Identities in between

Conflict and Identity Change among Exclusive and Dual Identifiers

Individuals can identify with multiple social categories at the same time (gender, class, occupation, religion, nationality, ethnicity, etc.), although not all categories are equally salient for all individuals or in all circumstances. ¹ National identification is an important category for the majority of people, something explained by the fact that the modern state is a powerful agent that shapes how individuals categorize and identify (Brubaker and Cooper 2000). In societies with territorially concentrated ethnic groups, identification with the nation often coexists with identification with an ethnic group.

Ethnonational identities can be thought as occurring in a continuum. At the extremes of this continuum, ethnic and national identities are exclusive; in intermediate positions, identities are compatible and inclusive.

This characteristic sets apart ethnonational identities from other identities that exist in a binary way, so that identification with a group precludes identification with equivalent groups (e.g., identification as “Protestant” precludes identification as “Catholic”).

Dual ethnonational identities are very frequent in the regions where ethnic groups concentrate. In fact, they are often the modal form of identification (Stepan, Linz, and Yadav 2011). For instance, in Flanders in 2010, dual identifiers were the largest category of identification: 42 percent of the population felt just as Belgian as Flemish, as compared to the next largest category where 28 percent felt more Flemish than Belgian (De Winter and Baudewyns 2014). In Quebec in 2011, the largest category, 35 percent, felt as Canadian as Quebecois and 31 percent felt more Quebecois than Canadian (Blais 2015). In the Basque Country in 2012, 39 percent of the population felt as Spanish as Basque and 23 percent felt more Basque than Spanish (Centro de Investigaciones Sociológicas). In Catalonia in 2013, 33 percent of the population saw themselves as both Spanish and Catalan and 27 percent as more Catalan than Spanish (Centre d’Estudis d’Opinió). In Scotland in 2013, 29 percent of the population defined themselves as equally Scottish and British and 29 percent as more Scottish than British (Scottish Social Attitudes Survey).

Family origin, parents’ purposive ethnic and national socialization, the language spoken at home, and experiences in the neighborhood influence the development of dual identities (Chernyha and Burg 2012; Barceló 2014; Hierro 2015). Once formed, identification with one’s ethnic or national group becomes stable, but not inalterable. Institutional reforms, such as educational reforms, can modify the demographics of ethnic and national identifiers across generations (Aspachs-Bracons et al. 2008; Clots-Figueras and Masella 2013). Identities can also change over the life course due to shifts in citizenship laws (Laitin 1998), electoral competition (Chandra 2004; Chandra and Boulet 2012; Eifert, Miguel, and Posner 2010), violent conflicts (Kaufmann 1996; Fearon and Laitin 2000; Kalyvas 2008; Rohner, Mathias, and Zilibotti 2013; Sambanis and Shayo 2013), riots (Wilkinson 2012), or threats of interstate conflict (Gibler, Hutchison, and Miller 2012).

Until now, the literature has argued that violent conflict shapes identities in a specific way: “Violence has the effect, intended by the elites, of constructing group identities in more antagonistic and rigid ways” (Fearon and Laitin 2000, 846). Although the effects of conflict on identity increase with the intensity of violence (Balcells 2012; Rohner, Mathias, and Zilibotti 2013), nonviolent political conflict between ethnic groups can also be expected to polarize ethnonational identities. This idea is also supported by classic social psychology studies, which suggest that even minimal forms of categorization and conflict have polarizing effects (Sherif et al. 1961; Tajfel et al. 1971). Competition between groups reinforces the “us” and “them” distinction, hardening rival identities.

Building on this previous research, we expect that nonviolent conflicts polarize the identities of people with an original weak identification with the out-group, that is, exclusive identifiers. The effects on individuals with dual identities are less straightforward and have not been explored in previous research. Individuals may react to political conflict in two different ways: by taking sides and prioritizing between their two identities or by withdrawing from the situation of conflict. The first possibility that dual identifiers choose between their two identities is consistent with the predictions of the literature on political cleavages, which suggests that party competition along the ethnonational cleavage contributes to reproducing and crystallizing the identities of both groups, and pushes individuals in between to choose among their loyalties (Lipset and Rokkan 1967). This would imply that individuals with dual identities prioritize one of their identities when these clash.

The literature on cross-pressured voters suggests a different reaction: that dual identifiers do not prioritize one identity, but withdraw from politics instead. Cross-pressures appear when individuals belong to two or more social groups with conflicting values or preferences, and receive opposing pressures from these groups (Lazarsfeld, Gaudet, and Berelson 1944; Berelson, Lazarsfeld, and McPhee 1954; Mutz 2002b; Brader, Tucker, and Therriault 2014). Dual identifiers can be seen as cross-pressured individuals at times of heightened ethnonational conflict. They experience social pressures coming from the ethnic and national groups, and they are conflicted by the opposing political interests and preferences of the two groups.

