Varieties of Popular Nationalism in Modern Democracies: An Inductive Approach to Comparative Research on Political Culture

Contemporary nationalism is typically framed as an oppositional ideology that legitimates the struggles of ethnic minorities for political sovereignty or, alternatively, justifies the xenophobic claims of nativist fringe groups. The emphasis on nationalism’s incendiary varieties, however, has led to the neglect of everyday popular nationalism—the routine and tacit acceptance of the nation-state as a primary object of identification and loyalty, as well as a fundamental unit of political organization. In an effort to address this gap in research, I examine the cross-national variation in popular conceptions of the nation-state using pooled-sample latent class analysis, a method that allows me to account for both withinand between-country heterogeneity and avoid reductive a priori assumptions about the national boundedness of culture. Having demonstrated that the resulting fourfold typology of popular nationalism is predictive of a wide range of political beliefs and is remarkably consistent across countries and over time, I show how the relative prevalence of the four types of nationalism shifts within countries in response to economic and political events that increase the salience of the nation-state. This study breaks new ground in the study of nationalism and offers a novel approach to the use of survey data in comparative research on political culture. Popular attitudes toward the nation-state have been linked to a variety of other sociologically relevant preferences and behaviors. Researchers have demonstrated that individuals’ beliefs about criteria of legitimate national membership can influence their voting and policy choices, particularly when the nation is made salient in political discourse (Sears 1993; Citrin et al. 1990, 2001). Likewise, restrictive conceptions of a nation’s social boundaries, high levels of national pride and attachment, and feelings of national superiority have been associated with in-group favoritism and out-group prejudice (Blank, Schmidt, and Westle 2001; Ceobanu and Escandell Bonikowski Shared Representations of the Nation-State 2 2008; Kunovich 2009). Historians have made similar claims with respect to elites, arguing that legislative support for more or less exclusionary policies toward immigrants and minorities has been historically shaped by elites’ idealized conceptions of the nation’s character (R. Smith 1988, 1997). Yet, despite the importance of popular understandings of the nation-state, we know relatively little about how these phenomena vary across countries and over time. Much comparative nationalism research conducted by survey analysts and historians has been reductive; all too often, it has identified individual nation-states with specific forms of nationalism without paying serious attention to within-country heterogeneity. This form of “methodological nationalism”, that is, a taken-for-granted view of the nation-state as a naturally bounded unit of analysis, has been widespread in the social sciences (Wimmer and Schiller 2002). Within nationalism research, this tendency is exemplified by the work of Hans Kohn (1944), which classified countries as espousing either a Western or Eastern variety of nationalism. Elaborating a dichotomy originally developed by Meinecke (1970 [1908]), Kohn argues that Western (or civic) nationalism is based on elective membership in a nation that is understood primarily as a political and territorial community, while Eastern (or ethnic) nationalism is based on ascriptive criteria of membership in a nation that is conceptualized first and foremost as a community of descent. Although this typology has become less rigid in its subsequent applications, the assumption that nation-state borders provide natural bounds for 1 The nationalism literature typically defines a “nation” as a group that views itself as legitimately deserving of its own state, due to the distinctiveness of its culture, language, ethnic roots, and/or historical territory. When the group’s boundaries are congruous with those of a state, the entity is referred to as a nation-state. Because the survey questions I rely on make reference to specific countries rather than sub-national groups, I treat all the countries in the sample as “nation-states” regardless if they are home to one or multiple national groups. Furthermore, I use the terms “nation-state” and “nation” interchangeably, even though I recognize the conventional distinctions between them. Bonikowski Shared Representations of the Nation-State 3 homogenous nationalist ideologies persists in contemporary comparative research (e.g., A. Smith 1991; Ignatieff 1993; Schopflin 1995). As Brubaker (2004) has persuasively argued, such “groupism” – the conflation of analytical categories (e.g., the nation-state) with empirical groups (e.g., a community with shared beliefs) – leads to theoretically untenable and empirically inaccurate conclusions. In contrast, this paper seeks to develop a rigorous analytical approach to the study of nationalist attitudes that takes seriously both withinand between-country heterogeneity. By analyzing pooled cross-national survey data using inductive methods, I avoid making a priori judgments about the national boundedness of culture. Instead, I identify common patterns of beliefs among all respondents from thirty countries and only subsequently examine the respondents’ national affiliations along with a variety of other individual-level attributes. Having mapped the heterogeneity of popular conceptions of the nation-state within each