The normative underpinnings of a democratic regime emphasize principles so desirable that citizens are expected to recognize its legitimacy. In a simplified form, this question is primarily related to political philosophy. Specifically, according to Dahl (1989) and Lijphart (1984), the relationship between responsibility and responsiveness lies at the core of democracy. Responsibility refers to the degree of accountability within the decision-making processes of a political regime. Responsiveness ensures that the decisions of a political regime adequately reflect public preferences (Lauth, 2020, p. 836). However, these two aspects of democratic rule are partly in conflict. The resolution of this contradiction is explored through various conceptual frameworks ranging from narrower to broader conceptualizations, each reflecting different strands of democracy theory (e.g., Habermas, 1992; Rawls, 1993; Peter, 2008).

Schumpeter (1942) minimalist understandings of democracy are all centered on the electoral process as a solution of both the chain of responsiveness and accountability (Bühlmann and Kriesi, 2013). Dahl (1989) extended this concept and included minimum conditions for democracy besides elections, which is commonly sub-summarized under the term liberal understanding of democracy. In practice, he suggested that democratic regimes—or in his words polyarchies—can be identified by the presence of certain key political institutions and civil liberties: (1) elected officials, (2) free and fair elections, (3) inclusive citizenship, (4) the right to run for office, (5) freedom of expression, (6) alternative information, and (7) associational autonomy. All these aspects are rooted in the rule of law, which asserts that no one, including those in power, is above the law (republicanism), while also ensuring that certain civil freedoms remain inviolable (liberal guarantee).

Some scholars argue that these criteria are not sufficient and have developed a broader understanding of democracy. For instance, another important normative principle associated with democratic rule is the concept of procedural and distributive justice. According to Buchanan (2002), political systems gain moral authority to exercise power not merely through democratic procedures, but also through a reasonable degree of justice in the distribution of material resources, the protection of fundamental human rights, and inclusive political participation. Since democracies are not necessarily just, their legitimacy would have to be questioned. Consequently, Diamond and Morlino (2005) argued that political equality could only be achieved through a certain degree of social and economic equality. Other refinements of democracy theory include deliberative and direct democratic elements (e.g., Pateman, 1970; Barber, 1984; Habermas, 1992). For instance, direct democracy encompasses mechanisms such as initiatives or referenda, which allow citizens to propose specific issues for the political agenda and challenge decisions made by those in power. These elements seek to enhance the direct involvement of citizens in the political process, thereby expanding the scope of democracy beyond representative systems.

It is therefore important to recognize the pluralistic nature of democracy theory and its wide spectrum. At one end, core liberal elements shape any (supposedly) democratic system. At the other end, broader approaches advocate for greater flexibility and variation in how democracy is understood and institutionalized. This distinction serves as a crucial heuristic for navigating the diversity of democratic models and their translation into related measurements intended to capture democratic legitimacy.

In contrast, we distinguish attitudinal components that reflect citizens’ legitimation of political rule on the micro-level, drawing primarily on theory-building in political sociology and political culture research. Typically, the focal point is—according to Weber’s Legitimitätsglaube (faith in legitimation)—the relationship between rulers and the ruled, and the extent to which the latter regards the former’s authority as justified (Weber, 1972).

Both the work of Almond and Verba (1963) and Easton’s conceptualization (Easton 1965, 1975) have become widely recognized as benchmarks, significantly shaping research on orientations toward the nation-state, its principles, institutions, and political actors. Almond and Verba (1963) distinguished four objects of political orientation at a descriptive level: (a) the political system as a whole, including its fundamental values and institutions, (b) participatory processes (input dimension), (c) the performance of the political system (output dimension), and (d) the self as a political actor.

Another influential concept introduced by Easton (1975) is the distinction between diffuse (or generalized) and specific support. While the former is characterized by long-term approval of the fundamental principles of political authority and abstract feelings toward the nation-state, the latter is based on the short-term evaluation of the performance of elected and appointed officials responsible for making and implementing political decisions. Both generalized and specific support—understood as a continuum rather than a dichotomy—are integral to the legitimation process of a political system.

Especially since the 1990s, Easton’s concept, originally open to all forms of government, has been further differentiated, with research showing a clear shift toward democratic systems as the primary focus (e.g., Buchanan, 2002). In addition to numerous empirical studies that address the operationalizability of certain components by using survey-based data to analyze different conceptualizations (cf. Kornberg and Clarke, 1992; Weatherford, 1992; Fuchs et al., 1995; Klingemann, 1999; Dalton, 2004; Anderson et al., 2005; Gilley, 2006; Westle, 2007; Booth and Seligson, 2009), conceptual-theoretical work that further developed the original conceptualization are particularly worth mentioning here. Emphasis is placed on the differentiation of possible support motives (e.g., Fuchs, 1989) or the specification of support objects and types (e.g., Westle, 1989).

