Ethnic Parties, Ethnic Tensions? Results of an Original Election Panel Study

Ethnic diversity is considered detrimental to national unity, especially if ethnicity is politically mobilized: Ethnic parties in electoral competition in particular are thought to increase the salience of ethnic differences and, with it, ethnic tensions. Yet the individual links of this psychological chain have only been examined cross-sectionally, and never together. This article employs original longitudinal survey data to simultaneously assess changes in ingroup identification, outgroup aversion, and national identification over the election period within one diverse society: Romania. While ingroup identification does increase, ethnic relations do not worsen. On the contrary, outgroup aversion decreases while national identification increases, for minority and majority Romanian citizens alike. I explain these findings with the common ingroup identity model from social psychology: Elections in ethnically diverse societies may not only increase the salience of ethnic groups but also that of the superordinate, national identity. The findings question the often assumed automaticity of

debate implicitly or (more rarely) explicitly locate the mechanism between politicization and polarization in psychological processes among the people constituting the different ethnic groups, this mechanism has not been studied directly: Little attention has been paid to how people perceive ethnic politicization and how politicization affects ethnic relations and national unity (see also Fox and Miller-Idriss 2008;Green and Seher 2003;Ishiyama 2011).
Only recently have studies begun closing this gap. Eifert and colleagues (2010) examine the salience of ethnic identities in cross-sectional surveys of 10 sub-Saharan African countries and find that it tends to be stronger nearer elections, particularly if these are more competitive. Higashijima and Nakai (2016) follow a similar empirical strategy in the Baltic states and conclude that ethnic identification strengthens with electoral mobilization. Both studies examine ingroup identification only, not allowing inferences to changes in ethnic relations. In this article, I make the often implicitly assumed links between ethnic politicization, identification, and ethnic relations explicit. I then employ an original longitudinal survey to adjudicate between different theories on the effects of ethnic politicization by simultaneously assessing changes in ingroup identification, outgroup aversion, and national identification over the election period within one ethnically diverse democracy: Romania. Although I find an increase in ingroup identification, I do not find evidence for worsening ethnic relations over this period. On the contrary, outgroup aversion decreases while national identification increases, for minority and majority Romanian citizens alike.
I explain these-from the perspective of the ethnic politics literature-unexpected findings with the common ingroup identity model from social psychology (see Gaertner and Dovidio 2009): Elections in ethnically diverse societies not only increase the salience of ethnic groups but may also increase that of the common superordinate, national identity. People hence see each other as fellow members of a shared group, and as a result, antagonism between the subordinate groups decreases. The argument is motivated by findings from social and political psychology that question the automaticity of intergroup threat, as understood in the literature on ethnic politics. Yet alternative theories to intergroup threat, such as intergroup contact theory and its offshoot, the common ingroup identity model, have so far been overlooked in the study of ethnic politics-perhaps due to, first, the field's strong focus on antagonistic intergroup relations and, second, its focus on ethnic identities at the expense of other identity dimensions, despite the constructivist literature's insistence that people have a multiplicity of identities whose salience depends on the context (see Chandra 2012).
Aside from the theoretical and empirical contributions, the article also makes two methodological contributions. First, I use data from an original two-wave survey panel conducted before and after Romania's 2016 electoral period. In contrast to cross-sectional surveys, panel surveys interview the same individuals at several timepoints. This allows identifying not only aggregate but also individual change over time, and as such for more confident identification of causality (Lynn 2009). Second, data collection was stratified by ethnic group relations, drawing representative samples of the politically and numerically dominant ethnic Romanian population and the nondominant Hungarian population, as well as of ethnic Romanians living in counties where Hungarians represent the majority. 1 This design allows for a more careful testing of the assumptions underlying the competing theories on intergroup relations, as any effect of ethnic politicization should be particularly pronounced among this third sample.

The Effects of Elections in Ethnically
Diverse Societies

Intergroup Threat
Studies on elections in ethnically diverse societies are predominantly based on theories of intergroup threat (see also Fearon and Laitin 2000). According to intergroup threat theory, relationships across ethnic group boundaries tend to be hostile rather than harmonious: We tend to show opposition to the outgroup while favoring our ingroup. This ingroup bias increases with the perception of threat to the ingroup by the outgroup, that is, especially in times of danger or contention (see Stephan, Ybarra, and Rios Morrison 2009). In electoral competition for political power, group members may perceive a threat through the respective outgroup, particularly when the ethnic boundary is emphasized by political actors like ethnic parties. Elections in ethnically diverse societies may therefore polarize ethnic relations by politicizing them (see also Michelitch 2015). Figure 1 depicts the hypothesized links of the chain in this polarization framework. First, ethnic politicization affects attitudes predominantly by making ethnicity and ethnic differences more salient, such that thought processes-and behavior-are more strongly influenced by ingroup-outgroup considerations (link A). Eifert and colleagues (2010) show that, across 22 survey rounds in 10 sub-Saharan African countries, respondents were more likely to emphasize their ethnic identity over other identities the closer the survey was conducted to an election.
In consequence, the theory goes, ingroup identification increases (link B). Not only does electoral competition make people more aware of the existence of ethnic differences, but also their ethnic identity becomes more important to them as a result. The mechanism may be either cognitive (Valentino, Hutchings, and White 2002) or rational (Eifert, Miguel, and Posner 2010): People may increasingly recognize a "utility of sticking together" during election periods. Others imply that the mechanism is emotional, that is, that people's ingroup identification increases in reaction to fear of the now more salient outgroup (Higashijima and Nakai 2016). This brings us to intergroup attitudes more generally. The salience of ethnicity-and specifically of ethnic differences-also increases outgroup aversion (link C). People are more aware