country, I ask how those understandings and their cross-national distribution have changed between 1995 and 2003 and how those changes relate to the countries’ evolving economic, political, and national security conditions. The results demonstrate that cross-national differences in popular nationalism are best understood in terms of the relative salience of multiple conceptions of the nation-state within countries rather than in terms of essential country-level differences. Furthermore, the content of the multiple shared representations of the nation-state is remarkably stable, but their relative prevalence within each country varies over time. I argue that this temporal variation represents a popular response to major macro-level events, whose impact on the nation is interpreted and framed by political and intellectual elites and the mass media. Bonikowski Shared Representations of the Nation-State 4 NATIONALIST ATTITUDES This paper is concerned with “nationalist attitudes,” but this concept is bound to be misunderstood if nationalism is viewed solely as a conscious ideology deployed by elites in the pursuit of political projects, like the founding of new nation-states or the reorganization of existing ones. Such elite ideology certainly falls under the rubric of nationalism, but it hardly exhausts its definition. Nationalism can also be understood as a pervasive cognitive orientation based on the taken-for-granted assumption that the nation-state is a natural and primary object of loyalty and identification, as well as a fundamental building block of the modern institutional order (Greenfeld 1995). I refer to this institutionalized and widely diffused perception as “popular nationalism.” The concept of popular nationalism places emphasis on everyday attitudes of ordinary people in all nation-states, including nation-states that do not experience flare-ups of nationalism’s more overt and incendiary varieties. Existing research has focused primarily on explicitly ideological nationalism— particularly as it is employed by radical political movements—rather than on everyday, popular nationalism. The emphasis stems in part from the obvious political and social significance of nationalist mobilization and its destabilizing consequences: nationalist movements to tend to challenge existing institutional arrangements, thereby producing conditions of political and social instability and, in extreme cases, widespread violence. Yet, such unsettled moments are relatively infrequent in established democracies. Therefore, for most observers, nationalism is something that happens elsewhere—typically in new, institutionally unstable, or ethnically fractionalized nation-states—or in the distant past, most notably in the successive waves of nation-state-building during the 19 and 20 centuries. Bonikowski Shared Representations of the Nation-State 5 It is a mistake, however, to equate the relative infrequency of nationalist unrest in contemporary societies with the absence of nationalism. As modernist theorists of nationalism point out, the nation-state is a relatively recent political and economic invention, whose ongoing success depends on the sustained popular belief in the reality of national communities, characterized by a shared sense of common history and culture (Gellner 1983, Anderson 1991 [1983]). The largely unquestioned legitimacy of the nation-state as a cultural and institutional form is continually reproduced by the educational system and the mass media, as well the countless routine interactions between national populations and powerful national symbols, from flags and history books to name places and currency (Billig 1995). Yet, despite the thorough institutionalization of the nation-state, which makes it difficult to think outside of its cognitive constraints, the specific manner in which people conceptualize and frame their own nation-states varies, both within countries and between them. Describing and explaining this variation is the primary objective of this paper. In order to map different ways in which people relate to their nation-states, we must first decide which types of attitudes should be taken into consideration. Past survey research has focused on four distinct types of attitudes: national attachment (how close one feels to one’s nation), national identity (what individual criteria one sees as important for legitimate membership in the nation), national pride (how proud one is of the nation’s achievements in a variety of domains), and hubris (how one’s nation compares to others). Analysts have often honed in on one or two of these categories, claiming to be measuring nationalism in general (e.g., Coenders and Scheepers 2003, Hjerm 2001). In contrast, my approach is to use indicators from all four categories, since they are all likely to be relevant for how people conceptualize and relate to their nation-states. Bonikowski Shared Representations of the Nation-State 6 COMPARING POPULAR NATIONALISM ACROSS COUNTRIES Social scientists are not immune from the tendency to internalize the belief in the ostensibly natural existence of nation-states (Wimmer and Schiller 2002). As a result, they often treat countries as internally homogeneous units of analysis. This has been true of comparative nationalism research in history (Kohn 1944), political sociology (Ceobanu and Escandell 2008), and comparative historical sociology (Brubaker 1992). The emphasis on attitudinal variation at the country level is itself a product of the thorough institutionalization of the nation-state system. Because we are comfortable thinking in terms of national differences, such comparisons seem to have face