Cross-pressures have been found to have two simultaneous different effects. First, cross-pressured individuals exhibit a greater understanding of the motives of both groups, and hence higher political tolerance than other citizens (Mutz 2002a). They engage in more intrapersonal and interpersonal dialogues between different points of view and experience both perspectives in a dispute as their own (Mutz 2002a). Because of their in-between position, dual identifiers should also be able to empathize with the two sides in the conflict and, consequently, understand their motives. Second, cross-pressured individuals demobilize politically by participating less frequently in elections and other political activities (Berelson, Lazarsfeld, and McPhee 1954; Mutz 2002b). Different factors lead to demobilization: ambivalence regarding their political stands (Mutz 2002b; Jang 2009); a psychological need to avoid conflict leading individuals to retreat from political discussion and resulting in lower political interest (Jang 2009; McClurg 2006); and the distress produced by being forced to choose between two stands (Mutz 2002b). We contend that duals find themselves in a similar situation to cross-pressured voters. Torn by the political conflict that confronts the two groups they belong to, dual identifiers should react by distancing themselves from the source of conflict (i.e., politics) and becoming less interested and more distrustful of politics.

None of the aforementioned processes, namely, the development of increased tolerance, and the growing estrangement from identity politics, are expected to lead dual identifiers to change their identities when exposed to conflict. Instead, we hypothesize that dual identifiers retreat from the political domain in order to avoid the discomfort derived from cross-pressures and become more disinterested and cynical about politics. Once they are disengaged from the conflict, they are also released from the two groups’ request to choose between the two conflicting identities.

In sum, we expect that the dynamics of conflict in a context with dual identities result in a polarization of identities at the extremes and absence of change, but a disengagement from politics of citizens with positions in between.

Case Study

Catalonia is an appropriate case study for examining the impact of ethnonational conflict upon the identification of exclusive and dual identifiers. The region hosts an intense ethnonational conflict and a large number of dual identifiers. Some people identify with Spain, but not with Catalonia. Other people identify only as Catalan and not as Spanish. Dual identifiers identify simultaneously with Spain and with Catalonia, and often to varying degrees. ²

Catalan identity is a very relevant identity in Catalonia and has language as its core ethnic attribute (Conversi 1997). Political conflict between Spain and Catalonia has a long history that dates back to the end of the nineteenth century. After the end of Franco’s dictatorship and the advent of democracy in Spain, Catalonia recovered its political autonomy in 1977. Self-government was granted to Catalonia in the 1978 Spanish Constitution, and in the 1979 Statute of Autonomy (i.e., regional constitution). Since 1979, Catalan political parties have pushed to enlarge Catalonia’s self-government powers. Statewide Spanish parties, in turn, have fought back to secure the status quo, arguing that Catalonia had been granted enough autonomy. In 2005, the conflict became more pronounced when the Catalan Parliament sponsored a new Statute of Autonomy with enhanced self-government responsibilities. The reform process was surrounded by divisive campaigns and negative statements made by political elites on both sides. After being approved in a referendum in 2006, the new statute was curtailed by a ruling of the Spanish Constitutional Court in 2010, which prompted Catalan civic organizations and political parties to start actions pushing for self-determination and secession from Spain. In the years since, the debate about secessionism has become the central issue in Catalan politics.

Alongside self-government, one of the most contentious issues in Catalan–Spanish relations has been the linguistic model in schools, which establishes Catalan as the main language of instruction. Criticisms against the model have mainly focused on the idea that it downplays the Spanish language, restricts the freedom of parents to choose their children’s language of instruction, and it is detrimental to students’ learning of Spanish.

Despite the environment of political confrontation, dual identities are very widespread in Catalonia. Mixed family origins, socialization in the two languages, exposure to the two cultures, or contact with out-group members are some of the reasons why people develop a dual identity. According to a representative survey of approximately 2,000 individuals administered in June 2013, the two largest groups in the population are those who feel “as Spanish as Catalan” (35 percent), that is, pure dual identifiers, and the “only Catalan” (32 percent). They are followed by those who feel “more Catalan than Spanish” (26 percent). The percentage of individuals who identify as “only Spanish” and as “more Spanish than Catalan” are about 3 in this survey. Table 1 presents the sociodemographic characteristics of each group using this study.

OPEN IN VIEWER

Table 1.

Sociodemographic Characteristics by Ethnonational Identification.