FIGURE 1 Hypothesized Changes in Intergroup Relations during Elections
Note: The figure, including both full and dashed arrows, depicts how the polarization framework expects intergroup relations to change over the electoral process due to the increased salience of ethnicity. The dashed arrows in addition signify the mechanisms that the integration framework expects to be blocked during the electoral process despite the increased salience of ethnicity.
of how different "the other" is and, as this happens in the context of competition, that this difference is negative.
Finally, studies on elections in ethnically diverse societies also use intergroup threat theory to infer how group members relate to the polity as a whole, the nation. 2 For members of the dominant group, particularly so if it is a titular group, identification with the nation is expected to increase (link D). For members of the nondominant group, however, national identification is expected to decrease as their attitudes toward the group nominally representing the nation worsen and their feeling of belonging to that nation decreases (link E). Overall, then, the polarization framework understands ethnic relations to be antagonistic in the longer term, with the antagonism being spurred in the shorter term by intense periods of politicization, such as elections (see also Weber, Hiers, and Flesken 2016).
However, some studies suggest that the links between ethnic group salience and attitudes toward the outgroup and the superordinate, national group theorized in intergroup threat theory may be blocked under certain conditions. In Figure 1, these blocked pathways are depicted with dashed arrows. According to this integration framework, the political salience of ethnicity may not manifest itself in outgroup aversion (link C) and low national identification (link E) among members of the nondom-2 In this article, the nation as an object to which attitudes or feelings are extended is understood to be the political community (see Easton 1965). inant group if the political system provides them with access to group representation. Large-scale cross-country survey analyses find that the minority-majority gap in feelings of national pride is significantly smaller in countries with greater political equality (Bühlmann and Hänni 2012;Ray 2018;Staerklé et al. 2010). 3 Examining ethnic group behavior, Birnir (2007) shows that the incidence of political violence is significantly lower where minorities have access to power through institutional means such as proportional representation.
Among members of the dominant group, the political salience of ethnicity does not lead to increased outgroup aversion (link C) if they learn, over one or more election cycles, that minority representation does not endanger their own interests. Hajnal (2001) shows that white Americans' attitudes toward blacks improve after experiencing black mayoral leadership. Similarly, Barack Obama's election as U.S. president reduced whites' prejudice toward blacks (Goldman 2012;Welch and Sigelman 2011).
Yet in this integration framework, both the attitudinal changes among minority and majority group members are the outcome of either longer-term processes or first-time experiences, in which members learn that they do not need to feel threatened by the respective outgroup and update their views on that group. But for an election in a political system with long-standing ethnic minority representation, as is the case in Romania, we would expect that this learning process has already taken place; as the links are blocked, the short-term electoral process itself should not have a discernible effect on intergroup relations, even if it temporally increases the salience of ethnicity (Figure 1, link A) and, with it, ingroup identification (link B).

