Consensus and fragmentation in affective culture: A Bayesian analysis of cultural competence and item difficulty

ABSTRACT Affect Control Theory (ACT) posits that the micro-foundation of the social order lies in affective meanings consensually shared within a culture. This study challenges the assumption of consensus in affective culture by integrating ACT with Cultural Consensus Theory and sociological perspectives, offering a view of affective culture as both consensual and fragmented. Using Bayesian methods and data from surveys conducted in France and Germany, the analysis reveals that despite broad consensus, systematic patterns of fragmentation emerge, particularly in the evaluation dimension, which interacts with potency and activity. Moreover, the study shows that cultural competence varies across social positions, reflecting broader social dynamics. These findings advance ACT research methodologically and contribute to understanding how cultural consensus and fragmentation interact in the affective culture. 1. Introduction Often slipping under the radar of common perception, the seamless coordination of individuals within society is a baffling phenomenon and one of the fundamental topics in social science (Lawler et al., Citation2015; Rogers, Citation2020). Affect Control Theory (ACT; Heise, Citation2007), rooted in symbolic interactionism (Berger & Luckmann, Citation1991), posits that the micro-foundation of this dynamically emergent social order lies in the affective meanings consensually shared within a language culture (Heise et al., Citation2015). These meanings, empirically captured by semantic-differential ratings along the dimensions of Evaluation, Potency, and Activity (EPA; Osgood et al., Citation1957, Citation1975), encapsulate the norms and values that guide social interactions through the mechanism of deflection avoidance (Heise, Citation2007). Members of a language culture rely on these shared affective meanings to form a web of reciprocal expectations, which – on the one hand – provide the socio-emotional foundation for coordination (Scholl, Citation2013), and – on the other hand – create prima facie social validity that legitimize status hierarchies (Ridgeway & Correll, Citation2006). Consequently, “it is cultural consensus [around affective meanings] rather than individual opinion that determines what is good or bad [Evaluation], powerful or powerless [Potency], active or passive [Activity]” (Heise et al., Citation2015, p. 183) within a given society. However, does the concept of consensus fully capture the complexity of affective culture, or should it be understood as a more fluid and context-dependent process that incorporates both shared norms and group or issue-based variations? While previous research has empirically demonstrated the assumption of consensus by identifying a dominant factor in the covariance matrix of respondents surveyed in semantic-differential studies (e.g., Ambrasat et al., Citation2014; Sewell & Heise, Citation2010), this study advances the existing literature by integrating recent developments in ACT, Cultural Consensus Theory (CCT; Romney et al., Citation1986; Romney et al., Citation1987), and the sociology of culture (Cerulo et al., Citation2021; DiMaggio, Citation1997) to provide a more nuanced understanding of the interplay between consensus and fragmentation at the aggregate level. Furthermore, at the individual level, it explores the relationship between cultural competence – specifically, whether individuals vary in their ability to produce consensual responses – and social position, and examines whether the degree to which social concepts are shared differs according to their connotation, thereby shedding light on the uneven terrain of cultural agreement within society. The analysis draws on six cultural surveys conducted in France and Germany, examining 2,278 social concepts along EPA dimensions using a representative sample of 4,175 participants. Crucially, the research employs the latest Bayesian developments in CCT (Anders & Batchelder, Citation2012, Citation2015; Anders et al., Citation2017) which offers a significant methodological advancement, accommodating multiple latent cultural subgroups, varying participant competencies, and item difficulties. By operationalizing uncertainty in consensus, this approach fits well with recent methodological advances in ACT research, moving from a frequentist to a Bayesian framework (Morgan et al., Citation2016; Schröder et al., Citation2016). 2. Theoretical background 2.1. Consensus and fragmentation in affective culture Over the past three decades, sociologists have increasingly challenged the traditional notion of culture as a coherent and unified construct. Instead, they advocate for a cognitively realistic view that recognizes culture as fragmented across different groups and contexts (DiMaggio, Citation1997; Cerulo et al., Citation2021). This shift in perspective highlights a key limitation within Affect Control Theory (ACT). By emphasizing the idea of consensus around affective meanings within cultures, ACT dismisses intrasocietal variance as merely idiosyncratic (Heise, Citation2007) and misses the opportunity to examine the conditions under which consensus and fragmentation arise within a given culture and across different social concepts. There are good theoretical and empirical reasons to not consider intrasocietal variance as mere noise, but rather as an indicator of cultural fragmentation (Ambrasat & von Scheve, Citation2022). At the aggregate level, fragmentation influences patterns of enculturation and cultural diffusion. On the one hand, affective culture is a form of