		Only Spanish	Spanish > Catalan	Spanish = Catalan	Catalan > Spanish	Only Catalan
Gender	Female	50.83	59.58	53.53	52.57	50.15
Age	Age	58.58	46.42	48.41	49.43	48.50
Origin	Born in other regions of Spain	82.46	53.28	30.81	8.28	3.19
	Both parents born in other regions of Spain	7.33	25.08	31.73	18.64	8.88
	One parent born in Catalonia	4.30	11.27	15.95	20.62	17.43
	Both parents born in Catalonia	5.92	10.37	21.52	52.46	70.50
Most often used language	Spanish	83.83	83.20	57.42	11.44	5.05
	Catalan	5.45	9.59	27.03	76.53	91.62
	Both	4.60	7.21	14.90	11.72	3.28
Education	Primary or less	66.75	42.18	48.77	36.08	31.24
Household income	<€1,000	38.29	19.81	16.64	14.03	9.79
	€1,000–€2,000	41.07	42.10	40.63	33.51	33.81
	€2,000–€3,000	6.05	24.93	28.40	30.95	30.34
	>€3,000	14.59	13.15	14.32	21.51	26.06
Observations	N	65	50	595	507	640

Source: CEO Barometer, June 2013.

Note: The entries are weighted column percentages and mean scores. Only individuals whose parents and themselves were born in Catalonia or in other regions of Spain are included in the analyses. Significant differences in the comparison to duals are marked in boldface.

Dual identifiers are quite similar to the rest of the groups in their age and sex composition, but they are in between Catalan and Spanish identifiers in other respects. Descent and use of language are the clear markers of exclusive identifiers, but the values of these variables are quite diverse for dual identifiers. First, dual identifiers’ origin is mixed: 31 percent were born in other regions of Spain, 32 percent were born in Catalonia to Spanish parents, and a noteworthy 22 percent were born in Catalonia to Catalan parents. Second, 57 percent of dual identifiers are Spanish speakers, but fully 27 percent are Catalan speakers and 14 percent speak Spanish as frequently as Catalan. There is a strong correlation between socioeconomic status and ethnonational identification. Exclusively, Catalan identifiers are significantly better educated and have higher incomes than average. “Spanish only” identifiers have lower levels of income and education. These differences are related to the socioeconomic origins of the only Spanish group, which mainly consist of low-educated and unskilled Spanish immigrants who arrived in Catalonia between the mid-40s and the mid-70s. Generally, dual identifiers are a quite heterogeneous group in the most adscriptive determinants of identity (origin and language).

Qualitative Evidence: Hardening among Some, Disenchantment among Others

Do nonviolent conflicts polarize identities? Do citizens with dual identities experience conflict differently than citizens who identify mostly with one group? We conducted thirty-eight semistructured interviews to explore how the Spanish–Catalan conflict affects the ethnonational identities of residents in Catalonia. The interviews were carried out in the province of Barcelona over the summer of 2008, the fall of 2010, and the winter of 2012. The quota sample design takes into account the origin of the respondents and their parents, their age-group, and educational level in order to get a typological representation of the population. ³ The interviews were structured in the following way: at the start, we asked respondents general questions about their biography, socialization, and social background. At the middle of the interview, they were asked about their national identification using the five-point scale and were invited to elaborate on their ethnonational identity and what it meant to them. The next section of the interview was about salient episodes in Catalan–Spanish relationships. Then, respondents answered a brief questionnaire. This section briefly illustrates different reactions in dual and exclusive identifiers and motivates our experiments.

The qualitative evidence suggests that duals had remarkably different discourses and attitudes toward the conflict than exclusive identifiers. People who mostly identify with one of the groups have “hot” reactions in the interviews: they are often aroused and angry about Spanish–Catalan relations when speaking about this topic. In sharp contrast to this, dual identifiers have “cold” reactions: they report feelings of disengagement and claimed that conflict has not affected their identities but has turned them away from politics.

The theme of identity hardening dominates among exclusive identifiers and it also appears among some respondents who feel more Catalan than Spanish. The dominant narrative among this group, which we call “reactive reaffirmation,” can be summarized as follows: some conflictive events in Spanish–Catalan relationships are experienced as an attack to the in-group, and the perception of being attacked makes people feel more distant from the out-group and more identified with the in-group. Many respondents, both Spanish and Catalan, explicitly state that they have changed and hardened their identity in recent years as a reaction to perceived attacks and mistreatment from the other side:

The (Constitutional Court) Ruling on the Statute, and the attitude that the central government is having regarding Catalonia’s financial system and other issues, I consider these serious injuries to the Catalan people, that is, to us (…) The Catalan part of my identity is the one that has felt more attacked. (Interview 37, woman, one parent born in Catalonia couple, Catalan speaker, born in 1949)

Among dual identifiers, by contrast, the theme of disengagement and alienation from politics dominates. When asked how the conflict between Catalonia and Spain has changed their views, members of this group provided dispassionate answers. The ethnonational conflict had led them to withdraw from politics and focus on their private lives.

