In many countries around the world, right-wing populists have turned against immigrants and their descendants, describing them as burdens on public welfare, illegal intruders, unfair competitors for jobs, abusers of the asylum system, or threats to national security. Who is targeted by anti-immigrant rhetoric and who is singled out as an especially problematic group varies over time and by country. Mexican and Muslim immigrants, for instance, are favorite targets of U.S. xenophobes today, while the vilification of Japanese Americans that occurred during World War II has subsided. A large literature in the social sciences has sought to explain which immigrants are the most unwanted and vilified and why.
Two kinds of arguments dominate this literature, which we review in more detail below. Many studies suggest that the skills of individual applicants matter most, as citizens prefer well educated and younger applicants who work in high-income jobs (such as engineers and doctors) because such immigrants may benefit the national economy the most (e.g., Hainmueller and Hiscox Reference Hainmueller and Hiscox2010). Other researchers argue that animosity toward racial others or Muslim immigrants (in Christian-majority countries) drives xenophobia (e.g., Gorodzeisky and Semyonov Reference Gorodzeisky and Semyonov2016; Polavieja et al. Reference Polavieja, Lancee, Ramos, Veit and Yemane2023; Yemane Reference Yemane2020).
In recent years, a handful of studies have introduced a third perspective, focused on the political relations between countries of origin and the host country, rather than the characteristics of immigrants themselves. They show that refugees are more likely to be welcomed by governments with a hostile relationship to the country of origin (e.g., Moorthy and Brathwaite Reference Moorthy and Brathwaite2019)—such as the dissidents and refugees from Eastern Europe who found open doors in the West during the Cold War. This research focuses on government policy toward refugees, however, rather than citizen preferences for immigrants, as we do in this study. But we build on this international relations perspective to argue that when it comes to immigrants (as opposed to refugees fleeing their own governments), the relationship is turned on its head: immigrants from allied states should be preferred over immigrants from hostile states because immigrants are generally perceived not as persecuted opponents to, but as everyday ambassadors of their governments (in line with He and Xie Reference He and Xie2022).
Increasingly, the literature on preferences for immigrants uses experimental methods to identify the characteristics of immigrants that may appear, in the eyes of natives, as potentially problematic and lead to their rejection. One strand of this research relies on field experiments, for instance through labor market correspondence tests (also known as audit studies; cf. Koopmans, Veit, and Yemane Reference Koopmans, Veit and Yemane2019). A second strand uses forced-choice conjoint or vignette experiments. In the conjoint design, respondents typically have to choose between two applicants for permanent resident permits, citizenship, or refugee status (Hainmueller and Hiscox Reference Hainmueller and Hiscox2010; Triguero Roura Reference Triguero Roura2021).
We follow the conjoint approach to test our international relations focused theory. Respondents were forced to decide between two applicants for permanent residence status, distinguished by their country of origin as well as a range of other attributes commonly used in the experimental literature to test if labor market concerns dominate preferences for individual immigrants. To disentangle preferences related to international relations from those related to perceived racial and cultural similarity, which stands at the center of the second approach in the literature, we randomly assign four countries of origin to the immigrant profiles: two countries of origin with a similar racial and cultural make-up as the majority of the survey respondents, one of which is a rival country and the other an ally (or at least a non-rival); and two countries with a racial and cultural make-up that respondents are likely to perceive as different from their own country’s majority race and culture, one of which is again a rival country and the other one an ally (or non-rival).
This experiment was fielded with nationally representative samples in 22 democracies, mostly in Europe and the Americas, but also in Asia as well as South Africa. The results strongly support the geo-political rivalry argument: in each of the survey countries, immigrants from non-rival countries are strongly preferred over those from rival countries. The effect is so large that it results in a net preference for immigrants from countries with a dissimilar racial and cultural makeup than the majority of the host country. We also delve into mechanisms and show that the greater the respondents’ sense of their own country’s superiority (i.e., the more “chauvinist” respondents are), the stronger the international relations of their governments are mirrored in their preferences for immigrants. Similarly, members of ethno-racial majorities are more prone to the rivalry effect because they are more strongly identified with their nation compared to minority members.
Our survey happened to be fielded during the leadup to and initial days of the Russian invasion of Ukraine on February 24, 2022—two countries we had featured as a rival/ally pair in 14 of the 22 survey countries. We are thus able to employ an “unexpected event during survey design” approach (Muñoz, Falcó-Gimeno, and Hernández Reference Muñoz, Falcó-Gimeno and Hernández2020) and examine how the ratcheting up of a geo-political rivalry, which was exogenous to our experiment, affected respondents’ attitudes toward prospective immigrants from the two countries. The analysis shows that the escalation of the rivalry to the level of a hot war strongly increased anti-rival and pro-ally bias. This encourages us to interpret the rivalry effect as causal, at least in this crucial test case, since we do not have to worry about confounders that do not vary quickly over time—such as regime type or the relative appeal of a country’s pop culture. For the other cases, we interpret the results with caution since countries of origin represent bundles of characteristics from which it is difficult to precisely isolate the foreign policy component. We deal with some of the most plausible confounders in supplementary analyses.
How important is this foreign policy component for our understanding of anti-/pro-immigrant sentiments? After all, many immigrants come from countries that are either of negligible foreign policy significance for the host country or are seen as neither allies nor rivals. We can offer a rough estimate of the relative relevance of our findings based on the OECD databaseFootnote 1 on immigrant stocks by country of birth or citizenship. By cross-referencing this information with a recently updated dyadic dataset on geopolitical rivalries and alliances (through 2020, see Diehl, Goertz, and Gallegos Reference Diehl, Goertz and Gallegos2021), we conclude that about one-third of migrants in highly developed OECD countries come from either allied or rival states. The mechanisms we document in this article thus play an important part in the overall dynamic leading to the selective rejection or acceptance of immigrants.
LITERATURE, THEORY, AND HYPOTHESES
Social scientists have long sought to understand anti-immigrant sentiment, as manifested in popular opinion as well as in political movements and party platforms. One strand of this research attempts to explain varying levels of hostility toward immigrants across countries, identifying as possible factors labor market conditions, the influence of right-wing populist parties, levels and composition of immigrant stocks and flows, global integration, the history of nation-state formation, and so on.
