good proxy for the xenophobic tendencies of the local electorates. If the discrimination is driven by taste, we would expect that the intensity of origin-based discrimination—as measured by the size of the disadvantage for Yugoslavian and Turkish applicants—increases with the local level of xenophobia.
In Table 7, we test this hypothesis by re-estimating our benchmark model with an interaction term between the anti-immigrant vote share and the country of origin indicator (we again restrict the sample as above). We use a fixed effects specification (Models 1 and 3) and a multilevel model in which we also allow the origin effect to vary by municipality (Models 2 and 4). The results support taste-based discrimination as the interaction terms are positive and highly significant across specifications. The magnitudes imply that a one percentage point increase in the anti-immigration vote share is associated with about a 0.5–0.7 percentage point increase in the relative origin-based disadvantage for applicants from (the former) Yugoslavia and Turkey. This result suggests that taste-based discrimination may account for about 60% of the overall origin-based discrimination. To give a clear interpretation of the interaction in Figure 3 we plot the estimated origin-based disadvantage as a function of the (demeaned) anti-immigrant vote shares (based on Model 2). In municipalities with the highest levels of xenophobia, Yugoslavian and Turkish applicants face a penalty of about 20–30 percentage points compared to observably similar applicants from richer northern and western European countries. This penalty is much lower at about 3–10 percentage points in the least xenophobic municipalities. Also consistent with taste-based discrimination, the models reveal that the level of xenophobia is uncorrelated with the proportion of “no” votes for applicants from richer northern and western European countries (the lower-order terms for the anti-immigrant vote are close to zero). As a robustness check, Table B.2 in the appendix replicates the models using two alternative measures of xenophobic tastes, the municipality-level vote shares from similar federal anti-immigration referendums in 1983 and 1988, and the patterns are very similar. Taken together, these results indicate that origin-based discrimination is largely driven by local voters’ xenophobic prejudice. We would not expect such strong patterns if the discrimination were purely statistical.Footnote 44
TABLE 7. Taste-Based Discrimination: Interaction Effect of Xenophobic Preferences and Country of Origin on Opposition to Naturalization Request
Note: Point estimates and parenthesized standard errors shown. Models 1 and 3 are OLS regressions with municipality fixed effects and standard errors clustered by municipality. Models 2 and 4 are restricted maximum likelihood multilevel regressions with varying intercepts and slopes for the country of origin indicator. The random effects are allowed to vary by municipality and their correlation is unrestricted. The random effect models include dummies for the municipality specific institutional features (i.e., voting location and constituency, respectively). For all models, only applicants originating from richer northern and western European countries or (the former) Yugoslavia and Turkey are used. Models 1 and 2 are based on the full sample of ballot box municipalities, Models 3 and 4 are based on municipalities where the ballots were cast at the polling place.
FIGURE 3. Municipality Specific Country of Origin Effects and Local Xenophobia
Note: Marginal effect estimate for the interaction of Yugoslavian and Turkish origin and municipal anti-immigrant vote share based on multilevel regression. The plot builds on the estimates of multilevel Model 2 of Table 7 which regresses the proportion of “no” votes on applicant characteristics, municipal characteristics, and (de-meaned) anti-immigration vote share in the 1982 referendum. Random intercepts and slopes for the country of origin indicator allowed to vary unrestrictedly by municipality. Random effects for each municipality are shown along with 90% prediction intervals.
The results thus far suggest that origin-based discrimination in naturalization referendums is driven by statistical and taste-based discrimination, and that the latter mechanism is perhaps more important. Although models of taste-based discrimination commonly treat tastes as exogenous, one interesting follow-up question involves the origins of voters’ xenophobic prejudice. A detailed examination of this issue is beyond the scope of this study, but the data allow us to briefly speculate about potential mechanisms.
One potential hypothesis is that animus against applicants from (the former) Yugoslavia and Turkey is driven by differences in religion if voters perceive these immigrants as Muslim (Adida, Laitin, and Valfort Reference Adida, Laitin and Valfort2010). We cannot test this claim directly since the voting leaflets typically do not contain information about the applicant's religion. However, for the group of applicants from (the former) Yugoslavia we can exploit the fact that applicants from Bosnia and Herzegovina, Kosovo, and Macedonia are more likely to be Muslim compared to applicants from Serbia and Montenegro, Macedonia, and Slovenia, since the share of Muslims is much higher in the former group of countries. If the prejudicial taste against applicants from (the former) Yugoslavia is primarily driven by Islamophobia, we would expect strong differences in the origin-based discrimination that these groups face. However, when we replicate the benchmark model with dummy variables for the groups of high and low Muslim countries, we find that the country of origin disadvantages are fairly similar (Table B.4 in the Appendix). Opposition to naturalization requests is slightly higher for applicants from high compared to low Muslim origins, but the differences are substantivally small and not consistently significant. Taken together, these admittedly limited tests suggest that Islamophobia is perhaps not the main source of animus propelling taste-based discrimination.