Look, I am going to confess something to you. Those issues related to self-government and regional politics bore me a lot. I mean, I understand that laws need to be agreed on, but most of the time I skip this news in the newspaper. (…) Issues related to our own politics bore me a lot. (Interview 22, woman, born in other region of Spain, bilingual, born in 1956)

Experimental Evidence: The Heterogeneous Effect of Conflict on Identities

The qualitative evidence appears to suggest that political conflict polarizes ethnic identities only among people with preexisting exclusive identifications with one of the groups and has no effect on dual identifiers. Testing this expectation empirically is challenging because concerns about omitted variables and reverse causality seriously limit our ability to draw inferences from observational data.

In Catalonia, the rise in political conflict since 2005 has coincided in time with other factors that may have affected ethnonational identification independently. For instance, the economic crisis initiated in 2008 may have reduced the appeal of the Spanish identity relative to the Catalan identity because Catalonia is richer and is perceived as less responsible for the economic crisis than Spain. Because ethnonational conflict, the economic crisis, and other contextual factors follow similar trends over time, it is difficult to disentangle the effect of each on identity. ⁴ A different version of this concern relates to reverse causality. For instance, some media are more likely to report on the conflict than others, and individuals with different ethnonational identities may choose to consume different media or no media at all. In that case, ethnonational identification would drive exposure to conflict.

In order to overcome the limitations of observational studies to address our research question, we designed two survey experiments that prime the Spanish–Catalan conflict to assess its effect on the identities of prior exclusive and dual identifiers. In order to judge the magnitude of the treatment effects, we need to consider the stability of the dependent variable and the nature of the treatment. First, identification is a stable individual trait difficult to change in the short term, and we do not expect dramatic shifts due to the treatment. Second, the magnitude of the stimulus effect must be read as the marginal effect of one additional exposure to a conflict prime and not as the full effect of conflict on identity. As already discussed in the Introduction, our sample has been very extensively pretreated on the Spanish–Catalan conflict due to its salience and wide coverage in the media, and we only expect small effects of one additional exposure to conflict. ⁵ The marginal effect of exposure is generally much smaller than the overall effect of the variable of interest (Druckman and Leeper 2012).

These characteristics of the dependent variable and the treatment have important implications for the interpretation of our results. Very large samples are necessary in order to detect small effects as the ones we expect. In our case, the problem of statistical power is compounded because we are interested in heterogeneous treatment effects between subgroups. As the American Statistical Association recommends in a recent statement (Wasserstein and Lazar 2016), when interpreting the results we evaluate p values in relation to the size of the subgroups and the magnitude of the effects.

Experiment 1: Priming the Catalan–Spanish Conflict

We embedded our main experiment, which primes conflict between Spain and Catalonia, in an online survey fielded by Netquest, a commercial polling company. Respondents were selected from a large panel using age, gender, region, and education quotas. To be eligible for the survey, respondents had to live in Catalonia and hold the Spanish nationality. The fieldwork was completed in July 2013 and interviewed 794 respondents.

We randomly assigned half of the sample to receive a prime about Catalan–Spanish conflictual episodes. ⁶ The priming treatment consisted of administering a block of survey questions to one group of respondents, with the aim of making the topic easily accessible in working memory. The design and the strength of this treatment are similar to other primes used in previous research (e.g., Kuo and Margalit 2012). The control group was not administered any of the questions. The four priming questions, administered as a block to all respondents in the treatment condition in the same order, concerned four well-known episodes in recent Catalan–Spanish relations about self-government and linguistic policy. The first question reminded respondents of a campaign against the Statute of Autonomy of Catalonia. The second priming question was about the 2010 Spanish Constitutional Court ruling that declared some articles of the Statute unconstitutional. The third and fourth priming question referred to conflicts about the language of instruction in schools. We provide the question wording in the Online Appendix.

To test our theory, we need to examine the impact of the conflict prime on identity polarization. The expectations drawn from our theory do not refer to the impact that conflict has on Catalan or Spanish identification, but on the difference between the two identities. In the online survey, two separate questions asked to what extent respondents identify with Catalonia and with Spain. The two questions were deliberately placed one after the other (in the same screen). This implies that respondents’ answers to the two questions cannot be detached. The response options for these two variables range from 1 not at all identified to 10 very identified. We subtract responses to the two variables to construct our measure of polarization. A value of 0 in this variable indicates that the respondent feels just as Spanish as Catalan and that polarization is not present. ⁷ A positive value implies that the respondent identifies with Catalonia to a larger extent than with Spain, such that the larger the value, the larger the identification differential, which indicates a high level of polarization. For instance, a respondent who reports being very identified with Catalonia but not at all identified with Spain has a score of 9 (10 − 1). Conversely, negative values indicate that respondents identify to a larger extent as Spanish than as Catalan, up to −9.