A second body of research is more directly relevant to the study presented here. It is not concerned with overall levels of anti-immigrant hostility but with its targets: which immigrant groups are the least wanted, most discriminated against, or perceived as the gravest threat to native interests? Indeed, many studies show that individuals do not evaluate all immigrants equally but make distinctions based on the latter’s places of originFootnote 2 and their individual characteristics. Broadly speaking, the experimental literature, on which we mainly focus here, offers three explanations for such preferences.
Racism, Cultural Distance, and Economic Competition
First, many authors argue that preference for and avoidance of specific immigrant groups are driven by racial resentment. The overwhelming majority of studies look at discrimination against non-white immigrants in white-majority countries.Footnote 3 Most recently, a series of arguments have emerged suggesting that blaming China for the COVID-19 outbreak resulted in a resurgence of “anti-Asian hate” and the rejection of Asian immigrants as “forever foreigners” (Li and Nicholson Reference Li and Nicholson2021; but see Daniels et al. Reference Daniels, Dimaggio, Mora and Shepherd2021).
Second and relatedly, other authors have focused on perceived cultural distance as a factor explaining preferences for or rejection of certain immigrant groups, most importantly Muslims in non-Muslim-majority countries. Most experimental studies that treat granting permanent residence permits or citizenship as the outcome report a strong and consistent anti-Muslim bias.Footnote 4 In a German labor market correspondence test, Koopmans, Veit, and Yemane (Reference Koopmans, Veit and Yemane2019) find that “objective” cultural distance (as measured through survey questions in origin countries) explains call-back rates across origin groups.
Third, most of the literature finds that the skills of individual immigrants determine their reception by natives. All studies using a conjoint experiment format (that we know of) arrive at the conclusion that highly skilled and educated immigrants, as well as those seeking work, are preferred over low-skilled immigrants and those not seeking (or not able to) work. Regarding possible mechanisms, the vast majority of studies find no evidence for labor market competition arguments (except in very specific niches, see Malhotra, Margalit, and Mo Reference Malhotra, Margalit and Mo2013) and support the idea that highly skilled immigrants are preferred because they are seen as benefitting the overall economy.Footnote 5
Theory and Hypotheses: International Rivalry and Political Competition
The literature has not considered a fourth possible factor that may also shape preferences for specific groups of immigrants: the global configuration of political alliances and hostilities between countries. In particular, immigrants from rival countries with whom the host country has a history of contentious and conflictual encounters are likely to be the least welcome while immigrants from allied countries may be the most welcome. Preferences for immigrant groups could thus represent a popular reflection of geopolitical relations of opposition and alliance (in line with the case study by He and Xie Reference He and Xie2022).
This argument rests on a particular tradition in the study of public opinion formation in foreign policy, according to which the public largely follows elite cues (see for example Berinsky Reference Berinsky2007). A recent article shows empirically for the United States that elites’ and citizens’ foreign policy preferences indeed closely mirror each other during most periods (Kertzer Reference Kertzer2022). Based on these results, it is reasonable to assume that individuals know and care about the foreign policy relationships with allied and rival countries. Our supplementary validation survey (discussed below and in Supplementary Appendix D)Footnote 6 confirmed that respondents’ perception of foreign policy alignments consistently mirrors those of their governments.
At the micro-level, the mechanisms translating geopolitical competition and rivalry into anti-immigrant sentiment likely include national identification and/or statistical discrimination. National identification leads respondents to see themselves as well as immigrants from specific countries as embodying their respective national communities (a form of “banal nationalism,” Billig Reference Billig1995). An immigrant from Japan, for example, thus comes to represent the Japanese nation—and, importantly for our international relations argument, the Japanese state. Respondents, in turn, identify with their own country and its government, assuming an international relations perspective when evaluating immigrants hailing from different countries.
Given that most people know very little about the composition of specific immigrant streams, they make rational assumptions about the average foreign policy dispositions of immigrants, in line with the statistical discrimination approach to stereotyping (Phelps Reference Phelps1972). Natives thus assume that the average Japanese migrant holds similar foreign policy stances as those of the Japanese government. This reinforces preference for immigrants from allied countries over immigrants from rival countries, all else being equal, given that the dispositions of immigrants from allied countries resemble those of respondents on average and given the fear of fifth columns that often accompanies nationalist thinking (Mylonas and Radnitz Reference Mylonas and Radnitz2022).
Our argument is agnostic to the relative weight of the national identification and the statistical discrimination mechanisms as they lead to observationally equivalent outcomes. Through either or both of these two mechanisms, dislike of a foreign country’s political and economic stances can spill over into dislike of the citizens of that country.Footnote 7 Anti-Israel positions can tip into antisemitism. Anti-China sentiment can slip into Sinophobia, and so on. Note that our argument does not take the role of national media into account, which is known to amplify local concerns about immigration (Hopkins Reference Hopkins2010). Given that news coverage about rival countries (and their citizens) is both more negative and more extensive (Hufnagel, Von Nordheim, and Müller Reference Hufnagel, Von Nordheim and Müller2023), the media are likely to amplify perceptions of rivalry and thus the rivalry effects posited by our theory.
Our perspective further implies that individuals distinguish between political refugees and immigrants when making evaluative judgments, as suggested by a long line of research.Footnote 8 From an international relations perspective, the distinction is crucial because refugees are often opposed to the policies of their governments (otherwise the latter would not persecute the former), leading to their acceptance if they hail from a rival country, as the warm welcome of dissidents and refugees from Communist countries during the Cold War illustrates. Immigrants who leave their countries for non-political reasons, however, are more likely to hold similar foreign policy views as their government, as discussed above, leading to their rejection if hailing from a rival country and their acceptance if they come from an allied country. Note here that this argument conceives of refugees in a narrow, legal way as citizens who are persecuted by their own government. Ukrainians should thus not be evaluated as refugees in the strict sense of the term, but as migrants, because their move across the border was prompted not by repression at the hand of their own government but by a foreign military invasion.