nondeclarative culture which is gradually acquired by internalizing underlying schemas through repeated exposure (Lizardo, Citation2017). Accordingly, empirical research demonstrated that patterns of affective enculturation slightly but consistently vary across social positions, revealing that factors such as race (Rogers, Citation2019; Sewell & Heise, Citation2010), class (Ambrasat et al., Citation2014), the size of personal social networks (Thomas & Heise, Citation1995), and gender (Dametto, Vieira, Nöel, et al., Citation2024; Heise, Citation2007) all play a role in shaping affective meanings. On the other hand, changes in affective meanings, much like cultural changes in norms and values, seem to follow a complex contagion dynamic (Centola & Macy, Citation2007). Correspondingly, shifts in affective meanings that challenge mainstream perspectives emerge within tightly connected peripheral groups, where frequent and intense local interactions foster these changes, rather than around central groups (Centola et al., Citation2005). This pattern is consistent with findings in the ACT literature, which show that cultural variance is more likely in peripheral and closed subgroups than in central clusters (Hunt, Citation2012; Smith-Lovin & Douglass, Citation1992). At the individual level, fragmentation affects cultural competence which refers to an individual’s ability to provide the consensual, or most common, response within a given cultural context (Romney et al., Citation1986). An individual who is culturally competent in a particular affective culture is able to interpret and apply affective meanings in a way that aligns with the majority’s understanding, thereby acting in accordance with prevailing norms, values, and also stereotypes. As such, detecting high levels of competence within specific sociodemographic subgroups may suggest that they have a vested interest or perceived obligation to align with mainstream affective representations. This adherence might be driven by social expectations, professional roles, or the desire to maintain status differences within the broader cultural framework (Ridgeway & Correll, Citation2006). By contrast, when individuals are tendentially less able to provide consensual responses, it may reveal that they either intentionally resist or unintentionally diverge from the majority’s interpretation of social concepts. These divergences could reflect a subgroup’s willingness to change the dominant cultural narrative or a disconnection from mainstream values, potentially signaling sociodemographic areas of cultural fragmentation. Therefore, competence at the individual level and the degree of consensus on affective meanings at the aggregate level are strongly influenced by social position and the patterns and intensity of social interactions, and conversely can reveal much about the latter. In the following sections, I will briefly review the “cultural survey” approach and the Q methodology (Heise, Citation2010), the techniques used within ACT to measure affective meanings and assess consensus, with the aim of highlighting their inherent limitations of analyzing fragmentation. 2.2. Methodological limitations to the analysis of cultural fragmentation Affective meanings are typically measured through cultural surveys, which serve a different purpose than the population surveys often used in sociological research. While population surveys emphasize variability to uncover social controversies and make causal inferences, cultural surveys are aimed at uncovering consensus, treating variability as an idiosyncratic component that obscures shared cultural norms. The goal is to minimize this “noise” in order to reveal the underlying social content that reflects the dominant culture (Heise, Citation2007). Traditionally, cultural surveys have used convenience sampling, selecting respondents based on their presumed expertise in a particular cultural context. For example, students – who are being trained for high-status roles – have been considered suitable representatives of the dominant culture. In other cases, respondents have been selected to represent specific subcultures (Hunt, Citation2012; Sewell & Heise, Citation2010; Smith-Lovin & Douglass, Citation1992). More recently, cultural surveys have also been conducted using representative samples, increasing the validity of generalizations to the population (Ambrasat et al., Citation2014; Dametto et al., Citation2023; Quinn et al., Citation2023). ACT builds on Cultural Consensus Theory (CCT), a research approach in cognitive anthropology that seeks to identify normative cultural features unknown to the researcher but shared across a population (Romney et al., Citation1986). Drawing from CCT, ACT researchers use Q methodology – a principal component analysis based on the correlation between respondents – to determine whether a single dominant factor explains a substantial proportion of the variability in responses, providing empirical support for the consensus assumption (Heise, Citation2007). However, Q methodology rests on three key assumptions that limit its applicability (Romney et al., Citation1986). (1) Single cultural truth. The methodology presupposes the existence of no more than one cultural truth, rendering it unsuitable for capturing multiple coexisting cultural perspectives. (2) Informant independence. It presumes that respondents’ answers are independent and reflect only their cultural competence, while research has demonstrated that social position at least partially influences the subjective meaning-making process. (3) Homogeneous competence. The Q methodology assumes that all informants have the same level of