We expect heterogeneous effects of the conflict prime depending on preexisting identification with Spain, Catalonia, or both. In order to test this prediction, we use an indicator of ethnonational identities placed at the beginning of the questionnaire as a moderator. The indicator is an anchored scale with five response categories: only Spanish (4 percent), more Spanish than Catalan (6 percent), as Spanish as Catalan (37 percent), more Catalan than Spanish (28 percent), and only Catalan (25 percent). ⁸ Because the two first categories contain very few cases, we collapse them in the analyses and create a category labeled “primarily Spanish.” ⁹

We examine if conflict polarizes identities by assessing whether the treatment had an effect on the identification differential, while subsetting on preexisting identities. ¹⁰ Table 2 reports the mean scores of the relative identification variable in the treatment and the control conditions. The difference in these scores is the treatment effect.

OPEN IN VIEWER

Table 2.

The Effect of Priming Conflict on Ethnonational Identification (Experiment 1).

		No Prime	Conflict Prime	Difference
Primarily Spanish	Identity	−2.54	−3.46	−0.91
	Standard error	(0.51)	(0.42)	(0.65)
	N	35	46
As Spanish as Catalan	Identity	0.29	0.55	0.27
	Standard error	(0.12)	(0.15)	(0.19)
	n	144	150
More Catalan than Spanish	Identity	4.87	4.79	−0.09
	Standard error	(0.21)	(0.20)	(0.29)
	n	111	113
Only Catalan	Identity	7.50	8.12	0.62**
	Standard error	(0.22)	(0.14)	(0.27)
	N	105	91

Note: Two-tailed significance tests. The entries are the values of the identification differential (standard errors), in which 0 means as Catalan as Spanish (and hence holding a dual national identity) and higher or lower values imply a more exclusive identity.

^† p < .1.

*p < .05.

**p < .01.

***p < .001.

The results clearly suggest that the conflict treatment made respondents who identify only with Catalonia, as measured at the beginning of the survey, become even more extreme in their exclusive ethnonational identification. The average score of the identification differential variable (our proxy for polarization) is 7.5 in the control group, but 8.1 (with an upper limit of 9) in the treatment group and the difference is statistically significant (p = .0234, two-tailed t-test). This change represents about 28 percent of a standard deviation, a very substantial change for a stable variable in a population that has been extensively pretreated, and in spite of ceiling effects due to the high value of the dependent variable in the control group. When focusing on respondents who report feeling primarily Spanish at the beginning of the survey, we observe a change in their relative identification in the opposite direction. The magnitude of the difference is even larger than for the Spanish-only identity group (−2.5 compared to −3.5). This effect is about one-third of a standard deviation. Despite the higher magnitude of the treatment effect in this group, we do not find statistical significance using a two-sided t-test, although this can be attributed to the small number of cases in the group comparison (thirty-five cases in the control condition and forty-six cases in the treatment condition). Among the groups in between the extremes, which have a much larger number of cases and hence are better powered, those with a purely dual identity and those who feel more Catalan than Spanish, the treatment does not appear to have any effect.

Taking into consideration the extensive pretreatment of the sample and the existence of floor and ceiling effects at the extremes, it is remarkable to find any effect at all in these groups. It is also noteworthy too that the identification differential remains very close to 0 also in the treated group of pure dual identifiers. This result is consistent with the claim that duals continue to self-identify as duals even after being exposed to the conflict.

One concern with this analysis is that the conflict prime may affect the identities of dual identifiers, albeit in different directions such that the contradictory effects cancel out on the aggregate. To assess this possibility, we examine if the distribution of responses is more spread and the standard errors of the dependent variable are larger in the treatment condition than in the control condition for dual identifiers, but we fail to find evidence consistent with this. More importantly, the null result is robust to subsetting dual identifiers by their origin (if they were born in Catalonia or not) or knowledge of Catalan. The effect of the treatment is somewhat larger among those not born in Catalonia than among those born in Catalonia (0.5 and 0.2), which is in fact counterintuitive. However, none of the differences are statistically significant, and the fact that the sign is positive for both groups suggests that it is not the case that contradictory effects cancel each other. Altogether, the additional analyses subsetting between different types of dual identifiers suggest that their identities are not moved by the prime.