In this study, we focus exclusively on migrants, rather than refugees. We note here, however, that our intuition is supported by recent research on the reception of refugees as noted above: governments are more willing to accept refugees from rival countries than from allied countries (Chu Reference Chu2020; Jackson and Atkinson Reference Jackson and Atkinson.2019; Moorthy and Brathwaite Reference Moorthy and Brathwaite2019) because refugees fleeing adversarial states can be regarded as allies, especially if a host country supports rebel groups operating on the rival’s territory (Turkoglu Reference Turkoglu2022) or if the rival country adheres to a different political ideology (Jackson and Atkinson Reference Jackson and Atkinson.2019). We are not aware of any research that explores this conjecture with regard to the attitudes of citizens, the focus of our analysis, rather than government policy.
We derive three observable implications from our arguments about how regular citizens evaluate immigrants from different countries of origin. First and most generally, individuals should show a clear preference for immigrants from allied countries and an aversion toward immigrants from rival countries with a history of competition or conflict with the respondent’s country (H1). Immigrants from neutral countries should be neither preferred nor discriminated against. Second, if the national identification mechanism operates as argued above, we expect that members of ethnic majorities will be more sensitive to international rivalries when expressing preferences for immigrants from specific countries (H2). This is because according to both social dominance theory (Pratto, Sidanius, and Levin Reference Pratto, Sidanius and Levin2006) and the in-group projection model (Mummendey, Kessler, and Mielke Reference Mummendey, Kessler and Mielke1999) in social psychology, national majorities tend to identify more strongly with their country than minorities.Footnote 9
Third, individuals who believe that their country is superior to other nations (an attitude commonly referred to as “chauvinism”) may be particularly attuned to geopolitical competition and the threat it can pose to their nation’s status. They are therefore likely to be more strongly opposed to immigrants from rival countries and more warmly disposed toward immigrants from allied countries than respondents with less pronounced chauvinist attitudes (H3). Note that H2 and H3 represent moderation arguments, which we test using interaction terms in the statistical models introduced later.
The logic of the overall theoretical argument is illustrated by some prominent examples from the history of the United States, including Germanophobia during World War I, anti-Japanese propaganda and persecution during World War II, or Islamophobia in the wake of the 9/11 attacks launched from Taliban-controlled Afghanistan. We discuss these and other cases briefly in Supplementary Appendix B.
THE SURVEY EXPERIMENT
We administered a survey experiment to large online samples from 22 democracies. Because the experiment was embedded in a larger collaborative survey project, the countries were not primarily chosen based on their foreign policy relationships. Rather, we focused on Western democracies supplemented with other stable democracies from around the world. This meant that we sometimes had to ask respondents about countries of immigrant origin that were less clearly identifiable as either rivals or allies than we would have wished for, a complication we discuss below in more detail. The final list of survey countries included Australia, Canada, France, Germany, Greece, Hungary, Italy, Netherlands, Poland, Spain, Sweden, the UK, the US, Argentina, Brazil, and Peru in the Western Hemisphere, and Turkey, India, the Philippines, Japan, South Korea, and South Africa outside of it.
Experimental Design
The surveys were fielded between late February and early March 2022 via Lucid Marketplace, a popular online survey platform. Supplementary Appendix C describes how we complied with APSA’s Principles for Human Subjects Research during and after fielding. To help maximize the external validity of our inferences, we used a quota-based sampling procedure, with quotas for age, gender, and education. The surveys were conducted in the national language and script. In Canada, the Philippines, and South Africa, we added a second national language and survey (French, Tagalog, and Zulu, respectively). In total, 46,549 respondents completed the survey and passed a two-part attention check (Aronow et al. Reference Aronow, Kalla, Orr and Ternovski2020). The survey included sociodemographic and attitudinal questions and several randomly ordered conjoint and vignette experiments, including the one examined in this article.
To test our theoretical claims, we conducted an immigrant officer conjoint experiment adapted from Hainmueller and Hopkins (Reference Hainmueller and Hopkins2015), which has been replicated in many different variations. We asked respondents to choose between a pair of applicants for permanent residence status with randomly assigned characteristics. The latter included variables that are relevant for labor market concerns: age, education, language competence, and profession, as well as length of residency. Each respondent was asked to make six successive choices between paired immigrants. Supplementary Appendix G illustrates the experimental setup with a screenshot from one of the surveys.
To disentangle the role of geopolitical rivalries from the perceived racial and cultural proximity to respondents, a core mechanism of an important strand of the literature discussed above, we selected four countries of origin for each survey country that varied across both race/culture and rivalry. The four countries thus included: (1) a country of origin whose majority population is likely to be perceived as racially and culturally similar by the majority of respondents and which has a non-rival relationship with the respondents’ country; (2) a country with a similar perceived racial and cultural makeup but a rival relationship with the respondents’ country; (3) a country with a population perceived as dissimilar in racial and cultural terms and a non-rival relationship; and (4) a country with a dissimilar racial and cultural makeup and a rival relationship. We validated our assumptions about perceived rivalries and alliances as well as perceptions of cultural and racial differences with a separately fielded representative survey in these 22 countries, conducted after the main, experimental survey had already been concluded (Supplementary Appendix D describes this validation survey). The results are discussed below.
Note that we do not measure racial distinctiveness directly, for example, by varying the skin color in images of immigrant applicants (as done by Harell et al. Reference Harell, Soroka, Iyengar and Valentino2012; Helbling and Kriesi Reference Helbling and Kriesi2014; Hopkins Reference Hopkins2015; Valentino et al. Reference Valentino, Soroka, Aalberg, Duch, Iyengar, Fraile and Hahn2019), nor does our experiment include attributes related to the cultural practices and beliefs of individual immigrants (such as their religion). This is because there is little plausible skin tone or religious variation among many of the country-of-origin populations (e.g., among Japanese people), making such a research design implausible from the point of view of respondents.
Instead, we choose pairs of rival and non-rival countries in close geographic proximity to each other, such that, from the standpoint of the respondents, they plausibly resemble each other in terms of average phenotypical features and culture, language, and religion. One of these pairs was situated continents away from respondents and was not tied through past migration and ancestry with majority respondents, while the other pair was in close geographic proximity or linked through ancestral ties to majority respondents, making it very likely that the latter pair would be perceived as culturally and racially more similar. In the validation survey, outlined in Supplementary Appendix D, we checked our initial assumptions against how a nationally representative sample of respondents saw rivalry and alliance relations as well as the degree of racial and cultural similarity or difference between their countries and the hypothetical immigrant-sending countries featured in the experiment. With very few exceptions, which we note below, our initial assumptions were validated.