Common Ingroup Identity
The field's strong focus on antagonistic intergroup relations results in an oversight of alternative theories to intergroup threat. Moreover, its focus on ethnic relations ignores one of the key constructivist insights that people have a multiplicity of identities of varying salience (see Chandra 2012); ethnic identity may not always be the most salient identity, even during electoral competition. The assumed automaticity of intergroup threat-barring the interference of contextual roadblocks-reduces constructivism to a form of "as-if primordialism": Although ethnic groups may be constructed, once they are constructed, they are seen to behave as if they were primordial (see also Kalyvas 2008). I argue that elections in ethnically diverse societies do not necessarily worsen intergroup relations but may even improve them, as the electoral process increases not only the salience of ethnic groups but also that of the common superordinate, national identity.
The argument is motivated by findings from social and political psychology that question the automaticity of intergroup threat, as understood in the literature on ethnic politics. First, increased ingroup identification does not necessarily lead to increased outgroup antagonism (Brewer 2001;Reicher 2004). Davis and Brown (2002) show that African Americans with a strong black social identity do not automatically harbor greater antipathy toward whites than those with weaker black social identity. Gibson's (2006) study on South Africa finds no evidence that strong ingroup identities lead to political or racial intolerance. Moreover, any ingroup bias is largely the result of enhancing the ingroup rather than of devaluing the outgroup (Brewer 2001)-and intergroup contact can reduce the ingroup-outgroup differential (Pettigrew 1998). This intergroup contact theory is the main competitor to the intergroup threat theory in social psychology.
A second set of studies questioning the automaticity of intergroup threat finds that increased ingroup identification does not necessarily lead to decreased national identification among minorities (Sidanius and Petrocik 2001). In the United States, de la Garza, Falcon, and Garcia (1996) find that Mexican American citizens are equally or even more patriotic than Anglo American citizens, whereas Citrin, Wong, and Duff (2001) conclude that ethnic and national identities are generally complementary rather than competing. In multicountry studies, Dowley and Silver (2000) only get mixed results, suggesting that exogenous factors affect the relationship, and Elkins and Sides (2007) find that minorities can feel attached to both the ethnic group and the nation at once. This simultaneity of nested group identities is taken into account in the common ingroup identity model in social psychology, aligning it with constructivist theory on ethnic politics. The model argues that a common ingroup identity such as national identity can extend or redirect "the cognitive and motivational processes that produce positive feelings toward ingroup members to former outgroup members" (Gaertner, Dovidio, and Bachman 1996, 271; see also Gaertner and Dovidio 2009). That is, when the common ingroup is made salient and people see each other as fellow members of a shared group, antagonism between the subordinate groups and, with it, any ingroup bias decreases.
In social psychology, the model is supported by various laboratory experiments, survey studies, and field experiments (see Gaertner and Dovidio 2009). In political science, Transue (2007) shows that making the national identity salient to U.S. study participants improves intergroup relations by reducing the social distance to citizens of different racial backgrounds. Similarly, Levendusky (2017) shows that highlighting the common American national identity decreases affective polarization between Democrat and Republican partisans. Beyond the U.S. context, increasing the salience of shared national identity has been found to increase pro-social behavior between Hindus and Muslims in India (Charnysh, Lucas, and Singh 2015), as well as trust across ethnic boundaries in Malawi (Robinson 2016).
In these studies, national identity was primed in experimental conditions, except for one study in Levendusky (2017), in which he examines the priming effect of Independence Day celebrations. Yet it is reasonable that general elections may also serve as a primer of broader national identity, particularly in its civic aspect of a political community. 4 First, ethnic parties are rarely concerned with issues pertaining to ethnic group interests alone but also campaign on general interest issues like taxation or foreign affairs. Second, ethnic parties are rarely the only, let alone the biggest, parties in any election. Often political parties portray their policy positions as affecting the country and its future as a whole, and the mobilization efforts of campaigns build political efficacy and social capital among the citizenry (Banducci and Stevens 2015;Rahn, Brehm, and Carlson 1999). Citizens may therefore also have a heightened awareness of their membership in a wider political community.
The common ingroup framework as well as the polarization and integration frameworks can be put to the test by observing changes in intergroup attitudes during an election period. Table 1 summarizes the changes in ingroup identification, outgroup attitudes, and national identification we would expect according to each framework for citizens from three different constellations of ethnic group relations: those in the overall majority and those in the overall minority, as well as those citizens of the majority group living in areas dominated by the minority.
The common ingroup framework would lead us to expect an increase in both ingroup and national identification for each group, and-as a consequence of perceiving the respective ethnic outgroup as part of the shared national ingroup-also more positive attitudes toward that outgroup (Table 1c). In both the polarization framework and the integration framework (Table 1a and Table 1b, respectively), ingroup identification also increases for all groups, though in the polarization framework the increase is likely stronger among the two minority populations, as for these groups the intergroup threat is felt more keenly. Accordingly, the change in outgroup attitudes and national identification is also stronger for these two groups than for the majority population, with the overall minority population experiencing a marked decline in national identification. In the integration framework, on the other hand, we would expect no change in outgroup aversion or national identification in any group over the short term once a learning process has taken place over the longer term.

Ethnicity in Romanian Politics
According to the 2011 census (Institutul Național de Statisticȃ 2011), Romania's slightly over 20 million people are divided into over 20 recognized ethnic groups, including ethnic Romanians, who comprise 83.5% of the population. The largest ethnic minority is that of Hungarians, with 6.1% of the population, followed by Roma at 3.1%, and Ukrainians, Germans, and Turks at 0.2% each. While ethnic minorities are scattered across Romania, they are most numerous in the historical region of Transylvania in the center and northwest of the country, where much of the Hungarian population is concentrated.
In the counties of Harghita and Covasna, Hungarians are even in the numerical majority, with 71.6 and 82.9% of the population, respectively. 5 Up until the beginning of the twentieth century, Hungarians were the socially, economically, and politically dominant population in Transylvania. This changed in the interwar period and particularly in the communist era, when the influence of Hungarians and Hungarian language and culture were curtailed. Following the fall of communism, the 1990s saw minority mobilization for community rights as well as political backlash among the majority population, with violent manifestations (Birnir 2007;Stroschein 2012). Nowadays, ethnicity still plays a role in Romanian elections. A number of minority organizations compete for reserved seats, one assigned to each recognized minority (see King and Marian 2012). The relatively large size of the Hungarian minority, however, allows Hungarian politicians to clear the 5% vote threshold, and they hence compete not for a reserved seat but for proportional representation on the general ballot. 6 The largest Hungarian party is the Democratic Alliance of Hungarians in Romania (DAHR; Uniunea Democratȃ Maghiarȃ din România / Romániai Magyar Demokrata Szövetség). Since the 1990s, the DAHR has consistently obtained at least 80% of the Hungarian vote and thus parliamentary representation (Kiss and Székely 2016).
While the DAHR is ideologically flexible enough to serve as coalition partner to Romanian parties from both the left and right in parliamentary politics, Hungarian minority rights are high on its official agenda and in its contact with its constituency (Andriescu and Gherghina 2012;Kiss and Székely 2016). According to the Ethnonationalism in Party Competition data set, which summarizes country experts' coding of the importance of ethnic issue positions to various parties in Eastern Europe (Szöcsik and Zuber 2015), ethnonationalism, cultural autonomy, education in and of Hungarian, the use of Hungarian language, and territorial autonomy are highly important to the party (coded as at least 9 on a 0-10 scale). According to a 2013 survey, the Hungarian electorate is confident that the DAHR effectively represents their community's interests in national government, more so than any other Hungarian party (Kiss and Székely 2016).