competence in relation to the items being assessed, yet the variability introduced by social positions also affects the salience of particular identities and concepts (Stets & Burke, Citation2000) and the availability of mental representations and intuitive associations (Shepherd, Citation2011). Moreover, one consequence of these latter limitation, as noted by Heise, is that by giving equal weight to all social concepts included in the analysis, the Q methodology cannot identify subgroups of participants in cases where cultural fragmentation occurs only in relation to a small subset of concepts (Heise, Citation2007, p. 177). To overcome these limitations, recent advancements in CCT offer valuable tools for the ACT community, providing a more nuanced understanding of the fuzzy relationship between consensus and fragmentation within the affective culture. 2.3. Bayesian approach to consensus analysis Anders and Batchelder (Citation2012), (Batchelder and Anders (Citation2012) extended the consensus analysis methodology originally developed by Romney and colleagues (Romney et al., Citation1986; Romney et al., Citation1987; Batchelder William & Romney, Citation1986) by introducing two significant enhancements. First, their model can account for multiple response keys, which represent different cultural “truths” corresponding to various latent subcultures. This allows the model to evaluate a participant’s competence relative to each specific truth key, reflecting the cultural diversity within a population. Second, their model can accommodate items with varying levels of difficulty, thereby offering a more nuanced analysis of responses and measuring consensuality around social concepts. 2.3.1. Multiple truth keys and participants’ competence The Multi-Culture General Condorcet Model (MC-GCM) is designed to identify latent subgroups, i.e. subcultures, within a sample of respondents. Membership in a latent subgroup is determined by shared responses that align with a specific “truth key,” representing a consensually agreed-upon answer within that latent subgroup. The model relies on three probabilities. The informant’s probability of knowing the truth (hit rate), the probability of guessing the truth and thus being correct by chance (false alarm rate), and the cultural membership parameter, which indicates the probability of belonging to a particular subgroup based on shared responses. Hierarchical Bayesian inference techniques are used to update beliefs about these three probability vectors. Cultural membership is initialized as a Bernoulli distribution, while the probabilities of knowing and guessing the truth are represented as mildly informative beta distributions, allowing the model to update these estimates as it processes more information (Anders & Batchelder, Citation2012; Batchelder & Anders, Citation2012). To use the MC-GCM, researchers must first determine the number of cultures within the sample by analyzing a scree plot of eigenvalues derived from the respondents’ correlation matrix. Traditionally, CCT relies on a rule of thumb where the first component must be at least three times larger than the second to confirm the presence of a single culture (Weller, Citation2007). However, Anders and colleagues suggest using the “elbow” method, which defines the number of cultures as the point before the scree plot transitions from a steep drop to a more linear trend (Anders et al., Citation2017). Once the model hyperparameters are defined, researchers validate their assumptions by comparing the model’s simulated data to the patterns observed in the empirical data’s eigenvalues. If these patterns align, the model is considered valid (Anders et al., Citation2017). The MC-GCM provides two key outputs for each participant: the latent subgroup they belong to and their competence score, which reflects their ability to provide answers consistent with their latent subgroup’s cultural consensus, adjusted for item difficulty. The MC-GCM has several potential applications in ACT research. It is particularly valuable when there is reason to believe that multiple distinct latent cultures exist within the sample. For example, previous research has shown that consensus varies across the EPA dimensions, with less agreement typically found in the potency and activity dimensions compared to the evaluation dimension (Dametto et al., Citation2023; Heise, Citation2007). The MC-GCM also addresses limitations in traditional consensus analysis conducted with the Q methodology, which cannot detect latent subgroups, when the number of disputed social concepts is relatively small compared to the overall set of stimuli, or when members of subcultures moderate or exaggerate rather than radically oppose dominant sentiments (Heise, Citation2007). Finally, by assessing participants’ competence, the model can reveal sociodemographic patterns that help identify which cohorts best incarnate the affective culture, offering insights into potential lines of cultural fragmentation. 