A perhaps more prominent explanation in the literature is group conflict theory, which posits that xenophobic prejudice arises as a defensive reaction by natives who view immigrant out-groups as a threat to the natives’ dominant position (Blalock Reference Blalock1967; Blumer Reference Blumer1958; Quillian Reference Quillian1995). Such threats may operate at an economic level, if natives are concerned about competing with immigrants for jobs and other scarce resources (Olzak Reference Olzak1994; Scheve and Slaughter Reference Scheve and Slaughter2001), or at a cultural level, if natives fear that particular immigrant groups “constitute a threat to the collective identity and the cultural, national, and ethnic homogeneity of the society” (Semyonov, Raijman, and Gorodzeisky Reference Semyonov, Raijman and Gorodzeisky2006, 428).Footnote 45 One of the main predictions from the group conflict model is that prejudice dynamically responds to a rising threat associated with rapid growth in the size of the out-group. Another prominent view in the literature is contact theory, which implies the opposite prediction. According to this model, an increase in the out-group can actually decrease prejudice over time because more frequent intergroup interactions help to dissolve natives’ stereotypes and preconceived judgments (Allport Reference Allport1979; Pettigrew and Tropp Reference Pettigrew and Tropp2006).
Our data provide a unique opportunity to evaluate these claims, since we behaviorally measure discrimination against immigrants from multiple origin groups over time, and Switzerland—apart from the general increase in immigration—also experienced rather dramatic shifts in its immigration composition during our time period. Figure 4 shows the trends in the shares of origin groups of the total foreign-born population over the 1980–2003 period.Footnote 46 The threat mechanism suggests that discrimination against applicants from Turkey and (the former) Yugoslavia should increase over time, since this immigrant group is relatively new and its relative size doubled during our time period. In contrast, discrimination against applicants from southern European should abate over time, since this origin group has a much longer immigration history in Switzerland and the group's relative size decreases considerably during our time period. The contact mechanism anticipates the opposite pattern.
FIGURE 4. Immigration Trends in Switzerland
Note: Plot shows the shares of immigrants from different origins on the total foreign-born population (Source: PETRA Database from Swiss Federal Office of Statistics).
In Table 8, we re-estimate the benchmark model for three periods (1970–1989, 1990–1999, and 2000–2003); Models 1–3 use all available data, and Models 4–6 restrict the sample to municipalities for which data are available in all periods. The results are more consistent with the dynamic threat mechanism as discrimination against different immigrant origin groups is positively correlated with growth in their relative sizes. Although the origin penalty for Turkish and (former) Yugoslavian applicants is about 5–7 percentage points in the 1970s and 1980s, it climbs to about 13–18 percentage points in the post-1990 period following the rapid influx of immigrants from these countries. In contrast, discrimination against applicants from southern Europe strongly decreases over time as the relative size of this group plummets. In fact, in the most recent period applicants from this group earn a premium of 4–5 percentage points compared to observably similar applicants from richer northern and western European countries (the rejection rates for this latter group remain fairly stable). This bifurcation indicates that natives’ discriminatory attitudes toward particular out-groups are not static, but dynamically correlate with the differential trends in the relative sizes of the origin groups. The threat mechanism might operate as a two-way street: rapid growth in the relative size of an out-group can stoke prejudice against it, but rapid decline in the group's relative size may also help to attenuate prejudice over time.Footnote 47
TABLE 8. Dynamics of Country of Origin Effects
Note: Point estimates and parenthesized standard errors (clustered by municipality) shown from OLS regressions. The models replicate the benchmark model for different time periods. Model 1 and 4 are based on applications in the years 1970–1989, Models 2 and 5 are based on applications in the years 1990–1999, and Models 3 and 6 are based on applications in the years 2000–2003. Models 1–3 are based on all available data; Models 4–6 are restricted to the same 29 municipalities for which data are available in all periods. All models are estimated with the full set of benchmark covariates and fixed effects for each municipality (coefficients not shown). Models 1 and 4 also include a decade fixed effect for the 1980s.