Furthermore, the conceptualization has been extensively reinterpreted as a hierarchical continuum, as seen in works by Diamond (1999) and especially Norris (1999, 2001), thereby offering a more nuanced account of its multidimensional nature. In this study, we rely on a conceptualization as proposed by Norris (2011, p. 23 and following) which we consider well-suited to the multidimensionality of democratic legitimation and combines the strengths of other attempts. The approach facilitates a decisive separation of political understandings, expectations, and behaviors relating to the normative ideas of democracy, institutional arrangements and their outcomes, therefore moving beyond the Eastonian dichotomy. It allows, for example, the exploration of why individuals may express strong commitment to democratic ideals while simultaneously exhibiting skepticism toward the political institutions that represent them and the politicians responsible for implementing policy. Most notably, the framework disaggregates political support into five analytically distinct components: (1) national identity, (2) regime principles, (3) regime performance, (4) political institutions, and (5) incumbent officeholders. This differentiation of support components is comparatively straightforward to operationalize, making the framework particularly well-suited for survey research.

It is important to emphasize that citizens’ notions of democratic legitimacy, that is, their perceptions of normative ideals, which may be established, unfulfilled, or subject to change in the prevailing political-constitutional order, and the processes of regime legitimation, both settled at the micro level, are closely intertwined and should not be viewed in isolation. They are inherently evaluative, reflect different degrees of support or disapproval toward political entities and principles, and operate at both conscious and unconscious levels. Nevertheless, many empirical studies implicitly conceptualize the multitude of surveyed indicators related to political support as a psychological orientation (e.g., Norris, 2011, p. 20), yet they often do so without explicitly considering the conceptual distinctions between various forms of values and attitudes. In our theoretical reflection, we aim to address these shortcomings by drawing on social psychology, which has extensively studied these concepts for decades (for a comprehensive overview, see Hewstone and Stroebe, 2020).

Values are conceptualized as normative ideals that are highly generalized and serve as fundamental guiding principles across broad areas of human life. Prominent theories posit that both the content of values and their structure exist across all cultures, as seen in the works of Inglehart (1971, 1997), Rokeach (1973), and Schwartz (1992). According to Rokeach (1973, p. 5), values are “enduring beliefs that a specific mode of conduct or end-state of existence is personally or socially preferable to an opposite or converse mode of conduct or end-state of existence.” Schwartz, for example, defines values as “desirable, transsituational goals, varying in importance, that serve as guiding principles in the life of a person or other social entity” (Schwartz, 1994, p. 21). Whereas Rokeach proposed a classification of value holders based on ranking scales, adopting a more person-centered perspective, Schwartz advocated the use of rating scales, which imply a linear relationship between adjacent values, thereby supporting a more variable-centered approach. Regardless, values tend to remain relatively stable over time as they are deeply internalized, allowing little room for questioning or change (Maio and Olson, 1998).

Attitudes are conceptualized as “a psychological tendency that is expressed by evaluating a particular entity with some degree of favor or disfavor” (Eagly and Chaiken, 1993, p. 1). In addition, the three-component model of attitudes (Rosenberg and Hovland, 1960; Eagly and Chaiken, 1993), also called the multi-component model (Zanna and Rempel, 1988), postulates that an attitude does not only contain feelings, which is the affective component, but also a cognitive dimension, which reflects attitude-relevant beliefs and finally a behavioral component. Notably, the behavioral attitude component assumes that people deduce their attitudes toward an object by recalling their past behavior relevant to the issue (Bem, 1972; Olsen, 1990). In principle, the behavioral dimension can also consist of intended behavior or hypothetical scenarios, which are empirically queried via item formulations. Therefore, it is also called the conative component of an attitude. Moreover, behavioral evaluations of an attitude object can be part of this dimension. Heyder et al. (2022, p. 4) suggested that, unlike values, attitudes might be classified as either generalized, which can be summarized into ideological concepts (Jost et al., 2008), or specific, directed at concrete attitude objects (Prislin and Ouellette, 1996).

Furthermore, with respect to the central aspect of evaluation, there are two basic views on the structure of attitudes. The one-dimensional and the bi-dimensional perspective (Maio and Haddock, 2015). The first is a view that all attitudes can be represented on a single bipolar evaluative dimension that extends from a maximum negative endpoint to a maximum positive endpoint and has a neutral center. The second postulates that two unipolar dimensions are necessary (positivity and negativity) to depict attitudes exhaustively. In most cases, Likert scales (Likert, 1932) are used to measure the intensity and direction of attitudes, following the first conception mentioned above.

Compared to attitudes, values serve as fundamental principles that stipulate what is considered right or wrong, whereas attitudes are more concrete to particular attitude objects and tend to be more flexible and context-dependent. In other words, values function as moral or ethical prescriptions that shape behavior more rigidly than attitudes (Maio et al., 2003). Accordingly, most researchers expect that changes at higher levels of abstraction (values) have a stronger influence on lower levels of abstraction, such as generalized and specific attitudes. For instance, Rokeach (1973) hypothesizes that a relatively small set of values can influence a larger set of attitudes.