(a) Polarization
Overall majority Overall majority + 0 0 Majority in minority Overall majority + + + Majority in minority + + + Overall minority + + + Note: The signs illustrate the expected changes in attitudes concerning intergroup relations (columns) among ethnic groups of different power status (rows) over the election period, according to the polarization, integration, and common ingroup identity frameworks. Two signs indicate that the effect is expected to be relatively stronger, compared to one sign.
Ethnic issues are not just a fringe concern of the minority parties in the reserved seats or of the DAHR: When examining the same issues for mainstream Romanian parties, issue importance ranges from 3.4 for the use of the Hungarian language to 6 on ethnonationalism, to which they are largely opposed. Adding the issue importance of the nationalist Greater Romania Party increases the numbers further (Szöcsik and Zuber 2015). That is, while possibly not as central to party programs as, for example, the economy or foreign policy, ethnic concerns do play a role for Romanian political parties too (see also Andriescu and Gherghina 2012).
The December 2016 general elections were no different. While the DAHR campaign covered general topics such as the economy, education, infrastructure, and rural development (DAHR 2016b) and advocated peaceful interethnic coexistence (DAHR 2016e), Hungarian minority representation was at the heart of the campaign. The party advocated the extension of existing minority rights (e.g., DAHR 2016i; DAHR 2016l) but also warned of their rollback, evidenced, it argued, among others by "an attack on the entire Hungarian community" in the form of the court-mandated closure of a Hungarian-language school in Târgu Mureș (e.g., DAHR 2016a).
To clear the 5% threshold and thus be able to preserve Hungarian rights, the DAHR increasingly called on Hungarians to vote. The message was reinforced by Hungarian Prime Minister Viktor Orban (DAHR 2016g) as well as by entrepreneurs, who likened DAHR representation to the "survival of Hungarians" (DAHR 2016j), and by church leaders, who equated the act of voting to a "religious duty" (DAHR 2016f) to protect the community (DAHR 2016d). On Election Day, the DAHR reported on mass text messages sent by Romanian nationalist politicians to encourage Romanians to vote, as this could prevent Hungarian representation in parliament (DAHR 2016k). This report followed warnings of increased levels of anti-Hungarianism in Romanian society more generally (DAHR 2016c). The DAHR campaign hence clearly politicized the ethnic boundary between Hungarians and Romanians and increased its salience. In the end, the party cleared the electoral threshold with 6.2% of votes, resulting in 21 seats in the lower chamber and nine seats in the senate. It does not form part of the government, but is in a confidence-and-supply agreement with the governing coalition of the Social Democracy Party (Partidul Social Democrat) and the Alliance of Liberals and Democrats (Alianța Liberalilor și Democraților).