validity. Identifying the similarities and differences between Americans and Spaniards appears to be more appropriate than doing so for Nebraskans and Andalusians. This leads scholars to look for ideal-typical features of each country’s culture, which necessarily downplays the cultural heterogeneity of national populations. Yet, it is possible that within-country differences are as large, or even larger, than those between countries. The pitfalls of methodological nationalism can be avoided by using analytic tools that are more sensitive to within-country heterogeneity. One such tool is latent class analysis (LCA), a well-documented survey analysis method that clusters respondents based on the similarity in the pattern of their responses to multiple survey items. LCA has been used in a variety of fields, from medicine (Sullivan, Kessler, and Kendler 1998) to marketing (Bhatnagar and Ghose 2002) and cultural sociology (Van Rees, Vermunt, and Verbrood 1999). LCA uses maximum likelihood estimation to model iteratively the relationship between multiple indicators in a data set and a predefined number of latent classes (more specifically, a single nominal variable, in which each value corresponds to a distinct latent class). The Bonikowski Shared Representations of the Nation-State 7 Table 1. Latent class analysis of hypothetical data set Observation Var1 Var2 Var3 A 1 3 2 B 3 4 1 C 3 4 1 D 2 1 3 E 1 3 2 F 2 1 3 G 2 1 3 H 1 3 2 algorithm seeks to find a solution such that the indicators become conditionally independent of one another—that is, their pairwise correlations are reduced to zero—once the latent classes are taken into account. The algorithm then calculates a posterior probability of class assignment for each observation in the data set and classifies the observations as belonging to the class with the highest posterior probability. Thus, LCA should make it possible to identify groups of people who share the same understanding of their nation-states. The logic of the LCA method is illustrated in Table 1, which presents a hypothetical data set consisting of eight observations (A-H) and three variables with values ranging from 1 to 4. All three variables in Table 1 are correlated with one another: the Pearson correlation between Var1 and Var2 is 0.196, the correlation between Var1 and Var3 is -0.385, and the correlation between Var2 and Var3 is -0.981. If these data were analyzed with LCA, the algorithm would group the cases into three latent classes based on the similarity of their responses, with the first class consisting of observations A, E, and H, the second class consisting of observations B and C, and the third class consisting of observations D, F, and G (in these stylized data, the probability of each observation’s assignment to the relevant class would be 1.0). Because the response patterns of the individual cases within each class would be identical, we could conclude that membership in the classes effectively removed all interdependencies between the three Bonikowski Shared Representations of the Nation-State 8 variables. That is, the variables became independent, conditional on the assignment of the cases to the latent classes. If we set aside national boundaries and analyze the pooled data from all the countries, we can use LCA to inductively generate groups (i.e., classes) of respondents who share similar response patterns—that is, similar ways of understanding the nation-state—regardless of their national affiliations. Having obtained this information, we can then ask where these respondents live. If countries do fall into distinct nationalist camps, such as those defined by civic and ethnic nationalism, we should expect some of the latent classes to be found only in some countries. On the other hand, if all the latent classes are observed in all the countries in the sample, we can dismiss the reductive view of nationalism and ask additional questions, like where each class is most prevalent, what are the individual-level predictors of assignment to each class, how does the content and distribution of classes change over time, and what accounts for such changes. It is important to be precise about what it is that an LCA approach to attitudinal data actually measures. A useful analytical tool for thinking about meaning-making in specific cultural domains is the concept of a cognitive schema, which originates in cognitive psychology (Fiske and Linville 1980) and has made its way into cultural sociology (DiMaggio 1997). Cognitive schemata are networks of association that impart coherence and order onto the messy and rapid flow of sensory information to which individuals are exposed in their daily lives. In addition to organizing and interpreting lived experience, schemata feature affective and evaluative components that make it possible for individuals to respond to stimuli in a manner consistent with their past experience and future aspirations. The schematic processing of information happens very quickly and without much deliberation (Lieberman et al. 2002). 