Building on our review of the relevant literature, we propose that common measurements of political support, drawing on Norris’s support continuum, can be situated at the intersection of two key analytical axes: (1) the processes of democratic legitimation at the micro level, which necessarily refer to the normative cornerstones of a political regime that claims legitimacy for itself, and (2) the underlying psychological construct that characterizes the nature of the measurement object. Specifically, we differentiate whether support components are conceptualized as social perceptions of group membership at a high level of abstraction (social identities), as idealistic guiding principles (values), or as cognitive, emotional, or behavioral evaluations of an attitude object (attitudes).

Following the value–attitude–behavior (VAB) model proposed by Kahle (1980, 1983, see also Homer and Kahle, 1988), we argue that social identities provide individuals with a foundational sense of political belonging, shaping the interpretation of values and influencing the formation of attitudes. They function as a relatively stable lens through which political meaning is filtered and expressed. Values, in turn, represent deeply internalized and enduring orientations that inform more specific and situationally responsive attitudes, which ultimately guide (political) behavior. As in Norris’s approach, our framework does not posit a strict unidimensional continuum reducible to a single latent structure. Rather, the components are analytically and empirically distinct—yet interrelated—and can be broadly ordered from abstract and stable (identities and values) to more concrete and evaluative (attitudes), thereby reflecting both the temporal depth and hierarchical organization of political support:

(1) National belonging (Norris, 2011, p. 25) may be defined as a form of social identity (Tajfel and Turner, 1979, 1986) and arguably represents the most uncontested component of political support, grounded in a coherent theoretical and a well-established empirical tradition. Often operationalized through indicators such as national attachment, pride, or patriotism, it represents not only a form of political allegiance but also an expression of in-group favoritism, and, in its more intensified form, national idealization (i.e., nationalism), as conceptualized in social identity theory (Blank and Schmidt, 2003; Roccas and Berlin, 2016). We argue that Norris’s conception aligns with these assumptions, emphasizing that group-level characteristics help individuals to categorize and structure their social environment. As a result, widely used operationalizations of national belonging capture not only positive or emotional attachment but may also imply elements of outgroup devaluation. At the national level, this component primarily reflects citizens’ social identity in relation to the perceived dēmos—that is, the political community or population regarded as entitled to legitimate authority within the regime.

(2) Support for regime principles can be understood as values that reflect individual guiding orientations within a given political-constitutional context (Norris, 2011, p. 26). However, this component remains conceptually ambiguous, as democracy is defined in multiple and often contested ways. Diverging views on which features are considered essential contribute to ongoing debates about legitimacy crises in established democracies (van Ham et al., 2017). Empirically, approaches vary: Some adopt minimalist indicators, such as abstract support for the idea of democracy, while others employ broader operationalizations that incorporate constitutional or institutional features and perceived gaps between democratic ideals and reality. Substantively, support for democratic principles can be distinguished by the values they prioritize. For instance, liberal democratic values emphasize the rule of law and the protection of individual rights, egalitarian values stress equality, pluralist values reflect commitments to societal diversity, and populist values highlight the need for increased citizen participation.

(3) Evaluations of regime performance refer to citizens’ assessments of the perceived quality of democratic governance (Norris, 2011, p. 28) and encompasses affective, cognitive, and behavioral attitudes. They reflect public perceptions of how effectively particular democratic procedures and institutions deliver fair representation, accountability, and policy responsiveness. Empirical research often relies on the widely used satisfaction with democracy (SWD) item as a proxy for generalized affective attitudes toward the prevailing regime. However, researchers have raised concerns about the conceptual ambiguity of the SWD measure, arguing that it conflates multiple dimensions of political evaluation (Poses and Revilla, 2022). In contrast, more specific survey items frequently focus on the regime’s performance in safeguarding rights, ensuring accountability, promoting economic welfare, or addressing societal needs, thus primarily capturing cognitive evaluations. Scholars increasingly emphasize the need for refined and valid measurement approaches to advance the empirical assessment of regime performance (Singh and Mayne, 2023).