Experiment 2: Criticism or Praise of the School Language Model in the News

We also examine if conflict polarizes identities using a different approach in a second experiment embedded in an online survey administered by Netquest using the same selection procedure as the previous experiment. The fieldwork was conducted between April and May 2012 and 1,500 citizens residing in Catalonia completed the survey.

Respondents read a piece of news in which an expert expressed an opinion either favorable or unfavorable to the Catalan linguistic model of education. The control condition describes a scenario with a lower level of conflict by presenting the linguistic model in a positive light. The treatment condition primes conflict by reminding respondents of one of the main arguments used against the model, namely, that teaching in Catalan reduces students’ proficiency in Spanish. Specifically, in the control condition, a linguistics professor “manifests his support to the Catalan model of education” and claims that this model “does not undermine the learning of Spanish.” In the treatment condition, he uses the reverse argument ¹¹ (see question wording in the Appendix).

The dependent variable, polarization, and the moderating variable are measured as in the first experiment. ¹² Again, a few respondents identify as only Spanish (6 percent) or more Spanish than Catalan (6 percent), and we collapse both categories. Table 3 presents the values of polarization measured in a scale that ranges from −10 to 10, subsetting by level of ethnonational identification.

OPEN IN VIEWER

Table 3.

The Effect of Messages about the Linguistic Model on Ethnonational Identification (Experiment 2).

		Low Conflict over Linguistic Model	High Conflict over Linguistic Model	Difference
Primarily Spanish	Identity	−2.84	−3.59	−0.75†
	Standard error	(0.29)	(0.33)	(0.44)
	N	94	86
As Spanish as Catalan	Identity	0.59	0.77	0.18
	Standard error	(0.10)	(0.10)	(0.15)
	N	275	280
More Catalan than Spanish	Identity	4.90	4.77	−0.13
	Standard error	(0.16)	(0.17)	(0.24)
	N	241	223
Only Catalan	Identity	8.19	8.69	0.49**
	Standard error	(0.17)	(0.14)	(0.22)
	N	151	150

Note: Two-tailed significance tests. The entries are the values of the identification differential variable (standard errors), in which 0 means as Catalan as Spanish (and hence holding a dual national identity) and higher or lower values imply a more hardened identity.

^† p < .1.

*p < .05.

**p < .01.

***p < .001.

We find the exact same pattern as in the previous experiment. In the control condition, which describes a low-conflict situation, respondents who identify as primarily Spanish at the beginning of the questionnaire had an average differential identification score of −2.8, but this value was −3.6 in the treatment condition. This suggests that the conflictual message hardens the identity of respondents with a preexisting Spanish identity. The difference is statistically significant at 10 percent (p = .090) using a two-tailed test and the increase amounts to 27 percent of a standard deviation. Conversely, respondents who identified as feeling solely Catalan earlier in the questionnaire were more likely to have a larger differential in the treatment condition (8.7) than in the control condition (8.2), which is a statistically significant difference (p = .027) of moderate magnitude (23 percent of a standard deviation). We find no effect of the conflict treatment on the two groups with ethnonational identities in between the extremes. Again, the results support the hypothesis that conflict hardens the ethnic identities of people at the extremes and suggest that conflict has no effect on the identities of people in between.

The first experiment primes the Catalan–Spanish conflict directly, while the second experiment uses a more indirect and subtle treatment. Strikingly, in spite of using two rather different treatments and extensive pretreatment, we are able to replicate the same results in both experiments: political conflict hardens preexisting identities among exclusive identifiers but has no any apparent effect on duals. It is very unlikely that our results are an artifact of the design. There is nothing in the design that would lead us to expect heterogeneous effects among duals. Replication across experimental treatments increases our confidence in the validity of our conclusions.

Observational Survey Data: Attitudes as a Mechanism

Because the experiments did not ask about feelings toward politics, they do not provide direct evidence about our main alleged psychological mechanism. The qualitative evidence reveals a prevalence of cynicism in the discourses of dual identifiers. This section mobilizes further observational data from public opinion surveys to explore whether the proposed psychological mechanism is at play.

The main implication of our theoretical argument is that disengagement toward politics has increased among dual identifiers, as conflict has intensified. To examine this prediction, we employ data from two separate surveys conducted in moments with a low (2002) and high level of conflict (2013) in Catalan–Spanish relations. ¹³

This section also addresses one criticism to our argument, which is that a general aversion to conflict may drive both dual identification and differential reactions to conflict. In this case, dual identification would not shelter people from the polarizing dynamics of conflict. Rather, it would itself be a manifestation of a general reluctance to get involved in conflict. This criticism does not undermine the validity of our experimental results, but it modifies their interpretation. If dual identifiers are more averse to conflict in general, we should observe that they are always more disengaged from politics than citizens in other identity groups, even before the increase in Catalan–Spanish conflict. Conversely, if cross-pressured individuals only became disengaged after the rise of ethnonational conflict, as we hypothesize, we should observe that dual identifiers are more disinterested and distrustful about politics in the more recent period. ¹⁴