We acknowledge that this design cannot isolate rivalry and cultural/racial distance in an unequivocal way as there are other characteristics of countries of immigrant origin that respondents may be reacting to as well. After all, countries of origin represent bundles of attributes. We are not aware of a research design that would have allowed us to do so. Shifting to individual-level variation (for example by providing information about the foreign policy stances of individual immigrants) would have created ecological or external validity problems, since outside of the experimental context such individual characteristics are unobservable for average citizens. We discuss some of the main possible confounders (such as regime type or the humanitarian circumstances under which migrants leave their country of origin) further below.
Forced-choice experiments like our conjoint offer several advantages over both standard survey instruments and vignette experiments (cf. Denney and Green Reference Denney and Green2021). Respondents are asked in concrete terms whether they would grant permanent residency to specific individuals, rather than more abstract questions about the desirability of immigrants from particular countries. The latter approach cannot disentangle compositional characteristics of a specific migration stream in terms of profession, age, or language competence from respondents’ preferences for that country of origin. The conjoint experiment allows us to do precisely that, resulting in less measurement errorFootnote 10 and better mapping onto real-world behavior than observational survey questions or vignette experiments (Hainmueller, Hangartner, and Yamamoto Reference Hainmueller, Hangartner and Yamamoto2015). Second, a conjoint experiment is helpful in minimizing social desirability bias since respondents’ preferences for certain countries of origin are inferred from several of their choices (Horiuchi, Markovich, and Yamamoto Reference Horiuchi, Markovich and Yamamoto2022). The prompts are framed as decisions about individuals, rather than countries of origin, allowing respondents to plausibly deny that their choices are influenced by country stereotypes. This is especially important for questions that could be understood as relating to ethnic or racial prejudice (An Reference An2015).
Countries of Origin
In choosing immigrants’ countries of origin, we relied on a broad understanding of rivalry. Following Thompson (Reference Thompson2001), rivalries represent dyads of states that “regard each other as a) competitors, b) the source of actual or latent threats that pose some possibility of becoming militarized, and c) [potential or actual] enemies” (Thompson Reference Thompson2001, 560). This understanding relies on perceptions, rather than government actions, and thus does not depend on the frequency of Militarized Interstate Disputes, in contrast to many other definitions. Alliances are defined by opposite features and are characterized by a shared focus on cooperation and mutual interests, as well as by trust in the peaceful and friendly nature of the relationship.
Before we describe the rivals and allies used in our experiment, it is important to note that in some cases, we had to make sub-optimal choices due to the fact that survey countries were not primarily selected with our experiment in mind and sometimes did not have clear-cut rivalries and alliances with both culturally/racially similar and dissimilar countries of origin. In a few cases, this forced us to choose countries of origin that were in a neutral or ambiguous relationship with the survey country. We discuss the specifics of these compromise choices below.
For all survey countries except Japan, we used China and Japan as rival and non-rival countries of immigrant origin with a majority non-white population and a large perceived cultural distance (in South Korea, they represented the culturally and racially more proximate pair). The rivalry between China and the West has broken into the open recently (cf. Mearsheimer Reference Mearsheimer2021), while Japan remains firmly in the camp of the anti-China coalition that has emerged over the past decade (Maizland and Cheng Reference Maizland and Cheng2021). From the point of view of many residents outside of East Asia, the citizens of China and Japan are racially similar and equally culturally distant, as the validation survey confirmed. Political relationships with China and Japan are less clear-cut for South Africa, where the official economic and financial cooperation with China, in the framework of the increasingly formalized BRIC alliance, was heavily criticized in recent years (Lu Reference Lu, Alden and Wu2021). We therefore classify this relationship as ambiguous, rather than as an instance of clear-cut rivalry. Correspondingly, the validation survey revealed that average South Africans do not see China as more hostile or friendly than Japan.
Across all our survey countries in Western Europe and Northern America, Russia has emerged as a second rival for regional and global dominance, from the autocratic turn under Putin onwards, and most openly since the annexation of Crimea in 2014 (Saad Reference Saad2019). Correspondingly, the Western public now sees Russia as an enemy as much as it did during the height of the Cold War. The obvious allied country with racially and culturally similar population characteristics is Ukraine, which has been driven into an even closer alliance with the West since the 2014 Maidan Revolution. Our post-experiment validation survey confirmed these assumptions with a single exception: in Hungary, Russia is seen by average Hungarians as more friendly than Ukraine, a reflection of Victor Orbán’s foreign policy stances.Footnote 11
For survey countries outside of Western Europe and North America (i.e., Argentina, Peru, Greece, Turkey, India, South Africa, the Philippines, South Korea, and Japan), we looked at the rivalry literature in international relations (Diehl and Goertz Reference Diehl and Goertz2001; Dreyer and Thompson Reference Dreyer and Thompson2012; Klein, Goertz, and Diehl Reference Klein, Goertz and Diehl2006; Thompson Reference Thompson2001) and analyzed existing rivalry datasets to identify suitable immigrant origin countries distinct from the Ukraine-Russia pair. We did not know in advance how active, persistent, strong, and publicly known the rivalries actually were in these countries in 2022, since no updated dataset was available when our surveys were fielded. Taking older rivalries into account was the most sensible solution, since a rivalry could leave a long-term legacy. We therefore include historical rivalries in some cases, even if those rivalries are listed as terminated in available datasets. The post-experiment validation survey, in which we asked respondents about their perception of how friendly or hostile they saw the countries of immigrant origin, allowed us to clarify which rivalries are still perceived as such in the average citizen’s mind. This was the case for all but one (discussed below).
We again chose one pair of rival and non-rival countries as distant and the other as proximate in racial and cultural (including religious) terms to the majority of respondents in order not to confound rivalry with perceived racial and/or cultural distance. We deviated from this principle in the Philippines, Greece, India, and Turkey, where we chose pairs of countries that differed significantly in terms of religion and that, in the case of the Philippines and India, were also racially dissimilar from the majority population. This deviation allowed us to choose a meaningful second rival (Libya for the Philippines, Turkey for Greece, Pakistan for India, and Greece for Turkey). The post-experiment validation survey revealed that in the case of Turkey, India, and the Philippines, racial or religious difference sometimes trumped, in the eyes of the survey respondents, other considerations of similarity, making these choices of countries of immigrant origin less than ideal for our purpose (see Supplementary Table D1 for details).