(4) Confidence in regime institutions and approval of political authorities reflect a combination of attitudes and social cognitions shaped by (collective) experiences, historical trajectories, and sociopolitical contexts. These attitudes emerge through processes of knowledge formation, activation, and diffusion, and vary depending on the level of abstraction or specificity of the evaluated object. Confidence in regime institutions typically refers to the legislative, executive, and judicial branches, as well as other core state institutions (Norris, 2011, p. 29). Conceptually, it remains contested whether this dimension reflects an affective attitude component, akin to interpersonal or social trust (Putnam, 2000), or whether it constitutes a more abstract form of systemic confidence (Luhman, 1989). As a result, empirical findings remain inconclusive as to whether political trust primarily signals diffuse or specific support, and whether it is essential for regime stability (e.g., Westle, 1999; Levi and Stoker, 2000; Denters et al., 2007; Harteveld et al., 2013). Approval of incumbent officeholders, by contrast, relates to citizens’ attitudes toward specific political leaders, parties, and public officials, including those occupying roles within the executive and legislative branches (Norris, 2011, p. 30). Yet debates persist regarding whether such approval is driven by short-term performance assessments, ideological proximity, or deeper trust in institutional authority. Moreover, it remains unclear to what extent citizens distinguish between political offices as formal institutions and the individuals who occupy them. Accordingly, common survey measurements tap into perceptions of trustworthiness, credibility, and accountability toward both institutional and personal representations of state authority.

Unsurprisingly, the body of empirical literature that addresses support for regime principles cross-culturally—at least from our perspective—is vast and multifaceted, highlighting both the theoretical and methodological challenges within this research domain. On the one hand, a remarkable diversity can be recognized in the use of terminology (see also Osterberg-Kaufmann et al., 2020, p. 306). For instance, while Kirsch and Welzel (2019) employ measurements fielded in the World Values Survey (WVS) to define liberal and authoritarian notions of democracy, Chapman et al. (2024) use the same scale to assess a typology of democratic conceptualization.

On the other hand, however, a major shortcoming is the lack of analytical sensitivity to cultural and national differences when comparing permutations of regime support. Examples include studies applying Mokken scale analysis with ESS data to examine the structure of Europeans’ views of democracy (Kriesi et al., 2016; Kriesi, 2025), single item responses as concepts of democracy with WVS data (Zagrebina, 2020), or non-compensatory composite scores to measure solid democratic support with Americas Barometer data (Moncagatta et al., 2023) and citizens’ democratic knowledge with WVS data (Wegscheider and Stark, 2020). Other studies, in contrast, calculate mean indices for Europeans’ expectations of democracy with ESS data (Heyne, 2018) or additive indices to capture core dimensions of democracy with EVS data (Seyd, 2020).

Yet relatively few studies explicitly stress systematic biases to enhance a broader comprehension of the underlying measurements among diverse cultural contexts. For instance, multi-group confirmatory factor analysis (MGCFA), Bayesian factor models, and alignment have been used to assess the comparability of liberal and authoritarian notions of democracy with WVS data (Sokolov, 2021; for Germany, see Jacobsen and Fuchs, 2020) and the evaluation of democracy with ESS data (Quaranta, 2018). Additionally, latent class analysis (LCA) has been applied to explore patterns that democracy takes in the minds of citizens with WVS data (Davis et al., 2020) or democratic ideals held by citizens with ESS data (Oser and Hooghe, 2018).

While this overview is neither exhaustive nor representative, it provides insight into survey-based approaches to assessing support for regime principles. Apparently, empirical research in this field reveals methodological diversity and is characterized by competing terminologies, insufficient conceptual differentiation of the constructs under study, and measurement strategies ranging from rudimentary to complex. Scholars predominantly adopt either an emic approach, which emphasizes country-specific explanations for regime support, or an etic approach, which seeks to develop universal theories incorporating cultural dimensions (Miller-Loessi and Parker, 2006). However, considering the state of the literature, fundamental ambiguities remain regarding the often-neglected questions (a) of what precisely constitutes the underlying psychological construct (b) and whether related measurements employing multiple indicators should be conceptualized as dimensions or typologies.

The “Europeans’ understandings and evaluations of democracy” module, included in the ESS round 10 and initially introduced in round 6 (see Ferrín and Kriesi, 2016), offers a valuable opportunity to study democratic values in times of turmoil. Surveys were conducted between September 2020 and September 2022 in 31 European countries. The national samples are considered representative for all persons aged 15 and over residing in private households. In round 10, nine countries transitioned from face-to-face interviews (CAPI) to self-completion modes (CAWI and PAPI) due to the impact of the COVID-19 pandemic, which also extended the fieldwork period (see Comanoru and Fitzgerald, 2025). Response rates varied significantly by survey mode, ranging from 20.9% in the United Kingdom to 72.8% in the Czech Republic. In our analysis, we mitigate potential biases from mode effects by restricting the sample to the 22 countries that administered the surveys as face-to-face interviews (ESS round 10, 2023a). Comprehensive details on sampling, sample sizes, questionnaires, and data access are available in the ESS Data Portal (ESS round 10, 2023b).

The module includes the CMoD scale, designed to measure democratic views by asking respondents to indicate the importance of a carefully selected set of elements drawn from theoretical models of democracy (Kriesi et al., 2016, p. 67). The conceptualization proposes five dimensions referring to the (1) liberal model which emphasizes procedural aspects, including minority rights, the rule of law, competitive elections, government accountability, and press freedom. The (2) social-democratic model expands on this by stressing income equality and protection from poverty. The (3) direct model focuses on referendums and direct citizen participation over representation. In contrast, the (4) populist model challenges the liberal approach by promoting an anti-elitist, anti-pluralist perspective, asserting that the people’s will should override constitutional safeguards, while the (5) multilevel model focuses on national sovereignty in decision-making processes (see Table 1).