We use two surveys that include questions about attitudes toward politics conducted in 2002 and 2013 by the Institut de Ciències Polítiques i Socials, a public research institution in Catalonia. Importantly, 2002 was a year with little conflict about ethnonational issues but ripe with other types of political conflict. There was intense controversy about issues such as the possibility of Spain participating in the Iraq war, a general strike, a reform of universities, and an oil spill off the Galician coast. These issues were hotly contested, and many cities hosted demonstrations attended by hundreds of thousands of people. Hence, the level of conflict was high in the two periods, but the content changed over time.

The two representative surveys consist of 2,000 and 800 respondents, respectively. The survey includes an item asking respondents which two feelings best describe their attitudes toward politics. The answer options were enthusiasm, compromise, interest, indifference, boredom, distrust, and anger. In our analyses, we focus on the feelings of more analytical interest for the claim that conflict leads dual identifiers to disengage from politics: interest and distrust.

We estimate four logistic regressions with interest and distrust in 2002 and 2013 as the dependent variables and identification with Spain and Catalonia as our key independent variable. ¹⁵ In agreement with our hypothesis that duals should turn less interested and more distrustful in 2013 as compared to exclusive identifiers, we model the relationship between identification and our dependent variables as curvilinear. We include sociodemographic controls. ¹⁶ Table 4 presents the four logistic models.

OPEN IN VIEWER

Table 4.

Logistic Regression Political Interest and Distrust in 2002 and 2013.

	Interest	Interest	Distrust	Distrust
	2002	2013	2002	2013
Identification	0.64** (0.22)	−0.49 (0.49)	0.23 (0.20)	1.10** (0.41)
Identification square	−0.09** (0.03)	0.13 (0.07)	−0.02 (0.03)	−0.18** (0.06)
Male	0.37*** (0.10)	0.29 (0.19)	−0.08 (0.10)	−0.41* (0.17)
Age	0.06*** (0.02)	−0.00 (0.03)	−0.00 (0.02)	0.05 (0.03)
Age square	−0.00** (0.00)	0.00 (0.00)	−0.00 (0.00)	−0.00* (0.00)
Education
Primary	−0.52* (0.23)	0.37 (0.36)	−0.50*** (0.21)	−0.71* (0.30)
Secondary	0.54*** (0.14)	0.90*** (0.26)	−0.01 (0.13)	−0.10 (0.22)
University	0.91*** (0.15)	1.26*** (0.29)	−0.15 (0.15)	−0.22 (0.26)
Size of municipality	0.07* (0.03)	0.01 (0.07)	−0.06 (0.03)	0.13* (0.06)
Descent
Child of mixed couple	0.02 (0.18)	0.24 (0.26)	−0.08 (0.17)	−0.40 (0.24)
Second generation	0.05 (0.16)	0.35 (0.29)	0.01 (0.15)	−0.07 (0.26)
Immigrant	0.17 (0.15)	0.11 (0.32)	0.03 (0.14)	−0.37 (0.28)
Subjective social class
Lower middle	0.18 (0.28)	0.39 (0.47)	0.02 (0.25)	−0.15 (0.37)
Middle	0.10 (0.27)	0.76 (0.45)	−0.22 (0.24)	−0.49 (0.36)
Upper middle/upper	−0.26 (0.37)	1.08 (0.59)	0.16 (0.33)	−0.26 (0.52)
Constant	−4.23*** (0.60)	−2.80* (1.22)	−0.50 (0.53)	−1.47 (1.02)
Observations	1,896	603	1,896	603
Log likelihood	−1,116	−332.2	−1,230	−386.8
McFadden’s adjusted	0.046	0.045	0.014	0.020
LR χ2 (df)	107.21 (15)	70.83 (15)	27.18 (15)	54.80 (15)
Prob. χ2	.00	.00	.03	.00

Note: LR = Likelihood Ratio. Standard errors in parentheses.

*p < .05.

**p < .01.

***p < .001.

To ease the interpretation of these results, we estimate the probabilities of reporting feelings of interest and distrust in 2002 and 2013 by individuals’ identification (see Figure I).

OPEN IN VIEWER

Figure 1.

Political engagement and identification in 2002 and 2013.

