Mounting evidence reveals that social context, electoral institutions, historical legacies, and partisan actors can affect the salience of ethnic cleavages, thus illuminating when and how people judge politics based on ethnicity (Birnir Reference Birnir2007; Chandra Reference Chandra2012; Dunning and Harrison Reference Dunning and Harrison2010; Eifert et al. Reference Eifert, Miguel and Posner2010; Laitin Reference Laitin1998; Michelitch Reference Michelitch2015; Posner Reference Posner2005; Rovny Reference Rovny2014). Prior work also establishes that ethnic identity can structure mass political preferences under varied circumstances (Adida et al. Reference Adida, Gottlieb, Kramon and McClendon2017; Lieberman and McClendon Reference Lieberman and McClendon2013; McCauley Reference McCauley2014). Relatively less understood, however, are the psychological mechanisms that heighten ethnic cleavages.
We propose a linguistic hypothesis of ethnic cleavage activation. Drawing on cognitive science (Boroditsky Reference Boroditsky and Nadel2003), we argue that language increases the political salience of ethnic divides in people's minds. Specifically, we claim that speaking a minority tongue strongly primes ethnic divisions, making it more likely that the speaker takes ethnic divisions into account when interpreting politics. In contrast, speaking a majority tongue weakly primes ethnic divisions, allowing other considerations (economic, ideological, etc.) to rise in prominence and affect political judgment.
Cognitive scientists have established that language can influence human thought by increasing the accessibility of specific content in people's minds (Danziger and Ward Reference Danziger and Ward2010). For example, when asked to name tourist spots, Mandarin–English bilinguals are more likely to identify Chinese sites when asked in Mandarin (e.g., Great Wall of China), but U.S. sites when asked in English (e.g., Grand Canyon) (Marian and Kaushanskaya Reference Marian and Kaushanskaya2007; Marian and Neisser Reference Marian and Neisser2000). Indeed, research shows that language can impact how people construe time (Boroditsky Reference Boroditsky2001; Boroditsky et al. Reference Boroditsky2011), whether they assign gender to inanimate objects (Phillips and Boroditsky Reference Phillips and Boroditsky2003), and how they decide if an event is accidental or intentional (Fausey and Boroditsky Reference Fausey and Boroditsky2011). These effects are generally driven by linguistic variation, rather than cultural differences between people (Boroditsky et al. Reference Boroditsky2011; Fausey et al. Reference Fausey, Long, Inamori and Boroditsky2010).Footnote 1
Political scientists also find that interview language robustly correlates with survey response (Lee and Pérez Reference Lee and Pérez2014). Laitin's (Reference Laitin1977) pioneering experiment with Somali–English bilinguals (N = 64) long ago suggested such patterns might arise from linguistic relativity: the notion that varied languages construe (political) concepts differently. But when people fail to share the same sense of a concept, it complicates the interpretation of language differences that emerge between speakers: are these effects due to variation between languages or variation between people's understanding of concepts (for an overview, see Pérez Reference Pérez, Michael Alvarez and Atkeson2015)? Newer work has rehabilitated linguistic relativity by suggesting that varied language speakers share the same concepts. What differs between concepts is their mental accessibility, which language can influence (Pérez Reference Pérez2016). Nevertheless, whether this accessibility directly impacts political opinions is less firmly established.