According to the underlying rationale of the CMoD, democratic principles and values would be assumed to be internalized or at least known and regarded as important or not by the governed (Kriesi, 2013). Consequently, support for a given regime and its realized or absent features would depend on the perceived (in)congruence between the status quo and individual notions of democracy. In our point of view, a key aspect here is the term importance, which conceptually as well as semantically follows the abovementioned classical works on the measurement of value orientations. Emphasis on individually perceived importance in combination with certain aspects of democratic rule suggests that the scale aims to capture enduring, deeply held guiding principles that transcend specific situations, reinforcing its alignment with the measurement of democratic values. While ESS items begin with a personal pronoun (“How important do you think…”), it should be noted that they also address democratic ideals at a generalized level (“…for democracy in general”). We argue, however, that the items unequivocally evoke individual value preferences, as it prompts respondents to reflect on how democracy should ideally be composed, based on their personal convictions.

Nevertheless, the theoretical foundations of the CMoD scale remain ambiguous concerning whether it is intended to capture a priori defined, more or less correlated value dimensions (variable-centered) or profiles, that is, relatively stable group patterns of individuals sharing similar configurations of democratic values (person-centered). Both methodological approaches would be plausible within a pluralist framework of democracy theory, given the diversity of political systems and citizens’ varying notions of an ideal regime. To illustrate this briefly, liberal and social-democratic elements may jointly form the core of democracy in some cases, while those of the social-democratic and populist model may melt together in others. Moreover, citizens may desire absent regime features like press freedom, while longstanding regime principles, such as minority rights, receive less attention. Conversely, where press freedom is widely seen as granted, importance may lie in unfulfilled or threatened minority rights as a crucial regime cornerstone. Consequently, it is uncertain whether latent value factors or profiles should be regarded as country-specific phenomena related to political support (emic perspective) or as cross-culturally equivalent and thus comparable (etic perspective).

Based on the premise that the CMoD scale is designed to measure democratic values across different sociopolitical and cultural contexts, yet both person- and variable-centered approaches may be suitable for exploring latent structures, our empirical analysis adopts an exploratory and circular approach, as illustrated in Figure 1. The guiding question, succinctly summarized by Horn and McArdle (1992, p. 117), is “whether under different conditions of observing and studying phenomena, measurement operations yield measures of the same attribute.” Since the principal investigators neither explicitly specify an appropriate analytical approach nor clarify whether the instrument and its underlying indicators should meet cross-country equivalence, eight measurement hypotheses were formulated and addressed step by step, each accounting for different assumptions about the occurrence and comparability of value factors and profiles (see Table 2). They reflect recommendations derived from the current literature on best practices for analyzing survey data with reflective indicators (Davidov et al., 2018; Leitgöb et al., 2023) and offer a systematic framework for assessing whether the patterns identified are consistent and comparable. Strictly speaking, only when the assumptions underlying the analysis are met, the observed differences can be interpreted as genuine variations in individuals’ support for regime principles, allowing for meaningful comparisons that extend beyond the investigation of latent structures emerging within each individual country.

Our analysis begins with an exploration of response distributions using the resquin package (Kraemer et al., 2024) in R Studio (RStudio Team, 2020), as we are interested not just in individual item performance but in overall response behavior across the full scale. This preliminary step helps assess whether latent variable models capture substantive constructs rather than being distorted by systematic response styles. Consequently, before estimating global results, we first investigate the underlying latent structures within countries, addressing the first two measurement hypotheses (H1 and H2). To this end, we conduct exploratory factor analysis (EFA) using lavaan (Rosseel, 2012) to determine the number of meaningful latent dimensions and their associated indicators. To improve interpretability, Oblimin (oblique) rotation is applied, allowing for correlated factors. Furthermore, we conduct latent profile analysis (LPA) in Mplus 8.4 (Muthén and Muthén, 2017) to identify response-based profiles at the individual level.

To compare the latent structures discovered cross-culturally, we employ multi-group confirmatory factor analysis (MGCFA) in Mplus to assess factorial measurement invariance (MI; H3–H5). Accordingly, we first investigate whether the empirically derived factor structure, which appears reasonable from our perspective, holds across countries (configural MI). Second, we test whether factor loadings could be constrained as equal (metric MI), a prerequisite for comparing structural associations. Third, we assess whether scalar measurement invariance could be secured, which would allow for meaningful comparisons of latent means.