Conversely, we also had to make some compromise choices regarding allies and rivals in order to maintain strict selection rules regarding racial/cultural similarity and difference. For South Korea, we chose Australia as a second culturally and racially distant country of origin besides the allied United States, knowing that the relationship between South Korea and Australia is best described as one of mutual neglect, rather than rivalry (Robertson Reference Robertson2021; Robertson and Gerszberg Reference Robertson and Gerszberg2021). Still, the validation survey (for details see Supplementary Appendix D) showed that South Koreans see the United States government as more friendly than that of Australia—but none of them as hostile. Relationships with Japan are ambiguous as well, torn between legacies of the colonial past and the shared security interests vis-à-vis China (Park Reference Park2008), which have become increasingly important over time.Footnote 12 Correspondingly, the validation survey showed that China is seen by South Koreans as more hostile than Japan.
Table 1 classifies the countries of origin based on their rival and non-rival status as well as their perceived racial/cultural similarity and dissimilarity. We mark with a single asterisk those cases (3/88) for which our validation survey did not produce clear differences in the perception of friendliness/hostility between assumed rival and assumed non-rival countries. Countries of immigrant origin with two asterisks are those for which our assumptions about perceived racial/cultural distance were not validated by the survey (also 3/88). For 82/88 pairs (or 93%), our assumptions were therefore validated. In Supplementary Figures A6a–c, we successfully replicate the main results without responses from survey countries where either of the two assumptions was not confirmed.
Table 1. Survey Countries and Immigrant-Origin Countries
Note: *The validation survey did not indicate that the assumed rival was perceived as more hostile, on average, by respondents; **the validation survey did not indicate that the population was perceived as more similar in cultural and racial terms on average, by respondents; in the case of Libyans in the Philippines, and Serbians in Turkey, only the perception of racial, but not cultural proximity diverged from our assumptions (see Supplementary Table D1).
Analytic Strategy
We present the results of the forced choice experiment using marginal means. We prefer them to average marginal component effects (AMCEs) because they are more appropriate for comparing preferences across subgroups of respondents (such as different survey countries) (Leeper, Hobolt, and Tilley Reference Leeper, Hobolt and Tilley2020), which we do below. Since marginal means are not dependent on the choice of a reference category, we are also able to compare their magnitude across immigrants’ countries of origin and other attributes. As a reminder, since the experiment was fully randomized and the assignment of conjoint attributes balanced across respondents of each survey country, we do not present results with respondent-level covariates.
By coincidence, the online survey experiment was launched a couple of days before Russia invaded Ukraine on February 24, 2022, and remained in the field for a few days afterward, allowing us to conduct an over-time analysis of responses for those survey countries where Ukraine and Russia were chosen as countries of immigrant origin. Building on the assumption that the timing of Putin’s declaration was as-if-random within the temporal window of our survey, we leverage this event to help us identify the causal effect of rivalry and alliance on preferences for immigrants. This effect would normally be difficult to estimate, since many countries of immigrant origin differ not only in the degree of rivalry with the survey country, but in many other, unobserved ways as well. We discuss sample balance issues as well as the possible role of humanitarian concerns for Ukrainians, which may confound the change in the intensity of the alliance, in the corresponding section below. With respect to other rival/ally pairs outside of the analysis of Russia and Ukraine, we will also attend to regime type as another possible confounder.
RESULTS
Geopolitical Rivalry and Perceived Racial/Cultural Difference
We start with the main results based on a pooled sample of all survey countries. Results by survey country will be discussed in later sections. We first present estimates for rivalry and perceived racial/cultural proximity as two separate attributes. The x-axis in Figure 1 represents marginal means, that is, the probability that an immigrant with certain attributes will be chosen by respondents, with higher values indicating a higher chance of acceptance. The vertical dotted line represents the absence of an experimental effect: that is, on average, an equal probability of acceptance or rejection. Confidence intervals overlapping with this line indicate that the corresponding attribute does not influence respondents’ choice at the conventional, two-tailed significance level of p < 0.05.Footnote 13 Note that the confidence intervals are too small to be visible due to the large sample size.
Figure 1. Effects of Rivalry and Perceived Racial/Cultural Similarity on the Probability of Being Granted Permanent Resident Status
Note: Plotted points are marginal means. Bars denote 95% confidence intervals (two-tailed). Standard errors are clustered by the respondent. For tabular results, see the APSR Dataverse repository, Full Model Results Tables, Table I.
Figure 1 provides the full set of immigrant attributes that we experimentally varied, thus offering comparisons for effect sizes. The four attributes listed at the bottom of Figure 1 provide strong support for the first hypothesis: the probability that an immigrant from a rival country is granted permanent resident status is 7.6 percentage points lower compared to an immigrant from a non-rival country. As we will see in a moment, the antipathy toward certain rivals and the sympathy for certain non-rivals is so strong that there is a net preference for country of origins with a larger perceived racial/cultural distance—the opposite of what important strands in the literature would expect. According to Figure 1, immigrants who are perceived as culturally and racially similar are 1.2 percentage points less likely to receive permanent resident status compared to immigrants who are perceived as racially and culturally more distant. How do these effect sizes compare to those of other often-studied attributes? The effects of rivalry are smaller than those of unemployment status or language proficiency, as Table 1 shows, but larger than the length of residency and gender. We will come back to the question of effect sizes further below.
To what extent are these results an artifact of our choice of rival and non-rival countries, in particular of the frequently used China–Japan as well as Russia–Ukraine pairing? Regarding the latter, the results are similar for the subsample of respondents (from 8 of the 22 countries surveyed) who did not choose between Ukrainians and Russians (see Supplementary Figure A1). In other words, the results presented in Figure 1 are not dependent on the strong anti-Russian sentiment generated by the war in Ukraine. How about the specific nature of the China-Japan comparison? We cannot directly investigate the possibility that the heterophilia effect is influenced by the pro-Japanese sympathy that we find in all countries except in South Africa, because Japan appears as a country of origin in all but the Japanese surveys. But we note here that respondents in Japan and South Korea also preferred dissimilar (specifically: American and Australian) immigrants over those from East Asia, thus contributing to the overall pattern.