The results confirm our expectations, especially for distrust. The relationship between identification and feelings toward politics is markedly different in 2002 and 2013. Dual identifiers were not disengaged from politics in 2002, and their attitudes toward politics were more similar to the attitudes of exclusive identifiers. However, in 2013, and compared to the exclusive identifiers groups, dual identifiers display a low level of interest on politics. The pattern is the opposite for distrust: with no group differences in 2002 and a significant inverted U-shaped relationship in 2013. Specifically, dual identifiers have an increase of 0.25 in the predicted probability of mentioning this attitude toward politics, compared to lower changes in the other groups. These results imply that cynicism attitudes were similarly present in all identification groups in 2002, but in 2013, dual identifiers show higher levels of distrust.

These findings demonstrate that dual identifiers do not have a stable psychological trait that moves them away from politics. Rather, the results are consistent with our argument that disengagement in the recent period is driven by the ethnopolitical conflict. ¹⁷

Conclusions

This article has argued that political conflict affects the ethnonational identities of exclusive and dual identifiers in heterogeneous ways. Individuals at the poles turn even more distant to the out-group when conflict intensifies. By contrast, dual identifiers, defined as individuals who identify with both the national group and the ethnic group to a similar extent, do not appear to be influenced by conflict in the same way. The experimental, qualitative, and survey evidence presented here broadly supports this claim. Using different conflict manipulations, our survey experiments show that exclusive identifiers become more extreme in their identification when they are reminded of episodes of conflict between Catalonia and Spain. A similar process, however, is not observed among those who feel as Spanish as Catalan. The qualitative and survey evidence suggests that disengagement from politics is a common reaction to conflict among this cross-pressured group.

Ethnonational identities are politically relevant. A broad literature in comparative politics finds that ethnic and national identities strongly influence individuals’ political behavior and political preferences (e.g., Blais and Nadeau 1992; Chernyha and Burg 2012; Clots-Figueras and Masella 2013; Muñoz and Tormos 2015; Henderson, Jeffery, and Liñeira 2015; Balcells, Fernández-Albertos, and Kuo 2015). Some of these studies admit that ethnonational identities are endogenous to the political context, but few test how the political context or, more specifically, the ethnonational conflict affects identity. Our article has sought to do so; hence contributing to the research agenda that sees identity as a consequence, rather than just as a cause, of conflict. Most of this latter research has focused on studying the effects of violent conflicts (Kaufmann 1996; Fearon and Laitin 2000; Kalyvas 2008; Balcells 2012; Rohner, Mathias, and Zilibotti 2013). Unlike these, our article has studied the effect of nonviolent conflict, an inherent feature of democratic politics, on ethnic and national identity. Our research suggests that violence is not necessary to polarize identities; intense nonviolent political conflicts also polarize people with identities at the extremes.

In addition, this article has paid attention to an important group that has been disregarded by this literature: individuals with dual identities. Some previous authors have stressed the political importance of dual identifiers, a numerous group in contexts with ethnonational cleavages, in preventing these societies to be torn apart (Linz and Stepan 1996; Linz 1997; Stepan, Linz, and Yadav 2011). However, this important claim has remained largely untested. Our research adds both empirical substance and some cautionary nuance to the claim that duals are key in preventing multicultural societies to separate. On one hand, our results suggest that, to the extent that dual identifiers do not polarize as a result of conflict, they are an important group that can hold the social fabric of a multicultural society together. On the other hand, the fact that duals become more disengaged from politics at times of heightened conflict may perhaps limit the possibility that this group counteracts the polarization of divided societies. Despite being well placed to do so, dual identifiers in Catalonia have failed to get organized and play a visible mediator role in the conflict between Spain and Catalonia so far. A similar situation can be found in other contexts with an intense ethnonational conflict (e.g., Scotland and Flanders).

As concerns to scope conditions, our theory applies mainly to contexts where ethnonational identities coexist. These include not only multicultural states but also immigrant societies where dual or hyphenated identities are frequent. First-generation immigrants who have been living in the country for many years can identify with the receiving society while keeping their original attachments. Second generations and children of mixed couples are also brought up with two identities. Future research should also investigate whether our finding for dual identifiers travel to other contexts with violent conflicts.

Our article also raises interesting questions about the role of emotions in the processes of identity hardening and duals’ political disengagement. Annoyance and anger appear in many of the discourses of people experiencing processes of identity reaffirmation, suggesting that hatred might play a key role, a claim that has also been stressed in previous studies (Balcells 2012; Petersen 2013). Research on cross-pressured voters claims, in turn, that cross-pressured individuals might feel that conflict threatens the harmony of their social relations (Mutz 2002b). Fear, therefore, could be the emotion driving dual identifiers’ retreat from conflict. Extensive research provides rational choice underpinnings to the process of identity change, but identification implies an affective and not just instrumental attachment to a group. Our research tentatively suggests that emotions and psychological mechanisms may have an independent role in driving identity change.

Supplemental Material

Supplemental Material