Although MGCFA is the predominant approach in testing exact measurement invariance, previous research shows that survey data often do not support strict model assumptions across many groups (Marsh et al., 2018). Therefore, supplementing our analysis, we apply alignment optimization in Mplus (Asparouhov and Muthén, 2014), an approximate and more flexible approach that allows up to 25 percent of model parameters to be non-invariant while still yielding trustworthy results. Additionally, we employ Bayesian approximate measurement invariance (BAMI) in Mplus (Muthén and Asparouhov, 2012), which allows small parameter variances between groups, thereby typically improving the cross-country comparability of latent means.

Moreover, we apply multi-group latent profile analysis (MGLPA) as a special application of mixture modeling with Mplus to assess the comparability of the initially discovered profile solution (Van Lissa et al., 2023). Unlike MGCFA, this approach assumes a categorical latent variable that represents the distribution of individuals sharing similar (dis-)approval patterns over all indicators of the CMoD. Here, measurement invariance relates to assumptions (H6–H8) on the occurrence of profiles across countries (heterogeneity), the similarity of item relationships within profiles (partial homogeneity), and the comparability of average item values across latent profiles (structural homogeneity).

All analyses using Mplus were retrieved with the MplusAutomation package in R Studio (Hallquist et al., 2024). For data manipulation and visualization, aimed at facilitating the interpretation of complex statistical analyses, we used the packages tidyverse (Wickham et al., 2019), tidyLPA (Rosenberg et al., 2018), here (Müller and Bryan, 2020), glue (Hester and Bryan, 2024), sf (Pebesma and Bivand, 2023), gt (Iannone et al., 2025), and a shapefile of European countries (Sevdari and Marmullaku, 2023).

Full information maximum likelihood (FIML) was used in all exploratory and confirmatory models as the estimation procedure to account for item non-response. In all group models, Belgium (selected alphabetically first) serves as the reference group to ensure result comparability. For factor models estimated using maximum likelihood (ML), the following cut-off criteria are applied to assess acceptable model fit: a comparative fit index (CFI) above .90, a Tucker-Lewis index (TLI) above .90, a root mean square error of approximation (RMSEA) below .06, and a standardized root mean square residual (SRMR) below .08 (Hu and Bentler, 1999). To evaluate nested models, we followed the recommendations by Chen (2007), considering ΔCFI/TLI below .01, ΔRMSEA below .015, and ΔSRMR below .03 as indicators of measurement invariance. In the Bayesian analyses, we test multiple prior values (.05–.0001) and visualize outcomes to assess latent mean stability (Arts et al., 2021). Prior selection can affect model fit, particularly the posterior predictive p-value (PPP), which may be less reliable with large samples (Asparouhov and Muthén, 2020). As values above .05 generally indicate acceptable fit, we also report Bayesian fit indices (BRMSEA, BCFI, BTLI) using similar cut-offs values. Given the extensive data with 22 country samples, the Gelman-Rubin convergence criterion was relaxed to .10 instead of .05. For mixture models focusing on latent profiles, model evaluation is commonly based on a combination of fit indices (Arts et al., 2021) and interpretability of the classes and their boundaries. We rely on the Bayesian information criterion (BIC), the Akaike information criterion (AIC), sample-size adjusted BIC (SABIC), entropy, and the Lo–Mendell–Rubin test (LMR) to assess model fit.

Figure 2 presents density and box plots, offering a comprehensive overview of distributional response indicators, including the mean, median, standard deviation across multiple items per respondent, and the Mahalanobis distance. An average standard deviation of 1.38 across all items suggests that individual responses typically vary by less than one and a half points from the overall mean. Respondents with a Mahalanobis distance above 5 exhibit highly divergent response patterns, though such cases are rare, as indicated by the third quartile remaining below this threshold. These outliers may signal potential data quality concerns. The mean response of 8.23 across all indicators of the CMoD suggests a distribution skewed toward high approval. With 50% of responses falling below approximately 8.6, half of the respondents provide values up to 9 for at least half of their answers. Thus, central tendency measures indicate a clustering of responses around 8 and 9, reflecting respondents’ inclination toward strong agreement.

Response style indicators, including midpoint response style (MRS), acquiescence response style (ARS), and extreme response style (ERS), reinforce this pattern, as shown in Figure 3. On average, respondents’ rate 10.5 out of 12 items as highly important (ARS). Additionally, respondents select extreme responses (ERS) for an average of about 5 items, typically rating them as extremely important. Notably, half of the sample did not choose a middle option. While the CMoD scale is unipolar, these tendencies highlight potential problems for model estimation: Many respondents assign the highest importance to most indicators. However, when examining straightlining—providing identical answers to items in a battery of questions using the same scale—we find that only 7.98% of respondents exhibit this behavior, suggesting that the scale does not primarily encourage careless responses.