To further disentangle rivalry from perceived racial and cultural similarity and to examine what is driving the preference for dissimilar immigrants, we interact racial and cultural similarity with rivalry, thus distinguishing between rivals of dissimilar racial and cultural backgrounds, non-rivals with dissimilar backgrounds, rivals with similar backgrounds, and non-rivals with similar backgrounds. The results are visualized in Figure 2.
Figure 2. Interaction Effects of Rivalry Status and Perceived Racial/Cultural Similarity on the Probability of Being Granted Permanent Residence Status
Note: Plotted points are marginal means. Bars denote 95% confidence intervals. Standard errors are clustered by the respondent. Other conjoint attributes are omitted from the plot. For tabular results, see the APSR Dataverse repository, Full Model Results Tables, Table II.
Clearly, the association between rivalry and receiving permanent residence status is pronounced for immigrants from both similar and dissimilar racial and cultural backgrounds, but it is stronger for the racially and culturally similar rivals. This indicates a stronger aversion against Russian immigrants compared to Chinese immigrants in the 14 Western countries that were offered this choice, as well as a strong anti-Chinese sentiment in South Korea and Japan. It is plausible that both are due to the high intensity of the rivalry in question: a full-scale war in Ukraine in the case of the West, and the spatial proximity to a rising and increasingly assertive China in the case of Japan and Korea. Indeed, the validation survey revealed that Russia is perceived in Europe and North America (with the exception of Hungary) as far more hostile, on average, than China, while the opposite is the case in Japan and Korea. We explore and support this interpretation in more detail in Supplementary Appendix E.
To further disentangle cultural/racial proximity from rivalry, we can also point to auxiliary analyses of the validation survey (described in Supplementary Appendix D). We asked respondents about how different/similar they perceived the culture and the race of the population of the countries of immigrant origin—offering a continuous rather than a dichotomous coding of this variable. To measure rivalry at the level of citizen perceptions, we asked how friendly or hostile (on a 5-point scale) respondents perceived the government of the countries of immigrant origin. Mirroring the outcome variable in the experiment, we asked how likely respondents would admit applicants for a permanent work visa from these countries. As Supplementary Figure A2 shows, perceived degrees of rivalry trump perceptions of racial or cultural proximity by far.
Moving on to a discussion of possible confounders, how confident should we be that these effects should be attributed to rivalry, rather than to regime differences between survey and origin countries? More specifically, do respondents dislike Chinese and Russian immigrants because their countries of origin are foreign policy rivals or because they are autocracies? Indeed, the average 2022 electoral democracy score from the V-Dem dataset (which ranges from 0 to 1; see Coppedge et al. Reference Coppedge, Gerring, Knutsen, Lindberg, Teorell, Altman and Michael2020) is 0.73 for survey countries (being a democracy was a sample selection criterion) and 0.63 for allied countries of immigrant origin, whereas it is only 0.2 for rival countries of immigrant origin (China at 0.075 brings down that mean). These differences reflect the fact that all-democratic dyads are less likely to develop rivalries in the first place compared to mixed dyads or autocracy-autocracy dyads (Conrad and Souva Reference Conrad and Souva2011; Hensel, Goertz, and Diehl Reference Hensel, Goertz and Diehl2000).
Still, we can disentangle rivalry from regime type differences, taking advantage of the full range of variation in our data. To that end, we regressed the marginal mean values for each country of immigrant origin generated by the survey experiments on a rival-ally dummy and on differences in the democracy scores between each survey country and country of immigrant origin (again relying on the V-Dem dataset). The results of this dyadic analysis with 88 observations indicate that regime difference is not significantly associated with the marginal means of immigrant preference (beta = −0.068; p = 0.416), whereas a rival geopolitical relationship clearly is (beta = −0.752; p = 0.000).
Over-Time Analysis of the Preference for Ukrainian Immigrants and Rejection of Russian Immigrants
A more direct way to isolate the international relations dimension from other attributes of the countries of immigrant origin is to identify cases where levels of rivalry/alliance change dramatically over time, while other attributes, including regime type, remain constant. As mentioned earlier, the survey experiment was fielded both prior to and after Russia’s invasion of Ukraine on February 24, 2022. The Russian aggression represents an unexpected exogenous shock within the duration of our survey.
We can therefore use the “unexpected event during survey design” (UESD) (Muñoz, Falcó-Gimeno, and Hernández Reference Muñoz, Falcó-Gimeno and Hernández2020), whose validity rests on several assumptions: excludability (that only the event in question affects the outcome variable); ignorability (that the pre- and post-event samples do not vary systematically); compliance (that respondents in the post-event sample are actually aware of the event); and lack of post-treatment bias (that the event does not affect survey responses to covariates and moderators other than the outcome of interest). We only view ignorability bias as a serious challenge for our design, to be discussed below together with humanitarian concerns as a possible alternative causal channel.
For the main analysis, we chose survey countries that contained the Ukraine-Russia pair and that had daily responses both before the war and for every post-war day until day 5. This generates a sample with about 23,000 respondents from Europe and the United States. Since daily sample sizes are small for many countries, we aggregate responses into two-day periods to arrive at sufficiently precise estimates for each period.
Figure 3 shows an increasing pro-Ukraine preference and a growing anti-Russian antipathy over time. The pro-Ukraine effect is especially pronounced, increasing from a 0.53 probability of being offered permanent residency status to nearly 0.6 over the course of a week. The alliance effect thus mirrors in magnitude the effect of the rivalry status, as was already the case in Figure 2. Note that initially, before the outbreak of the war, respondents did not express a preference for Ukrainian over Japanese or for Chinese over Russian immigrants, but in the aftermath of the invasion, these preferences increasingly diverged. Note also that according to Figure 3, the anti-Chinese preferences are not significantly affected by the war, thus offering a placebo test that further supports the rivalry hypothesis.
Figure 3. Preference for Russian and Ukrainian Immigrants Before and After the Russian Invasion
Note: Plotted points are marginal means. Bars denote 95% confidence intervals. Standard errors are clustered by the respondent. Other conjoint attributes are omitted from the plot. For tabular results, see the APSR Dataverse repository, Full Model Results Tables, Table III.