The Romanian Election Panel Study
The election campaign thus had the potential to polarize ethnic relations, particularly so in the counties in which the Hungarian population is concentrated and in which Romanians are in the minority. To assess the dynamics of ethnic relations, I conducted a panel survey surrounding the general elections on December 11, 2016. The first wave was conducted 4-6 weeks before Election Day, before the official campaign start on November 11 (OSCE/ODIHR 2016); the second wave was 4 weeks after, in mid-January 2017. 7 The survey included questions concerning respondents' political interest; party affiliations, voting behavior, 7 Wave 1: October 24-November 7, 2016; Wave 2: January 9-24, 2017. The DAHR launched its election manifesto on November 6, 2016 (DAHR 2016h). and perceptions of electoral fairness; satisfaction with the economy, government, and democracy; social trust; national identity; group identification and relations; and demographic information. In this article, I examine three key sets of questions covering ethnic relations: belonging to one's own ethnic group, evaluation of the respective outgroup, and belonging to the national political community.
To assess the effects of elections on ethnic relations, I surveyed representative samples of voting-age respondents living in three different constellations of ethnic group relations: (a) Romanians living in the majoritarian Hungarian counties of Harghita and Covasna (n = 401 in both waves); (b) Romanians in majoritarian Romanian counties (n = 417; in the following, "national Romanians"); and (c) Hungarians in counties in which they make up a substantial share (n = 423). 8 The initial response rate was 65.8%, in line with other large-scale representative surveys, and the attrition rate was comparatively low at 8.3%, with no apparent bias in attrition across samples. 9 With the three samples, it is possible to examine all links in the theoretical chains described in Figure 1 and Table 1, by observing changes in the majority population and the minority population, as well as the majority population living in minority-dominant counties. Among the latter two samples, any election effect should be particularly strong because it is here that the DAHR focuses its electoral campaign, whereas mainstream parties present themselves as "defenders of the 'Romanian minority'" in Harghita and Covasna and unite against the Hungarian vote in more balanced counties (Kiss and Székely 2016, 604).
The panel survey design has several advantages over other research designs. In contrast to cross-sectional surveys, it allows identification of not only aggregate but also individual change over time, and thus also for more confident identification of causality (Lynn 2009). In contrast to many experimental designs, the panel survey design lends external, and specifically ecological, validity to the results, as it covers attitude change among representative samples and in real-world contexts rather than in simulations (Holbrook 2011). 8 The Hungarian survey was conducted in Bihor, Cluj, Covasna, Harghita, Mures, and Satu Mare (see Figure A1). 9 See Table A1. Initial response is unlikely to be biased along interethnic attitudes, as the survey was introduced as an election study, without mention of a focus on ethnic relations. Neither is there an apparent bias by media attention or political interest (see Table A2). Data on demographic representativeness are presented in Table A3.

Results
I examine the data for their fit with the expectations derived from the polarization, integration, and common identity framework (compare Table 1). I first examine differences between samples and changes within samples for the three outcomes of interest-ingroup identification, outgroup attitudes, and national identificationfollowed by additional tests of the common ingroup identity framework and of alternative explanations. The variables and methods used are described in the respective sections.

Ingroup Identification
To test the first set of hypotheses concerning ingroup identification, I use two questions: 10 "On a scale of 0-10, how important is being [Romanian/Hungarian] for you personally?" "I am going to ask you to use a scale like a thermometer to express your evaluation of members of this group. 100 degrees means you typically evaluate them to be extremely favorable, 50 degrees means neither favorable nor unfavorable, and 0 degrees means extremely unfavorable." Responses to the importance and feeling thermometer questions are strongly associated with each other ( 2 w1 = 513.49, 2 w2 = 862.06, p < .001), and after rescaling the latter to range from 0 to 10, I create an index out of their mean.
The results for ingroup identification do point to changes associated with the election, as hypothesized in all three frameworks (Figure 2a). 11 In Wave 1, identification is relatively high for all three samples, with mean values 8.76 to 8.90 on a 0-10 scale; the means do not differ significantly between samples. The values increase even further in Wave 2, by an average of 0.22 points among national Romanians and 0.45 among Hungarians. But the strongest increase occurs among Romanians living in the Hungarian-majority counties of Harghita and Covasna, with a mean increase of almost 1 point. All changes are significantly different from 0 (p w2-w1 ࣘ .004). although the changes may appear relatively small, they amount to an increase of up to 11 percentage points, and that despite a likely ceiling effect. The results are in accordance with intergroup threat theory, and hence the polarization and integration frameworks, in that they suggest that the election more strongly affected the majority population in counties in which they are not dominant, perhaps as a backlash against Hungarian election rhetoric. However, they do not inform us whether perceived threat is indeed the mechanism behind increased ingroup identification.

Outgroup Attitudes
For the second set of hypotheses concerning outgroup attitudes, I use the same thermometer question as above, but for the respective outgroup. Figure 2b describes between-and within-sample differences in outgroup attitudes. Note that values below 5 signify outgroup aversion, 5 neutrality, and those above 5 affinity. In Wave 1, all samples report outgroup affinity. Among national Romanians, the extent of affinity is limited at 5.34, but among Romanians in Harghita and Covasna and among Hungarians, affinity is relatively high-and significantly higher than among national Romanians-at values above 7. This pre-election result alone suggests that intergroup contact rather than threat theory is applicable in the Romanian context: In counties with a higher probability of intergroup contact, between-group affinity is stronger than in counties with a lower probability of intergroup contact. Just as for ingroup identification, outgroup affinity increases between waves for all three samples. Among national Romanians and Hungarians, the increase is about 1 point, while among Romanians in Harghita and Covasna, the increase is almost twice that with 1.84 points. All increases are statistically significant (p w2-w1 < .001). The election thus did not have a polarizing effect on ethnic relations: On the contrary, outgroup affinity increased among all samples, and most among Romanians who would be most strongly affected by the minority's increase in power. Hungarians, too, report increased affinity toward Romanians, so neither can it be said that the minority is being incited against the majority.