Using the case of ethnic cleavages, we argue that this language-accessibility mechanism matters beyond simple recall, impacting directly people's opinions about politics. Our contention draws on belief-sampling frameworks, where people construct opinions based on considerations that become salient when a topic is broached (Lodge and Taber Reference Lodge and Taber2013; Tourangeau et al. Reference Tourangeau, Rips and Rasinski2000; Zaller Reference Zaller1992). Specifically, when a person speaks a minority language, we expect ethnic divisions to become more salient to them, thereby impacting the character of their political views. That is, speaking a language that is distinct from the one spoken by the majority of a population should remind individuals more vividly about the multi-ethnic character of their polity. Because information about ethnic divisions is more mentally accessible to them, speakers of a minority language are more likely to consider it when expressing political views. In contrast, speaking a majority language should make individuals less sensitive to ethnic divisions. This is because the language they use does not make this consideration as salient, which means it is less mentally accessible and therefore less likely to affect expressed opinions. Hence, speaking a majority language should lead people to downplay ethnic divisions when making political judgments, allowing other political cleavages (economic, moral, ideological, etc.) to become more prominent and shape political views.Footnote 2
Insofar as our reasoning is correct, these language effects should manifest in how prominently ethnic considerations feature in people's attitudes about politics. We test two observable implications of our reasoning here: speaking a minority language should cause individuals to rate ethnic relations as a more important political issue (H1) and improve people's ability to recall anti-minority political actors (H2).
BACKGROUND
We test our claims in Estonia, a historically homogeneous nation that was once occupied by the Soviet Union and experienced substantial immigration of Russians and Russian speakers from Soviet republics (Laitin Reference Laitin1998).Footnote 3 Lauristin et al. (Reference Lauristin, Kaal, Kirss, Kriger, Masso, Nurmela, Seppel, Tammaru, Uus, Vihalemm, Vihalemm and Jufereva2011) estimate that 61% of Estonians identify as Estonian speakers and 29% as Russian speakers. This linguistic diversity is a byproduct of Soviet occupation, with Russian widely considered a colonial tongue. Under Soviet rule, most Estonians acquired some knowledge of Russian, and among the Russian-speaking population, proficiency in Estonian is growing. Currently, about 44% of Estonians and 36% of Russians are proficient in the other tongue, setting the national share of bilinguals at around 37% (Lauristin et al. Reference Lauristin, Kaal, Kirss, Kriger, Masso, Nurmela, Seppel, Tammaru, Uus, Vihalemm, Vihalemm and Jufereva2011).
RESEARCH DESIGN
We fielded two phone surveys in Estonia (Study 1, N = 262; Study 2, N = 1,200), which randomly assigned bilingual adults to interview in a minority (Russian) or majority (Estonian) language.Footnote 4 This makes our manipulation stringent—i.e., can simply interviewing in a different language make ethnic cleavages more salient?Footnote 5 Recall that we do not claim that language generates an “us versus them” mentality (i.e., ethnocentrism) (Kinder and Kam Reference Kinder and Kam2009). Rather, we contend that language simply makes ethnic cleavages more mentally accessible, which leads people to directly integrate this consideration into their views about ethnic politics. After assessing language fluency and other covariates, respondents were informed:Footnote 6
Based on your answers to some of the previous questions, it appears that you are fluent in both Estonian and Russian. Therefore, we will let the computer program randomly select which language we continue this interview in [SHORT PAUSE].
[Estonian/Russian] was selected. This means that after this point, the rest of the interview will take place in [Estonian/Russian]. This is not a language test. We are simply interested in your opinions as an [Estonian/Russian] speaker.
We used this wording across studies, except Study 1 omitted “We are simply interested in your opinions as an [Estonian/Russian] speaker.” This casts some doubt on demand effects. Asking respondents to report opinions as Estonian (Russian) speakers might lead them to think there is a “proper” way to respond, encouraging them to adjust their reports beyond any language effect. Finding similar effects across slightly varied manipulations will suggest that demand effects are unlikely driving our results (Campbell and Stanley Reference Campbell and Stanley1963).Footnote 7 Although our studies are not replicas, their overall structure is similar (e.g., few pre-treatment covariates followed by random assignment). Our treatment's delivery mode and content is also comparable, save for the minor wording change. Moreover, our outcomes vary by whether they were administered well after the treatment (Study 1) or immediately after it (Study 2). And, in both studies, no respondents dropped upon learning their assigned interview language. Thus, our samples are not comprised of bilinguals who are only comfortable interviewing in one of their tongues, which could bias upward a language effect.Footnote 8
We also note that random assignment of interview language helps to ensure that any observed effects are driven by interview language rather than ethnicity.Footnote 9 In addition, we use pre-treatment data on preferred interview language to proxy for ethnicity. Insofar as assigned interview language propels our treatment effects, they should remain intact after adjusting them for preferred interview language, thus affirming our studies’ internal validity (Campbell and Stanley Reference Campbell and Stanley1963). SI.2 reports information on study administration, identification of bilinguals, sample composition, script for language assignment, and randomization and balance checks.