To explore the latent structures and assess both cross-cultural consistency and deviation, we conducted country-wise EFA, testing factor solutions ranging from one up to five factors, what aligns with the five models of democracy that the CMoD is designed to capture theoretically. Based on eigenvalues (Kaiser-Guttman criterion) and scree plot analysis (see Figure 4), a common three-factor solution was uncovered in most countries, explaining a substantial share of the variance. However, this dimensional structure did not hold in Bulgaria, Greece, Croatia, Hungary, Ireland, Italy, Lithuania, Montenegro, North Macedonia, and Slovakia, where, at first glance, a two-factor solution appeared to provide a better fit to the data.

Subsequently, a global EFA employing the three- and two-factor solution was conducted on pooled data to assess its cross-country validity. As robustness check, we repeated the analysis on two separate subsets: one comprising only countries where the three-factor structure did not emerge and the other including all remaining countries. Additionally, we tested a two-factor solution for both groups to further validate the findings. The suitability of the number of factors extracted was supported by a significant Bartlett’s test of sphericity across all six models. A comparison of the three-factor and two-factor solutions—regardless of whether the Kaiser-Guttman criterion was applied—consistently showed superior model fit for the three-factor solution (see Appendix A). Regarding the first measurement hypothesis, H1, a three-dimensional structure appears empirically viable across all countries, albeit with certain limitations that will be addressed later.

However, contrary to the CMoD’s theoretical rationale, no clear empirical distinction emerges between the direct-democratic and populist models. Instead, they form a unified dimension (see Figure 5), with strong factor loadings (>.70) for unrestricted majoritarian rule and elite criticism, a solid loading (.51) for national political sovereignty, and a modest but acceptable loading (.45) for direct democracy. This pattern persists even in countries where the EFA did not support a three-factor solution. Notably, despite Norway’s non-EU status, the indicator on subsidiarity still loads satisfactorily. Since this dimension encapsulates key aspects of populism as a conception of political rule—namely, the belief that politics should directly reflect the unified will of the “true people,” often in opposition to elites and perceived institutional constraints—we label it populist democratic values.

Both indicators addressing material equality (.91) and redistribution (.67) in democracy exhibit satisfactory factor loadings on a factor we label social-democratic values, following Ferrín and Kriesi (2016). Similarly, items concerning the rule of law (.56), minority rights (.56), free media (.69), party pluralism (.67), and fair elections (.82) load convincingly onto a single factor representing liberal democratic values, elements that many theorists consider as minimal criteria for liberal democracy. Vertical government accountability (.30) falls below the threshold for sufficient validity and exhibits substantial cross-loadings in the pooled-data analysis, suggesting that it might represent a regime feature associated with support for democracy more broadly. Country-level analyses revealed outliers that affected the average validity coefficients; however, the indicator was retained in the measurement model due to its theoretical relevance and its satisfactory associations with the liberal democratic values factor in most countries.

To apply the person-centered approach empirically, we conducted an LPA with model solutions ranging from 1 to 11 classes, reflecting the total number of indicators in the CMoD. The results indicate that model fit improved with an increasing number of profiles, as shown by decreasing sample-size adjusted BIC values (see Figure 6), particularly between the one- and two-profile solutions. While BIC and AIC values continued to decline for higher-profile solutions, entropy values generally dropped between the second and third profile, suggesting greater classification accuracy for the two-profile solution. Exceptions include Hungary, Iceland, Italy, Portugal, Slovenia, and Finland, where the suitable number of profiles is varying to some extent (see Appendix B). Additionally, the substantive meaning of latent profiles was a key factor in model selection. Balancing statistical fit indices with theoretical interpretability, we determined that a two-profile solution best fits the data.

Figure 7 illustrates the latent profile distributions across the indicator scales. One class predominantly consists of respondents who consistently selected high importance values (i.e., 8, 9, and 10), while the other comprises individuals who avoid extreme response categories. This pattern is moderated for items with highly skewed distributions—such as fair elections and rule of law—where agreement is relatively high across both classes. Given the potential influence of measurement artifacts, particularly due to skewness because of extreme acquiescence, we refrain from interpreting this as indication of distinct types of regime supporters (e.g., strong vs. weak liberal democrats; cf. Kriesi et al., 2016; Kriesi, 2025). Rather, the findings suggest that respondents tend to evaluate all regime features either as uniformly important (acquiescence profile) or not (nuanced profile), without meaningful patterns of value-holder groups based on the proposed models of democracy. Still, in line with measurement hypothesis H2, the results indicate a consistent number of latent profiles across countries.

Following the establishment of the latent structures, we proceeded with confirmatory analyses. Model fit indices for the MGCFA are reported in Table 3 (see Appendix C for country-level CFA results). The configural model indicates a moderate fit, with a CFI exceeding .92 and TLI above .90, although RMSEA falls slightly below the cut-off. When comparing the configural and metric model, ΔSRMR surpasses the .015 threshold, and ΔCFI falls just below the .01 cut-off value. In contrast, ΔTLI and ΔRMSEA remain within acceptable limits. Given that CFI and TLI both reach .90, the metric model is acceptable under lenient fit criteria, whereas the scalar model is clearly rejected.