2.3.2. Heterogenous item difficulty The MC-GCM captures variability in responses by considering both multiple latent subgroups and heterogeneous item difficulty. In traditional models with homogeneous item difficulty, each item is assumed to have the same likelihood of being answered correctly. For example, in a typical nine-point EPA semantic differential scale, the probability of randomly answering an item correctly is 1/9. However, because participants do not respond randomly, the difficulty of items varies, and this variability is not constant. At the item level, the MC-GCM estimates difficulty by considering both the informants’ responses and the characteristics of the item itself. Items that receive consistent, consensual answers from informants of different competence levels are considered easier. On the other hand, items that elicit diverse responses from many informants are classified as more difficult. Importantly, this difficulty measure depends on the informant’s latent subgroup, meaning that an item’s difficulty can vary between latent subgroups (Anders & Batchelder, Citation2012; Anders et al., Citation2017). At the model level, the Variance Dispersion Index (VDI) is used to assess the assumption of heterogeneous item difficulty. The VDI is calculated as the sum of the within-item variance relative to the sample mean variance, i.e. it is the variance of the item variances. A high VDI indicates that there are significant differences in item difficulty – some items are more or less consensual than others – while a low VDI suggests that all items have similar difficulty levels (Anders & Batchelder, Citation2012; Anders et al., Citation2017). To test whether this assumption of heterogeneity holds, Anders and colleagues developed a method that compares the VDI of empirical data against the range of VDI values produced by the simulated data. If the empirical VDI falls within this range, the assumption is validated. However, if the VDI is too small, it suggests homogeneous item difficulty, while very high VDI values indicate difficulty in converging on a shared cultural truth within the population (Anders & Batchelder, Citation2012; Anders et al., Citation2017; Batchelder & Anders, Citation2012). While accounting for heterogeneous item difficulty generally improves the precision of results (Anders et al., Citation2017), this feature also has significant implications for ACT. For example, a difficult item might indicate that (1) a social concept is not widely known or is cognitively challenging, resulting in a lack of shared representation and leading participants to hold different affective meanings. Difficult items might also suggest (2) that a social concept is highly controversial or (3) that the concept holds specific meanings within a subculture. 3. Analysis strategy 3.1. Materials and rating procedures The current study is based on semantic-differential ratings of the 1,139 social concepts in French and German – a total of 2,278 items – contained in the “French-German EPA Dictionary,” a repository of affective meanings along the dimensions of Evaluation, Potency, and Activity (EPA) in the two languages (Dametto, Vieira, Blaison, et al., Citation2024). These social concepts are divided into three grammatical categories: 387 gendered social identities, 337 behaviors, and 415 settings. Each concept was rated in both French and German, ensuring semantic alignment through a back-translation method. A French native speaker translated the concepts from English to French, followed by a German native speaker translating them from French to German. Another German native speaker then translated them back to English. Any discrepancies in translation were resolved through team discussions to ensure accuracy. The selection of identities and behaviors was based on a cluster analysis of U.S. affective dictionaries (for identities and behaviors: Smith-Lovin et al., Citation2016; for settings: Smith-Lovin, Citation1987) to ensure an even representation of the affective space (for a similar approach cf.: Mostafavi et al., Citation2024). Participants rated one social concept at a time on the EPA dimensions, using a nine-point scale. The order of social concepts was randomized for each participant, and there was no option to skip items. A copy of the rating instruments is available in the Supplementary Information (SI-1). The contrasting anchors of the bipolar scales were already well established in German (Schneider, Citation1989; Schröder, Citation2011) and were identified in French through a factorial analysis by Dametto et al. (Citation2023). The units of analysis are the ratings provided by a total of 2,088 French and 2,087 German respondents who participated in six cultural surveys, covering the three categories of social concepts (identities, behaviors, and settings) in both countries. These surveys form the basis of the “French-German EPA Dictionary” (Dametto, Vieira, Blaison, et al., Citation2024). A balanced incomplete block design was used to randomly assign participants to different sets of stimuli. Each set was rated by a sample representative of the population in terms of age, gender, and region. Where available, U.S. ratings were used to affectively balance the stimuli presented to participants across sets. To ensure comparability, the stimuli sets rated by the German participants were identical to those rated by the French participants. A specific description of the samples for each survey is available in the supplementary information (SI-2). Overall, each of the 1,139 social concepts was rated by ca. 140 (70 French and 70 German) for a total of N = 4,175 participants. The sample size far exceeds the recommendations of Heise (Citation2007) for cultural surveys and those of Anders and Batchelder (Citation2012) for implementing the MC-GCM model. Additionally, participants provided sociodemographic information, including age, gender, region, education level, income, city size, and whether they were native speakers of the survey language. For the sociodemographic analysis of competence, only 3,819 participants who provided information on income level and city size were considered. 