MEASUREMENT
Our studies were part of longer surveys. Study 1 included an item that most directly appraises language effects on the political salience of ethnic divisions: the importance respondents assign to ethnic relations relative to other issues (Importance of integration). Study 2 fielded an item (Nationalist) evaluating respondents’ knowledge of Estonia's nationalist political party, which is publicly anti-minority.Footnote 10
Our argument concerns the prominence of ethnic divisions relative to other considerations in people's minds. We first test this claim with Importance of integration. Interviewers first read respondents a list of current issues in randomized order: “performance of the economy,” “immigration and refugee crisis,” “integration of the Russian-speaking population,” and “unemployment,” with the third one tapping Estonia's ethnic relations. The phrase “integration of the Russian-speaking population” is how ethnic relations are generally discussed; strategies like secessionism or segregation are not prevalent in Estonian political discourse (Kallas et al., Reference Kallas, Vetik, Kruusvall, Saar, Helemäe, Kirss, Leppik, Seppel, Kivistik and Ubakivi-Hadachi2015; Püvi, Reference Püvi1999).Footnote 11 Respondents then ranked these four issues as the most, second most, and third most important facing Estonia. If ethnic divisions become more salient for respondents interviewing in Russian (minority language), they should assign more importance to the integration of Russian speakers. We code as “4” those who picked this issue as the most important; “3,” those who chose it as the second most important; “2,” those who deemed it the third most important; and “1,” those who ranked it as the least important.Footnote 12
Our Nationalist item asked respondents to report their knowledge of the most nationalist party in Estonia's parliament. The response options were (1) reform party, (2) center party, (3) union of Pro Patria and Res Publica, and (4) “don't know,” with (3) being the right answer.Footnote 13 Unlike our ranking item, which is more open to extraneous variation during survey response (Tourangeau et al., Reference Tourangeau, Rips and Rasinski2000), this knowledge item is a conservative hypothesis test. Factual items like these are considered direct reflections of accessibility (Delli Carpini and Keeter, Reference Delli Carpini and Keeter1996), since answers are overwhelmingly driven by whether people (do not) have the knowledge to report. We coded correct answers as “1” and all other replies as “0.”Footnote 14 Although Estonian and Russian speakers are both exposed to this fact, we anticipate that this nationalist party's anti-minority rhetoric tightens any mental association between this fact and the Russian language (Pérez, Reference Pérez2016).Footnote 15 Thus, inasmuch as speaking a minority language primes ethnic divisions, those interviewing in Russian will generate more correct answers because this information is more accessible.
RESULTS
Table 1 displays our results from ordered logit (Importance of integration) and logit (Nationalist) models. Language appears to heighten the political salience of ethnic divisions.Footnote 16 Model 1 shows that compared to those who were assigned to interview in Estonian (majority tongue), those who were assigned to interview in Russian (minority tongue) rank minority integration as a more important issue (0.42, p < 0.04). Among those assigned to interview in Estonian, the likelihood of rating minority integration as the least important issue is 0.52; yet among those assigned to interview in Russian, this probability drops ten percentage points to 0.42 (FD = −0.10, p < 0.04).