CONCLUSION
Many studies of anti-immigration sentiment are based on subjective survey data and limited to examining attitudes toward immigration in general. Using behavioral data from naturalization referendums in Swiss municipalities, our analysis demonstrates that immigration preferences in fact vary dramatically regarding different types of immigrants and over time. In particular, we show that observably similar applicants face dramatically different rejection rates depending on the applicants’ country of origin, which matches the legal definition of discrimination according to the Swiss Constitution.
In particular, we find that immigrants from (the former) Yugoslavia and Turkey face the highest probability of being rejected compared to immigrants from all other origins. Other immigrant attributes, such as immigration history and economic credentials, also affect naturalization success, but to a much lesser degree. Language skills and an immigrant's assessed immigration status play almost no role in naturalization success. Disentangling the causal pathways of origin-based discrimination, the results suggest that about 40% of the measured differences in the opposition to naturalization requests may be attributable to statistical discrimination and about 60% to taste-based discrimination. Overall, these results are consistent with arguments linking anti-immigrant sentiment to widespread prejudice against particular immigrant groups (Dustmann and Preston Reference Dustmann and Preston2007; Hainmueller and Hiscox Reference Hainmueller and Hiscox2007; Kinder and Sears Reference Kinder and Sears1981; Sides and Citrin Reference Sides and Citrin2007).
How generalizable are our results? The data indicate that the municipalities in our sample are not very different from municipalities in other parts of Switzerland, and our results are consistent across subsamples. Hence, our main findings may generalize to the country as a whole, or at least to the German-speaking regions. External validity beyond Switzerland is much more difficult to judge, and we advise against over- or undergeneralizing the results to a cross-national context. On the one hand, Switzerland is unique in many respects. Given the fierce immigration debates and the lasting success of the Swiss People's Party, the discrimination against immigrant out-groups that we found might represent an upper bound compared to other countries where right-wing parties have been less successful in recent decades. On the other hand, Switzerland, given its exceptionally large immigrant population and varied history involving the peaceful integration of four different language traditions, is a vanguard, rather than a laggard, country in terms of cultural heterogeneity (Linder Reference Linder2010). Another interpretation of our findings, then, is that cultural and ethnic tensions between the native and immigrant populations in Switzerland are perhaps not that different from the xenophobic rifts emerging in many other European countries similarly struggling with conflicts about immigration and citizenship policies. Studies of anti-immigrant sentiment have consistently shown that prejudices against immigrants run high in many other European countries (Dustmann and Preston Reference Dustmann and Preston2007; Hainmueller and Hiscox Reference Hainmueller and Hiscox2007), and right-wing parties have also experienced significant electoral support in several countries. It is therefore not readily apparent why we would expect to see dramatically different results if voters in these countries were given the opportunity to voice their preferences regarding immigrants in popular votes. In the end, external validity is of course best addressed by comparing the results of several internally valid studies conducted in different circumstances and at different times, and thus we hope this study will stimulate more behavioral research on anti-immigrant sentiment and discrimination in other countries.
Our results also have implications for the sizable literature on direct democracy in which scholars still profoundly disagree about the impact of direct democracy on minority interests (Frey and Goette Reference Frey and Goette1998; Gamble Reference Gamble1997; Hajnal, Gerber, and Louch Reference Hajnal, Gerber and Louch2002; Maskin and Tirole Reference Maskin and Tirole2004). Given the specific focus of our study, concluding that referendums generally harm minorities would be unwise. However, our microlevel results shed light on at least one central issue: access to citizenship in a country where local voters used their direct democratic means to discriminate against particular minority groups. The profound disadvantage in naturalization success for applicants from (the former) Yugoslavia and Turkey emphasizes concerns raised by previous studies that primarily out-group minorities suffer when their civil rights are put to a popular vote (Vatter and Danaci Reference Vatter and Danaci2010). These concerns are further supported by a follow-up study in which we exploit the Federal Court ruling and examine how naturalization outcomes changed, once the ballot box municipalities were forced to transfer the decision about naturalization requests from popular referendums to an elected municipality council (Hainmueller and Hangartner Reference Hainmueller and Hangartner2012). There we found that naturalization rates soared once accountable legislators, rather than the people, voted on naturalization requests. Moreover, the increase in naturalization rates resulting from replacing direct with representative democracy was largest for immigrants from (the former) Yugoslavia and Turkey, the groups that fared the worst in naturalization referendums. More research is clearly needed, but taken together, this evidence helps to empirically ground the heated policy debates about naturalization procedures in Switzerland and raises concerns about recent proposals to restore secret ballot referendums to decide on naturalization requests.
Comments
No Comments have been published for this article.