2 In this hypothetical data set, the correlation between the variables within classes would be undefined (because the variables become constants—that is, they do not vary within classes), but in most practical applications of the method, classes have some variation within them, leading to correlations that approach zero but remain defined. Bonikowski Shared Representations of the Nation-State 9 Schemata are domain-specific, so it should be possible to map them for the cultural domain of the nation-state. There are many ways in which such research can be carried out. One option, for instance, would be to record associations evoked by country-relevant cues, such as potent national symbols. Another would be to conduct experiments that detect changes in behavior after an experimental group is primed with national symbols. A less contextually sensitive approach, but one that allows for large-scale comparisons of national populations, is to ask respondents a series of survey questions about their perceptions of their nation-state. This is the strategy employed in this paper. Given that this approach requires a certain amount of data aggregation, the resulting cultural models cannot be interpreted as direct measures of cognitive schemata, which are inherently individual-level phenomena. I will refer to them instead as shared representations of the nation (Durkheim 1964 [1895], Moscovici 1984, Thompson and Fine 1999). Such representations average over, and hence abstract from, individually held understandings, but as such they may be able to approximate the kinds of cultural repertoires that are available to individuals in broader political culture. The profiles generated by LCA are shared because multiple respondents subscribe to them (to a lesser or greater degree) and they are representations because they consists of a set of interrelated attitudes that reflect people’s perceptions and understandings of a specific domain of social life. It should be stressed that the attempt to identify distinct shared understandings of the nation-state, or any other domain for that matter, differs considerably from the standard variablebased approach in attitudinal and public opinion research. The assumption here is that cultural representations should be viewed holistically, as the sum of all their constituent parts, rather than as sets of discrete attitudes that can be examined in isolation from one another. The interest is not in people’s opinions per se, but in their cognitive understandings of a given social domain, which Bonikowski Shared Representations of the Nation-State 10 themselves reflect symbolic representations that exist at the supra-individual level (in media accounts, popular culture, political speeches, mass education, etc.). Furthermore, the meaning conveyed by cultural models is inherently relational. A long tradition in structural linguistics and semiotics has demonstrated that symbols derive their meaning from their relationship to other symbols and not from their individual essence (Saussure 1960 [1916]; Levi-Strauss 1963; Mohr and Duquenne 1997; Yeung 2005). Consequently, to understand symbolic structures—that is, culture—we must take into account the constituent parts of those structures, as well as the relations of similarity and opposition among those parts. Leaving out some of the elements is akin to omitting variables in a regression analysis: it produces biased results. It is for this reason that I focus the analysis on all four types of nationalism variables mentioned earlier: national attachment, identity, pride, and hubris. DATA To model the shared representations of the nation-state and analyze their distribution across countries and over time, the paper uses data from the 1995 and 2003 National Identity Supplements to the International Social Survey Programme (ISSP). The ISSP is a representative multinational survey administered independently in each participating country (in the U.S., it is incorporated into the General Social Survey). The survey focuses on respondents’ attitudes on a variety of topics and features a wide selection of sociodemographic covariates. The Aspects of National Identity II supplement was administered in 34 countries between 2003 and 2005. Residents of former East and West Germany were sampled separately, as were 3 It is important to bear in mind that the central question in this project concerns not how people define the nationstate in general, but rather how they define their particular nation-state. A response to the question “what does the concept of a country mean to you?” is likely to consist of references to generic properties of nation-states, like sovereignty, borders, taxation, or shared culture, whereas the question “what does the United States mean to you?” is likely to yield more specific sentiments. Of course, cognitive schemata of particular countries are necessarily linked to schemata of the nation-state as a generic concept. Bonikowski Shared Representations of the Nation-State 11 Israeli Jews and Arabs, bringing the number of separate samples to 36. For the purposes of the analysis, the East and West German samples were combined using the appropriate sample weights and four countries were excluded: Bulgaria, Latvia, and Israel were dropped because their questionnaires omitted a number of nationalism items; and Taiwan was excluded because of the dearth of covariate data stemming from its ambiguous administrative status. All respondents under the age of 18 and over the age of 65 were deleted from the data, as were non-citizens and cases with missing values on more than two nationalism items. The final sample size consisted of 27,790 observations from thirty countries, with an average of 926 respondents per country. To analyze attitudinal change, I rely on data from the 1995 national identity supplement to the ISSP (administered between 1994 and 1996), which featured twenty of the countries included in the 2003 survey. The final sample size for the 1995 data was 18,613, ranging from 608 respondents for Hungary to 1,767 for Australia. The size of the comparative 2003 sample was reduced to 17,574. The two national identity supplements include twenty-six indicators of the four dimensions of nationalism, which are listed in detail in Appendix