National Identification
To test the third set of hypotheses on national identification, I also create an index of the mean of the responses to two questions: "To what extent are you proud of being a citizen of Romania?" and "On a scale of 0 to 10, to what extent do you agree with the following statement? It is important to me to promote a positive image of Romania in contact with foreigners or when I go abroad" ( 2 w1 = 1,240.2, 2 w2 = 838.06, p < .001). Figure 2c describes between-and within-group differences in national identification. In Wave 1, the mean is, not surprisingly, lowest among Hungarians, but still relatively high at 6.61 on a 0-10 scale. And even though it is higher among Romanians in Harghita and Covasna, at 8.09, this value is still significantly smaller than that of national Romanians, which lies at 8.64. The latter value does not substantively change in Wave 2, possibly due to the already high value in Wave 1. However, national identification increases among Romanians in Harghita and Covasna by more than 1 point to 9.16, and among Hungarians by nearly 2 points to 8.51. The changes differ significantly from each other as well as from zero.
The data therefore indicate that in the time between the survey panels, national identification among national Romanians was relatively stable and that that among Romanians in Harghita and Covasna and among Hungarians increased substantially. Moreover, national identification among Romanians in Harghita and Covasna is now significantly higher than among national Romanians. Both this and the stark increase among Hungarians again strongly suggests that the election campaign did not have a polarizing effect, as otherwise national identification should have decreased or, at the very least, remained stable among Hungarians.

Individual-and Subsample-Level Changes
The aggregate analysis so far has shown that ingroup identification, outgroup affinity, and national identification all increased in all three samples, pointing away from both the polarization and integration frameworks and toward the common ingroup framework. To avoid ecological fallacies, I also examine whether this association holds when disaggregating the data.
First, the national Romanian sample includes respondents from counties in Transylvania, other than Harghita and Covasna, and the rest of the country. As intergroup tensions have historically been highest in Transylvania, it is necessary to examine whether the threat-based frameworks apply to this region, if not outside of it. However, looking at Romanian respondents living in Transylvania (specifically, in Bihor and Cluj) supports the argument set out so far: Although ingroup identification increases significantly, again suggesting higher salience of intergroup relations in these counties, national identification also increases and outgroup attitudes do not worsen in Transylvania either (see Figure A2 in the supporting information).
Second, the Hungarian sample includes respondents from counties where they are in the clear majority (Harghita and Covasna) and from counties where they are in the minority. Threat and contact dynamics may work differently in these contexts; for example, majority Hungarians may be more resistant to the idea of a common ingroup identity than minority Hungarians, who have more opportunity for outgroup contact. However, majority Hungarians show a significantly and substantially higher increase in ingroup identification, outgroup affinity, and national identification than minority Hungarians ( Figure A3 in the supporting information), suggesting that the election period fostered a common ingroup Note: Numbers represent correlation coefficients (r) between changes in the respective attitudes between the pre-and postelection survey wave at the individual level. H&C = Harghita and Covasna identity particularly among those for whom it may not have been as prevalent before. Finally, I examine whether the simultaneity of increases in ingroup identification, outgroup affinity, and national identification holds not only at the aggregate level but also at the individual level. A series of simple correlations shows that the change scores of all three outcomes are positively correlated with each other in all three samples (Table 2). Among national Romanians, the correlations between change in ingroup identification and outgroup attitudes are not particularly large, but positive (r = .21). The largest correlation in this sample is that between the change scores in ingroup and national identification, which indicates that the ingroup identity of Romanians, as the titular group, is strongly associated with national identity. Among Romanians in Harghita and Covasna, as well as among Hungarians, change in ingroup identification and outgroup attitudes is more strongly positively correlated (r = .46 and .55, respectively). That is, among respondents in the samples with the highest rate of intergroup contact, we see a stronger increase in outgroup affinity where there is also an increase in ingroup identity. To reiterate, this result aligns more strongly with the general intergroup contact theory on which the common ingroup framework is based than with the intergroup threat theory explicitly or implicitly underlying much of the literature on ethnic conflict in political science.