Table 1 Language Heightens the Salience of Ethnic Divisions
Note. Entries are (ordered) logit coefficients with standard errors in parentheses. Dependent variables are in column headings. ***p < 0.01, **p < 0.05, one-tailed.
We believe interview language drives this sizeable effect. Randomization guards against ethnicity yielding this result, since respondents, regardless of (un)observed characteristics, are more likely to view ethnic divisions as an important political issue when they interview in a minority tongue. We test this result's robustness by using preferred interview language as a proxy for ethnicity (e.g., I prefer to interview in Estonian because I am ethnically Estonian).Footnote 17 We code Prefer Russian interview as “1” for those preferring to interview in Russian, and “0” for those preferring to interview in Estonian. Model 2 shows that after adjusting our treatment effect for this covariance, assigned interview language still impacts the importance assigned to the integration of Estonia's Russian-speaking minority. This underscores language's direct impact on the weight assigned to ethnic divisions relative to other political issues.Footnote 18
Model 3 further supports our argument, showing that those assigned to interview in a minority tongue (Russian) are also more likely to correctly report the Union of Pro Patria and Res Publica as Estonia's most nationalist party (0.50, p < 0.01). Among those assigned to interview in Estonian, the likelihood of correctly reporting this fact is 0.35; yet among those assigned to interview in Russian, this probability climbs 12 percentage points to 0.47 (FD = 0.12, p < 0.00). This supports our claim that interviewing in Russian (minority tongue) enhances the accessibility of knowledge about this nationalist, anti-minority party. Moreover, as Model 4 reveals, the effect of assigned interview language remains substantively unchanged after including Prefer Russian interview as a covariate.Footnote 19
Nevertheless, since Nationalist captures knowledge, it is plausible that survey fatigue hinders recall of correct information. Yet we asked this item early in Study 2. Moreover, if survey fatigue is present, then recall should be more difficult in the language in which political information is less available. In Estonia, national news and partisan messages are more likely to be available in Estonian than in Russian. Thus, recall should have been easier in Estonian than in Russian, yet we find the opposite pattern.Footnote 20
CONCLUSION
Across two experiments, language appears to heighten the political salience of ethnic divisions. Compared to people who were assigned to interview in a majority language (Estonian), those assigned to interview in a minority tongue (Russian) rated ethnic relations as a more important problem than other political issues, and were more likely to correctly recall the nationalist, anti-minority party in their political system.
Our results contribute to scholarship on the activation of ethnic cleavages (Chandra Reference Chandra2012; McCauley Reference McCauley2014; Posner Reference Posner2005): a research area actively interested in better understanding the psychological mechanisms behind them (Adida et al. Reference Adida, Combes, Lo and Vernik2016b). Specifically, our findings indicate that a minority language can reliably prime ethnic cleavages and that, once activated, ethnic divisions are more likely to affect individuals’ political opinions.
Our results also imply that interview language is more than an administrative variable, for it appears to provide access to the mental content people draw on to report opinions. Our finding that a language's status (minority versus majority) can shape opinions about politics opens a door for subsequent research to explore other ways in which language affects people's political views and, possibly, behavior.
Our experiments have limitations, however. We examined a context where a minority language has unequal status and where, historically, antagonism (but no violence) between ethnic groups exists. This setting is special, but not unique. The origin of Estonia's minority population is similar to most post-Soviet states and partly reflects a colonization experience, which other countries have experienced. Furthermore, like other nations (e.g., South Africa, the United States), Estonia has a clear divide between a dominant ethnic group and language, and a subordinate ethnic group and tongue. Yet our findings’ robustness to settings where the ranking of languages is more (un)stable is an open question. Hence, future research should consider whether our results differ across more (less) conflictual settings than Estonia (e.g., nations where debates about ethnic relations revolve around secessionism) or in contexts where a minority population is not a result of foreign occupation.
SUPPLEMENTARY MATERIAL
To view supplementary material for this article, please visit https://doi.org/10.1017/XPS.2018.27