A, and all of which were included as indicators in the LCA model. The variables were recoded so that higher scores correspond to stronger feelings of attachment, more importance attached to each criterion of national membership, higher degrees of pride, and greater levels of hubris. The observations were weighted using individual-level sampling weights provided by the ISSP, as well as population weights to ensure that all countries contributed equally to the solution. The dataset also includes a variety of covariates, which will be used to predict the likelihood of subscribing to a particular understanding of the nation. These include age, gender, marital status, education, religiosity, urban/rural location, political party affiliation, and parents’ citizenship status. In addition to the ISSP data, I compiled country-level variables from a variety Bonikowski Shared Representations of the Nation-State 12 of sources that will make it possible to examine the causes of change in the distribution of shared representations across countries. SHARED REPRESENTATIONS OF THE NATION-STATE When conducting a latent class analysis, it is up to the analyst to decide how many classes the algorithm should identify. This decision is typically based on measures of goodness of fit, such as the Bayesian information criterion (BIC). Not surprisingly, when one is dealing with over 27,000 sets of responses to twenty-six survey questions, the goodness of fit measures are likely to favor solutions with a large number of classes. However, the number of classes has an inverse relationship with interpretability. The more classes there are, the smaller are the differences between them and the lower is the analytical utility of the overall classification system. Also, more classes result in fewer observations per class, which makes it difficult to meaningfully analyze the correlates of class assignment. It is possible to think of the choice of classes in an LCA model as the resolution with which one wants to view the attitudinal variation. At maximum resolution, 27,790 observations will yield up to 27,790 attitudinal profiles; at minimum resolution, they will yield one attitudinal profile. The statistically preferable solution, one that yields the lowest BIC, is found somewhere in between these two extremes. However, pragmatically, a solution with only a few classes is preferable for the purposes of interpretability. To identify this optimal tradeoff point, it is possible to rely on a similar method to that routinely used in principal component and factor analysis for selecting the most appropriate number of factors. The method relies on a scree plot, which maps the number of factors against the additional information provided by the inclusion of Bonikowski Shared Representations of the Nation-State 13 !1,650,000!! !1,700,000!! !1,750,000!! !1,800,000!! !1,850,000!! !1,900,000!! 1! 2! 3! 4! 5! 6! 7! 8! 9! 10! 2003!Full!Sample! 2003! !1,050,000!! !1,100,000!! !1,150,000!! !1,200,000!! !1,250,000!! 1! 2! 3! 4! 5! 6! 7! 8! 9! 10! 1995!and!2003!Comparison!Samples! 1995! 2003! Figure 1. BIC values by number of classes each additional factor. The optimal stopping point is represented by an “elbow” in the plot, at which the amount of additional information generated by each new factor begins to level off. Analogously, when evaluating a series of LCA models, it is possible to identify an elbow in a two-way graph of BIC by the number of classes included in each model. Figure 1 illustrates the relationship between BIC and number of classes in the ISSP nationalism data. The first graph reflects the full data for the 2003 sample of thirty countries and the second graph reflects the data for the 1995 and 2003 comparison samples of twenty countries. The graphs show BICs for solutions ranging from one to ten classes. For all three samples, the optimal tradeoff point is produced by the four-class solution: At first, the inclusion of additional classes produces large payoffs in BIC improvement, but beyond four classes the payoff declines considerably. A supplementary method for evaluating model fit is to examine how well the model is able to assign individual cases to the latent classes. The assignment process consists of two steps. First, the algorithm calculates a posterior probability of every respondent’s assignment to each latent class. Second, every respondent is assigned to the class for which he or she has the highest Bonikowski Shared Representations of the Nation-State 14 Table 2. Posterior Probabilities of Class Assignment, ISSP 2003. Threshold Class 1 Class 2 Class 3 Class 4 All > 0.9 0.707 0.710 0.673 0.759 0.711 > 0.75 0.848 0.844 0.823 0.880 0.846 > 0.5 0.987 0.990 0.981 0.988 0.986 posterior probability. In the four-class solution, for instance, hypothetical probabilities of 0.5, 0.2, 0.1, and 0.2 would result in assignment to the first class. The ability of the modal assignment process to produce unambiguous results can itself be used as a measure of model fit. If the majority of respondents have high posterior probabilities then we can be confident that the classes provide a reasonable fit to the data. The proportion of cases in the full 2003 sample that exceed the 0.5, 0.75, and 0.9 probability thresholds for class assignment are presented in Table 2. The proportion denominators are the counts of all the cases assigned by the algorithm to the corresponding class using the modal probability method. The results demonstrate that the assignment process in the four-class solution is quite accurate, with 71.1 percent of cases having a posterior probability greater than 0.9 and 84.6 percent of cases having a probability