Common Ingroup Identity
The stark increase in national identification throughout the population suggests that the common identity has indeed been primed over the election period in between the two survey waves. The data may lend additional Note: n = valid observations; w1 = pre-election wave; w2 = post-election wave; p w1 and p w2 = significance values of two-tailed t-tests of mean differences between the Romanian (national) sample and the other two samples in the respective wave; p w2-w1 = significance values of one-tailed t-tests testing whether the change in within-sample means is negative; H&C = Harghita and Covasna.
evidence for the common ingroup framework. First, if the common ingroup framework is correct and an emphasis on the superordinate identity refocuses ingroup bias from the ethnic group to the national ingroup, we should also see a decrease in social distance between groups (see also Transue 2007). I examine social distance with the differences between the ingroup and outgroup feeling thermometers. Table 3 reports the values for ingroup bias per sample in Wave 1 and Wave 2, and their differences. It is evident that national Romanians generally have a significantly and substantively higher ingroup bias than the other two samples, with feelings toward the ingroup nearly 3.5 points higher than feelings toward the outgroup. This again points toward the intergroup contact theory, with those respondents in areas with higher intergroup contact (specifically, Harghita and Covasna) having lower biases. But among national Romanians, just as for Romanians in Harghita and Covasna, the ingroup bias decreases significantly in Wave 2, showing that the election period indeed coincides with an improvement in interethnic relations. Among Hungarians it decreases too, and although it is only significant at the .1 level, this is so despite the already very low level of ingroup bias among Hungarians before the elections: 79.3% of all Hungarians show an ingroup bias of 0 or 1 in Wave 1, whereas only 65.7% of Harghita and Covasna Romanians do so.
Second, the common identity framework may be further tested by examining attitudes toward pro-social behavior. Field experimental studies on the nation as common ingroup identity have shown that salience of national identity triggers greater support for redistribution and pro-social behavior across ethnic boundaries (Charnysh, Lucas, and Singh 2015;Robinson 2016;Transue 2007). De Cremer and van Vugt (1999) explain such changes with increased value being assigned to the public good. Therefore, if the superordinate, national identity and, in particular, its political aspect have been effectively primed, we would also expect an increase in respondents' support for the public good, here examined with their willingness to support the state by paying their taxes in full. As Table 4 shows, this is indeed the case: Respondents in all three samples are significantly more willing to pay their taxes in full after the elections than before. Both ingroup bias and tax attitudes hence provide further evidence for the common ingroup framework.

Discussion
The positive association between ingroup identification, outgroup attitudes, and national identification over the election period rejects the assumptions of both the polarization and integration frameworks and instead suggests that national elections serve as a primer of a common ingroup identity. Here, I discuss potential alternative explanations. First, might the changes have been caused by factors other than the elections? Commemorative national holidays and sporting contests, terrorist attacks and natural disasters, or other short-term events putting a spotlight on the nation, such as a census, may evoke a sense of national identification (e.g., Bonikowski 2016). Yet, of these, only a few international games and a national holiday occurred between the two survey waves. The games were nothing out of the ordinary, and the Romanian teams usually found themselves on the losing side. The Romanian national holiday on December 1 is for many merely another day to go out and celebrate, without heeding the national meaning, which makes it unlikely that its effects are still felt in January. Moreover, the day is not perceived as a holiday by Hungarians, as it signifies the unification of Transylvania and Romania in 1918 and to them represents a loss of power; rather than a day of celebration, it is a "day of mourning" (Fox 2006, 223). Neither the sporting contests nor the holiday celebrations could thus have had the observed effects.
Second, might the timing of the observations not tell the whole story? For example, although the official campaign period began only after completion of Wave 1, it may be that parties began campaigning earlier and that therefore the attitudes observed in Wave 1 did not represent "base attitudes" but those already affected by campaign rhetoric. However, media use for campaigning is strictly regulated (OSCE/ODIHR 2016), and if the DAHR already campaigned informally beforehand, it is unlikely to have reached a large enough number of citizens, particularly Romanians, to have had a substantial impact. And if it did, there are two possibilities: The attitudes observed in Wave 2 show either (a) a continuation or (b) a reversal of this impact. In the first case, the argument advocated so far that the elections improved intergroup attitudes holds, but because the true base attitudes are not known, it is not possible to pinpoint the extent of the improvement. In the second case, the observations would indeed suggest a polarizing campaign effect, but one that does not last and is already forgotten a few weeks later; the substantive effect of ethnic party rhetoric would not be as strong as feared. The same holds for the argument that polarization should have been strongest shortly before or on Election Day, which remained unobserved. Which case is correct can only be resolved with a follow-up study with more waves over a longer time period. But in either case, we still observe that changes in ingroup, outgroup, and supergroup attitudes are positively associated with each other, rather than negatively as predicted by the intergroup threat theory, underlining this article's arguments that it is necessary to look beyond that theory and to examine the different attitudes separately, rather than making inferences from one to the other.
Third, might the findings be the result of the questionnaire design? Surveys are prone to social desirability bias, in which respondents do not report their true attitudes but those they perceive to align with social norms.
If social norms portray ethnic tolerance and multiculturalism as desirable, respondents may not be willing to express the true extent of their ingroup bias. However, the question most explicitly inquiring about attitudes toward other groups, the feeling thermometer, was phrased identically to that inquiring about ingroup attitudes, and the setup was identical in both waves. Social desirability bias hence cannot explain the change in attitudes between waves. Instead, attitude change may be explained by learning effects, or panel conditioning, with respondents updating their response patterns after answering the same questions earlier. However, updating responses requires remembering the previous responses, which is unlikely over the more than 2-month gap between the survey waves. Moreover, it would be unclear what triggered the learning effect, save perhaps that the election period acted as a reminder of any social norms on intergroup relations. 12 Even if that were the case, it would qualify the common ingroup framework only to the extent that it is not indeed the common ingroup itself but the idea of a common ingroup that effected the change.
Finally, might the changes not have been driven by considerations of identity, but by those of material benefit? Romanian residents of Harghita and Covasna may recognize that there are benefits of being represented by the DAHR, as the indivisible spoils of government participation are distributed in the region as a whole: Romanians there, too, will benefit from investments in infrastructure, housing, and the job market. In this case, we would expect an increase in affinity not only toward the Hungarian outgroup but also toward the DAHR. However, whereas in the pre-election wave, party affinity for the DAHR is higher among Romanians in Harghita and Covasna than in other regions, it decreases in the post-election wave, such that party affinity is the same across the country. 13 That is, DAHR support cannot explain why minority Romanians report increased affinity for Hungarians. 14 Moreover, any such consideration would still not explain the changes among Hungarians and Romanians elsewhere.