To examine the impact of different measurement assumptions on model estimation and data fit, we plotted the factor loadings for each item across all countries (see Figure 8). Substantial variation in loadings across models with increasing constraints may indicate problems of model misspecification. For the populist democratic values dimension, the items referring to majoritarian rule and elite criticism exhibit strong and consistent factor loadings across countries, with only minor deviations. In contrast, the items representing direct citizen participation and, in particular, subsidiarity demonstrate weaker and less stable measurement properties, as indicated by low reliability coefficients. The social-democratic values factor shows satisfactory and stable loading patterns across all countries. For the liberal democratic values dimension, the picture is more complex. As noted in the exploratory analysis, the indicators for vertical accountability and party competition reveal insufficient cross-cultural measurement quality. Furthermore, the item on minority rights falls below acceptable thresholds in Bulgaria and Czech Republic. By contrast, the remaining items exhibit satisfactory factor loadings. Nonetheless, according to these findings, imposing exact equality constraints across all groups may be overly rigid.

Given that even minor cross-country differences can lead to the rejection of substantively meaningful models, we extended our analysis using approximate measurement invariance (MI) approaches, which relax strict equality constraints while preserving the comparability of latent structures. Table 4 presents the results of alignment, distinguishing between invariant and non-invariant parameters. Of the 264 estimated factor loadings in the three-dimensional model, 78 were identified as non-invariant, with substantial variation across indicators. For example, only two loadings were non-invariant for the will of the people item, while up to 12 parameters (54.5%) were non-invariant for the referendum item. A higher degree of non-invariance was observed for the intercepts, with 145 identified parameters. In total, 223 out of 528 parameters (42.2%) were found to be non-invariant—substantially exceeding the 25% threshold proposed by Asparouhov and Muthén (2014).

Interestingly, only one of the tested Bayesian approximate MI models successfully converged with a PPP of zero, which typically indicates poor model fit. However, given the large number of country groups and the small prior variances applied, these results must be interpreted with caution (Asparouhov and Muthén, 2019; Arts et al., 2021). At a prior variance of .001, the BCFI with .89, BTLI with .88, and BRMSEA with .09 fail to meet the thresholds for acceptable model fit. Overall, the findings suggest that even under relaxed assumptions of parameter equality, the data do not support scalar invariance.

Figure 9 presents the country-wise ranking of estimated latent means derived from models that would support robust cross-national comparisons, namely, the scalar MI model, alignment, and the BAMI model, thereby illustrating the substantive implications of lacking scalar invariance. In this ranking, a value of 1 corresponds to the highest estimated latent mean, while 22 indicates the lowest. Crucially, the comparison reveals notable fluctuations in latent means, as reflected in shifting country rankings across different modeling approaches. For the populist value dimension, only France and Montenegro retain stable positions across all models. In the social-democratic dimension, 10 countries consistently maintain their relative rankings, suggesting higher measurement stability. Regarding the liberal value dimension, Norway, Slovenia, France, Great Britain, and Montenegro show no change in ranking across models. Notably, across all value dimensions and modeling strategies, no clear or consistent cultural or regional clustering emerges in terms of high or low ranks in democratic values.

Regarding measurement hypotheses H3, H4, and H5, the findings indicate that the data support meaningful dimensions of democratic values that hold cross-nationally (H3). The relationship and relevance of certain indicators vary somewhat, but metric invariance can be achieved (H4). However, comparisons of latent means remain problematic due to unsecured scalar invariance, regardless of whether strict or more flexible approaches are applied (H5).

To conclude our analysis, Table 5 presents the model results from the MGLPA, focusing on the cross-national comparability of the two identified latent profiles. The log-likelihood difference tests indicate that more restrictive models are not supported by the data. Only the heterogeneous model shows an acceptable fit, while the full homogeneous model failed to converge. Figure 10 illustrates profile membership across Europe on a map, alongside the estimated item means for each profile. As demonstrated in the previous analyses, most indicators show pronounced skewness toward strong agreement, particularly those related to fair elections and the rule of law. Accordingly, a key distinction between the acquiescence and nuanced profiles lies in their overall level of agreement across the scale. The observed response patterns may therefore reflect a widespread consensus on the importance of all regime features, rather than substantively distinct value-holder profiles, for instance, individuals who favor social-democratic over populist democratic values. Although our analysis excludes several European states whose inclusion would allow for a more comprehensive assessment, the profiles do not reveal any plausible clustering, for instance, along cultural traditions or regime similarities between countries.

In summary, with respect to measurement hypotheses H6, H7, and H8, the two-profile solution is supported across countries (H6). However, profile membership probabilities vary substantially between countries (H7), and the estimation problems of the homogeneous model (H8) indicates that profile-specific item means cannot be meaningfully compared across national contexts. Table 6 provides a brief tabular overview of all measurement hypotheses, corresponding results, and their respective implications.