greater than 0.75 (a lower but still highly discriminating probability threshold). A probability greater than 0.5 is the minimum threshold for necessarily unambiguous class assignment and, as the table illustrates, this threshold is exceeded by the vast majority of cases (98.6 percent). There is some variation in the average posterior probabilities across the four classes, with Class 3 having the lowest modal probabilities and Class 4 having the highest modal probabilities. It appears that assignment to Class 3 is somewhat more ambiguous than assignment to the remaining classes, but the cross-class differences are small. One way to test the impact of lowprobability class assignment is to perform post-estimation analyses with a sample restricted to 4 The proportions are similar for the 1995 and 2003 comparative samples, with 69.9 percent of cases having a posterior probability greater than 0.9, 83.3 percent of cases having a probability greater than 0.75, and 98.5 percent of cases having a probability greater than 0.5. Bonikowski Shared Representations of the Nation-State 15 high-probability cases and compare the results with those generated from a full sample. This strategy was employed for all of the analysis in this paper and no differences were found between the full sample and the restricted sample, further demonstrating the robustness of the four-class solution. In addition to relying on the BIC metric and the accuracy of class assignment, it is also possible to test the robustness of the classes by comparing their content between the full 2003 sample and the 1995 and 2003 reduced comparison samples. This technique will be employed in a subsequent section of the paper. The next step in the analysis is to examine how respondents in each of the four classes structure their understandings of the nation-state. After the classes are identified, the LCA algorithm calculates the probabilities of specific survey responses conditional on class assignment. Based on those probabilities, the algorithm then predicts the distribution of responses to each nationalism measure in every class. By examining these predicted responses, we can get a sense of the attitudinal profile of each class. Given that there are twenty-six variables, each of which has between four and five response categories, the easiest way to compare the classes is to use variable means. 5 The class means for the twenty-six nationalism measures are presented in Figure 2. It is important to remember that the values shown on the graphs represent central tendencies. Consequently, given that most of variables were measured using a forced-choice four-point scale, the 2.5 mark on the graph represents not an individual’s lack of agreement or disagreement with a particular survey question, but rather an underlying distribution of positive 5 The appropriateness of reporting mean values for ordinal data has been debated because the distances between the individual categories may not be equal. I use means here in the interests of parsimony. Having compared each mean value with the underlying variable distribution, I am confident that the means accurately capture the response patterns found in each of the four LCA classes. Bonikowski Shared Representations of the Nation-State 16 and negative responses among all the respondents assigned to that class. For instance, the mean score of 2.49 on the item “Living in a country most of life” corresponds to the following proportions of responses: not important at all, 0.12; not very important, 0.37; fairly important, 0.42; and very important, 0.09. To aid in the interpretation of the results, it is useful to label the latent classes based on their content. I have labeled the four classes as liberal, critical, populist, and ultranationalist. To the extent that the labels capture the most distinct aspects of each class, they represent a first step in the development of a comparative typology of nationalist attitudes. Class 1: Liberal nationalism. The defining characteristics of the first class, which comprises 36.46 percent of the sample, are its moderate scores on the attachment, identity, and hubris variables, combined with a high degree of pride in all domains of the nation-state. Respondents in this class feel close to their region (mean of 2.99) and country (3.37) and are ambivalent about their attachment to their continent (2.64). Their notion of who is a legitimate member of the nation tends toward civic nationalism, with more emphasis placed on elective criteria, like respect for institutions and laws (3.38), language ability (3.34), and subjective feeling (3.30) than on ascriptive criteria, like religion (1.88), ancestry (2.32), and birth (2.71). The pattern of responses to the pride items stands in contrast to the moderate values on the attachment and identity variables: members of Class 1 exhibit a high degree of pride in all aspects of the nation-state. In fact, only Class 4 has higher mean pride scores. None of the means for the pride variables in Class 1 fall below the mid-point of the response scale (a score of 2.5) and only two of the means indicate ambivalence toward the specific domains of the nation-state: pride in the armed forces (2.72) and pride in the equal treatment of groups (2.73). 6 I interpret the middle of the response scale, ranging from 2.25 to 2.75, as reflecting ambivalence within the class concerning the specific domain evoked by the survey question. Bonikowski Shared Representations of the Nation-State 17 Figure 2. Variable means by latent class, ISSP 2003. a Att = Attachment; Id = Identity, Prd = Pride, Reg = Hubris, Oth = Shame 1.000! 2.500! 4.000! A>:!State!