Conclusion
Ethnic parties are often seen to undermine national unity, as they are thought to politicize ethnic differences and 12 Research on attitudinal panel conditioning is in its infancy (Sturgis, Allum, and Brunton-Smith 2009). 13 See Table A6 in the supporting information.
14 Neither is the decrease in party affinity specific to the DAHR but affects all parties. raise their salience, and in doing so increase ingroup cohesion and outgroup rivalry. The longitudinal analysis conducted in this article does not support this polarization framework. Although ingroup identification did increase, neither outgroup relations nor national identification worsened.
Neither does the analysis support the integration framework, according to which intergroup threat is blocked by contextual factors such as power relations and the history of intergroup conflict: One may argue that the elections did not worsen intergroup relations because Hungarians make up only 6.1% of the population and are regionally concentrated in Transylvania. As such, Romanians outside of Transylvania might not have been exposed to much of the campaign, and Romanians may feel that even when Hungarians gain representation, their power is relatively easily contained. DAHR's junior partner position in a variety of government coalitions may reassure them. That is, the uneven power relations together with the absence of evidence of a Hungarian threat may explain the lack of a polarization effect among Romanians. The same factors may also explain the lack of a polarization effect among Hungarians, as they have learned that, despite the uneven power relations, proportional representation provided them with the opportunity to enter parliament and even government. However, the data do not evidence attitudinal stability but an increase in outgroup affinity and national identity throughout the electoral process. That is, the findings at least qualify the mechanisms suggested to underlie the integration framework, in that the demographic and institutional context not only block more or less automatic threat dynamics engendered by ethnic politicization, but also allow identities other than ethnicity to become more relevant.
The results invite consideration of alternative hypotheses on intergroup relations. Here, I explained the observed simultaneous increase in all three attitudes with the main contestant of intergroup threat theory underlying the polarization and integration frameworks: intergroup contact theory and its offspring, the common ingroup identity model. The national nature of elections raises the salience of a common ingroup identity and inspires national identification, and, as a result, antagonism between the subordinate groups decreases-despite ethnic party rhetoric. The effect of ethnic parties may hence be negligible for interethnic attitudes, whereas national elections can have a positive effect.
As the discussion of the integration framework shows, the Romanian context may only allow generalization to similarly stable political regimes with similarly low levels of ethnic tensions; we may obtain a different picture if we repeated the analysis in societies marred by systemic discrimination and ethnic violence. Even so, the findings generalize to a large number of countries with ethnic diversity and ethnic parties, such as Belgium, Czech Republic, Finland, Mauritius, Namibia, New Zealand, or Peru (Lublin 2014): There, too, ethnic party rhetoric may not necessarily lead to worsened ethnic tensions. And in less stable countries, it is necessary to identify how ethnic party rhetoric may work together with other factors in fueling ethnic tensions.
The findings presented here question the assumed automaticity of intergroup threat and invite future studies to make one's assumptions about the psychological processes at work explicit, and then to empirically assess each of its component parts, rather than to make inferences from one part about the others. The field's tendency to equate ingroup identification with outgroup aversion or to assume that minority groups and national identification are mutually exclusive has resulted in disproportionate emphasis on intergroup threat to explain ethnic relations. This also implies that we may not easily make inferences from attitudes to behavior, and vice versa: For example, while voting behavior may follow ethnic lines, this does not necessarily indicate nationalistic tendencies or outgroup aversion on the part of either the minority or the majority. 15 Future research needs to more clearly disaggregate interethnic attitudes and behavior in theory and practice, and to explore these in both international and intranational comparison.

Supporting Information
Additional supporting information may be found online in the Supporting Information section at the end of the article.
Online appendix Ethnic parties, ethnic tensions? Results of an original election panel study Figure A1: Percentages of ethnic Romanians and Hungarians in Romania by county Figure A2: Mean values per survey wave for Romanians (nat.) in counties with a Hungarian population above and below national average of 6.1 percent Figure A3: Mean values per survey wave for Hungarians in majority and minority Hungarian counties Table A1: Sample sizes and response and attrition rates Table A2: Media attention and political interest shares per sample Table A3: Population structure, database structure, and weights by age groups, residence, and gender per sample Table A4: Question wording Table A5: Between-and within-samples differences in mean values of ingroup identification, outgroup attitudes, and national identification Table A6: Party affinity towards DAHR per sample and wave