Coming back to the question of effect sizes, the rejection of Russians and the preference for Ukrainians after the war broke out (and the rejection of Chinese in both Japan and South Korea, see Supplementary Figure A3) exceed the magnitude of all individual attributes—with the exception of being a doctor or of being unemployed or lacking any language skills. This reflects the widespread sympathy toward Ukrainians and the generous welcome they received upon migration to Western and Eastern Europe after the outbreak of the war, irrespective of other desirable attributes of migrants such as profession or language skills. In other words, the rivalry effect varies with the intensity of the rivalry/alliance (see also Supplementary Appendix E) and is thus particularly pronounced in extreme cases such as during the Ukraine war or the threat posed by China to its neighbors in East Asia—and presumably during the Cold War, the Vietnam War, and other such intense historical conflicts.
This over-time experimental evidence demonstrates that the country-of-origin effect for the same two countries changes as the intensity of rivalry/alliance increases, while other country-of-origin characteristics, including regime type, the appeal of Ukrainian or Russian popular culture, or the frequency and nature of respondents’ previous encounters with Ukrainians or Russians all remain constant. We note here that Bansak, Hainmueller, and Hangartner (Reference Bansak, Hainmueller and Hangartner2023) also find that Ukrainian refugees are far more welcome in 2022 than they were in their previous experimental study with an identical design conducted in 2016.
We do, however, have to be concerned about ignorability, one of the conditions for a valuable UESD mentioned above. We observe some imbalance in the sample composition before and after the invasion of Ukraine, particularly on demographic variables associated with immigration preferences. To address these concerns, we used optimal matching to balance the samples on observables and then reran the conjoint analysis on the matched sample. The results, illustrated in Supplementary Figure A4, are consistent with the main findings reported above.
Another objection relates to a possible alternative causal channel linking the war to immigrant preferences. Could it be that Ukrainians were increasingly preferred by our respondents because they felt sympathy for people who had to flee from an unprovoked and violent military assault, independent of the intensifying alliance with Ukraine? First and as shown in Figure 3, the penalty for Russian immigrants increases steadily over the days before and after the invasion, which clearly cannot be attributed to a decrease in humanitarian concerns for Russians. Second, we can rely on other data to see if humanitarian concerns systematically influence the reception of refugees, for which such concerns are obviously of greater relevance compared to immigrants. Indeed, refugees who were personally tortured or otherwise persecuted were preferred over other immigrants in prior experimental research (e.g., Bansak, Hainmueller, and Hangartner Reference Bansak, Hainmueller and Hangartner2016; Reference Bansak, Hainmueller and Hangartner2023). It is unclear, however, what to expect at the aggregate country-of-origin level, where other considerations (including foreign policy relations with the respondent’s country) come into play as well (cf. Moise, Dennison, and Kriesi Reference Moise, Dennison and Kriesi2023).
To explore the possible role of humanitarian concerns, we relied on the results of two conjoint experiments, one from 2015 (Bansak, Hainmueller, and Hangartner Reference Bansak, Hainmueller and Hangartner2016) with European respondents and one from 2019 with U.S. respondents (Steele, Abdelaaty, and Than Reference Steele, Abdelaaty and Than2023). They both calculate preferences for refugees from different countries of origin, net of other applicant characteristics. We use the relative rank of each country of origin as a dependent variable and regress it on two indicators of the degree of civilian victimization in that country: the average annual death count of civilians in the years before the survey and the level of state terror against citizens during the year of the survey. The results, shown in Supplementary Figure A5, suggest that refugees from countries of origin with a greater degree of civilian victimization are not systematically preferred by Western respondents. We therefore do not think that escalating humanitarian concerns after Russia’s invasion of Ukraine sufficiently confound the observed rivalry/alliance effects to undermine our interpretation of the results.
Results by Individual Survey Country
Next, we analyze the results by survey country. For ease of interpretation, Table 2 provides a summary of the findings. The corresponding plots can be found in Supplementary Figure A3. The foreign policy effects are clear-cut: in all 22 survey countries, immigrants from non-rival or allied countries are viewed significantly more positively than immigrants from rival countries. If we look at each pair of countries of origin separately, however, there are four pairs (out of a total of 44) that do not produce a clear preference for non-rivals.
Table 2. Summary of Results by Survey Country
How do we interpret these exceptions—setting aside the possibility of Type II error? In the case of Australia and the United States as countries of immigrant origin in the Korean survey, the results are not surprising, as we chose Australia on the basis of its perceived cultural and racial dissimilarity knowing that it was not a rival, but maintained a relationship of mutual neglect with South Korea, as discussed above.Footnote 14 We are also not surprised by the case of Paraguay versus Ecuador in Peru, because the intense territorial rivalry between Peru and Ecuador concluded more than 20 years ago with the 1998 peace agreement (Schenoni et al. Reference Schenoni, Goertz, Owsiak and Diehl2020), and correspondingly, the validation survey came back showing that average Peruvians do not distinguish between Ecuador and Paraguay (the non-rival) in terms of perceived friendliness/hostility. In the case of South AfricaFootnote 15 and Turkey,Footnote 16 upon further reflection and with the benefit of hindsight, our choice of one of the two rival or non-rival countries, respectively, turned out to be less than optimal.
To verify whether our results are robust to the exclusion of all survey countries for which our assumptions about foreign policy relationships or the perceived cultural and racial distance were not validated by the supplementary survey, we re-ran the main analyses presented above on a reduced sample. As is shown in Supplementary Figure A6a–c, the results are largely unchanged and all our hypotheses are supported.Footnote 17
Moderation by Degree of National Identification
We are now ready to explore the second and third hypotheses, which regard treatment heterogeneity: the rivalry effect should be moderated by respondents’ majority or minority status as well as by their belief in the superiority of their country. Note that these moderation analyses should not be interpreted in causal terms as we do not know what other (observed or unobserved) respondent characteristics correlated with chauvinism or minority status could drive the observed effects.
Figure 4 explores the hypothesis that majorities react more to the foreign policy relationships with origin countries than do minorities. We included in each survey country-specific questions, often drawn from national censuses, that measured respondents’ membership in linguistic, religious, racial, or other ancestry-based minority groups. Figure 4 shows that the rivalry effect is considerably stronger for members of national majorities. But the rivalry effect is still pronounced for minorities as well. Note also that minorities and majorities do not react to racial or cultural proximity of potential immigrants in different ways, as one would expect if these considerations would drive overall responses.
Figure 4. Effects of Rivalry and Racial/Cultural Similarity by Respondents’ Majority Status
Note: Plotted points are marginal means. Bars denote 95% confidence intervals. Standard errors are clustered by the respondent. Other conjoint attributes are omitted from the plot. For tabular results, see the APSR Dataverse repository, Full Model Results Tables, Table IV.
The third hypothesis maintains that the rivalry effect is stronger for respondents who view their own country as superior to all other countries. Figure 5 shows exactly this. To determine whether this is the consequence of the Ukrainian war taking place in Europe, where most of the respondents lived who had to choose between Ukrainian and Russian immigrants, we carried out a further robustness analysis limited to the subsample of respondents who were presented with different rival-ally pairs. The results are again broadly similar, as Supplementary Figures A7 and A8 show.
Figure 5. Effects of Rivalry and Racial/Cultural Similarity by Respondents’ Perceptions of National Superiority
Note: Plotted points are marginal means. Bars denote 95% confidence intervals. Standard errors are clustered by the respondent. Other conjoint attributes are omitted from the plot. For tabular results, see the APSR Dataverse repository, Full Model Results Tables, Table V.
CONCLUSION
This article introduced a novel international relations argument into the literature on anti-immigrant sentiment. In addition to the racial, cultural, or labor-market characteristics of immigrants, the exclusive focus of the existing literature, we argue that the political relationship between origin and destination countries influences how migrants are perceived and evaluated. Drawing on the rivalry literature in international relations, we showed that immigrants from countries with a contentious and conflictual relationship with respondents’ countries are less welcome than those hailing from allied countries. The preference for immigrants from allied countries and the discrimination against those from rival countries trumps considerations of racial and cultural similarity, at least in this sample and with our choice of countries of immigrant origin.
In the specific empirical set-up of our study, aversion against Russian immigrants in Western Europe and against Chinese immigrants in East Asia produces a net preference for immigrants of dissimilar racial and cultural backgrounds. A more detailed analysis of the data by survey country (see Supplementary Appendix E) showed that there is little evidence of a broad anti-Asian or a more specific Sinophobic bias beyond the rivalry effects. This conclusion holds across survey countries, continents, and origin countries.
Some limitations of our approach have been noted. Since all surveys contained a choice between Chinese and Japanese immigrants, we do not know the extent to which the specificity of an apparently widespread Japanophilia drives the results regarding the net preference for racially and culturally dissimilar immigrants. We also were not able to identify clear-cut rival and non-rival pairs from racially and culturally dissimilar/similar origins for all the 22 survey countries. Moving beyond these limitations would demand a larger sample of survey countries, a more precise and up-to-date measurement of rivalry, and a larger choice set of immigrant origins for respondents to choose from.
Another note of caution concerns the generalizability of our findings. First, we do not know if rivalry would matter more than racial or cultural preferences if we had included other non-white immigrants besides East Asians in our design. Similarly, our findings have little to say about the widespread evidence of anti-Muslim bias in Western countries—itself a consequence, our theory would suggest, of decades of violent conflict and wars between Western powers on the one hand and states (such as Afghanistan, Iran, or ISIS) and political movements (such as the Taliban or al Qaida) associated with radical Islam on the other hand. Future work would do well to investigate these possibilities by including survey countries that are involved in rivalries and alliances with Muslim-majority countries as well as non-white and non-Asian-majority countries.
Our study has larger implications for the study of anti-immigrant sentiment. It shows the importance of going beyond the preoccupation with the individual background characteristics of migrants and embedding the study of xenophobia within the global context of political competition and alliances between countries. This obviously does not preclude the possibility that racial animus or Islamophobia may develop independently or in tandem with these global political forces. Delving deeper into these questions would demand a different research design where racial and cultural features (such as religion) can plausibly vary among immigrants from the same country of origin.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit http://doi.org/10.1017/S0003055424000753.
DATA AVAILABILITY STATEMENT
Research documentation and data that support the findings of this study are openly available at the American Political Science Review Dataverse: https://doi.org/10.7910/DVN/ZZYSIZ. The experimental design, hypotheses, and primary analytical strategies were pre-registered at https://osf.io/kh2ft (see Supplementary Appendix F).
ACKNOWLEDGMENTS
We thank Duy Trinh from Princeton University for data cleaning and preparation. We gratefully acknowledge funding support from the Tokyo Foundation for the main experimental survey and from the Canadian Institute for Advanced Research for the supplementary survey.
AUTHOR CONTRIBUTIONS
A.W. and Z.F. conceived of the project and A.W. led its design and execution, including writing the article. Z.F. did the initial analyses. B.B. contributed to the project’s conceptualization and writing and produced the final figures. C.C. led the design and implementation of the multi-project survey, collected all data, and oversaw research design and analysis. K.T. wrote the grant, served as PI for the multi-project survey, and secured funding from the Tokyo Foundation. Together with C.C., he coordinated all projects and contributed to the design and implementation of the survey. B.B., M.G., and A.W. contributed to the design of the survey. C.C., M.G., and Z.F. contributed to the writing of the article. The supplementary validation survey was designed and executed by A.W., B.B., and C.C. A.W. secured funding from the Canadian Institute for Advanced Research.
FUNDING STATEMENT
The main experimental survey was funded by the Tokyo Foundation for Policy under the research grant “Populism and the Future of Democracy” and the supplementary survey by the Canadian Institute for Advanced Research under grant number CF-0371-CP24-043.
CONFLICT OF INTERESTS
The authors declare no ethical issues or conflicts of interest in this research.
ETHICAL STANDARDS
The authors declare the human subjects research in this article was reviewed and approved by Dartmouth College under protocol number STUDY00032391 (for the main experimental survey) and by Columbia University under protocol number AAAU8663 (for the supplementary survey). The authors affirm that this article adheres to the principles concerning research with human participants laid out in APSA’s Principles and Guidance on Human Subject Research (2020), as detailed in Supplementary Appendix C.
Open access
Comments
No Comments have been published for this article.