3.2. Applying the MC-GCM Given the high-dimensional space defined by each respondent’s ratings of the social concepts, the posterior distribution must be approximated by simulation methods rather than computed directly. To fit the model, Anders and colleagues provide the R package “CCTPack” (Anders, Citation2017; Anders et al., Citation2017), which relies on the open-source software JAGS (Plummer, Citation2003). This package uses Gibbs sampling, a type of Markov Chain Monte Carlo algorithm. Gibbs sampling works by iteratively sampling from the conditional distribution of each variable while keeping the other variables fixed. The results discussed below were generated using CCTPack. Each model was fitted by drawing 6,500 samples from the posterior distribution, with the first 2,000 samples discarded as “burn-in” to allow the model to stabilize. This process was repeated across three separate chains to ensure robustness. Given that the ratings were provided on a nine-point scale, the input data were treated as ordinal. The ratings of French and German participants were analyzed together. This was done for two reasons: on the one hand, previous research examining the affective meanings attributed to social identities suggests a considerable cultural consensus between French and Germans (Dametto et al., Citation2023). On the other hand, the aim of the analysis was to assess possible sources of cultural variation. It is therefore more interesting to consider language culture as a possible source of variation alongside sociodemographic factors, but within the same Franco-German cultural context, rather than dividing the dataset based on a priori assumptions about cultural cleavages between nations. For each category of social concepts (identities, behaviors, settings), a separate model was fitted to each stimulus set for each EPA dimension. This resulted in 30 (stimuli sets) ×3 (dimensions) = 90 models, each based on 140 participants rating approximately 38 items. All models considered item difficulty as heterogeneous, and the number of cultures was determined using scree plot analysis. All data and analysis scripts are openly available at https://osf.io/8974f/?view_only=01d9d99a1dc44386ae7dcc718e5d9f35. 4. Results 4.1. A priori and posterior analysis As a form of Bayesian inference, the MC-GCM model requires (I) prior knowledge of the number of cultures present in the dataset and (II) an assumption of heterogeneity or homogeneity in item difficulty. A posterior assessment of the model fit should then be performed. Since the data were collected using an unbalanced block design, the a priori and posterior analysis must be performed at the stimuli set level. Below, I present an exemplary case and refer the reader to the Supplementary Information for the complete a priori (SI-3) and posterior analyses (SI-4) that validate the model hyperparameters. The top panel of shows the scree plots of eigenvalues derived from the respondents’ correlation matrix for the first set of the behavior survey across the dimensions of evaluation (left), potency (center), and activity (right). According to the “elbow” method proposed by Anders et al. (Citation2017), the point before the scree plot transitions from a steep drop to a more linear trend defines the number of cultures a priori. This point is located at the first factor for the evaluation dimension and at the second factor for the potency and activity dimensions. The bottom panel shows the posterior checks, where the black line representing the empirical data is “overlaying or highly similar to the gray lines [produced by the simulated data]” (Anders et al., Citation2017). In this case, the simulated data slightly underestimates the second factor with respect to the activity dimension, but matches the patterns found with respect to the evaluation and potency dimensions. The second posterior analysis addresses the issue of heterogeneous item difficulty, as measured by the Variance Dispersion Index (VDI). Anders et al. (Citation2017) suggest that the VDI should fall between the 10th and 90th percentiles. The first set of the behavior survey shows VDI values of 24.8 for the evaluation dimension, 59.2 for the first latent subculture, and 47.6 for the second latent subculture on the potency dimension, and 59.6 and 78.4 for the first and second latent subcultures respectively on the activity dimension. This confirms the assumption of heterogeneous item difficulty. As documented in the Supplementary Information, despite some variations, the patterns related to the number of cultures and the assumption of heterogeneous difficulty presented above are consistent across all sets of concepts and dimensions. Approx. 3 out of 30 sets for the evaluation dimension, 6 out of 30 sets for the potency, and 9 out of 30 sets for the activity dimensions could have been eventually fitted to a different number of latent cultures. Similarly, the assumption of heterogeneity was not satisfied for 4 out of 30 sets for the evaluation dimension, and for 2 out of 60 latent cultures for the potency and activity dimensions. Therefore, to ensure comparability across stimuli set and within dimensions, the MC-GCM models for the evaluation dimension were all fitted to a single culture, while the models for the potency and activity dimensions were fitted to two latent subcultures.Footnote1 All models were set to have heterogeneous item difficulty. The results presented below are based on two datasets that aggregate the results of all 90 models. The first dataset contains information on the difficulty level and truth key for each item. As the potency and activity dimensions were modeled with two latent subgroups, the difficulty parameter and truth key for these dimensions are estimated twice. Easier, i.e. consensually shared, items have negative difficulty values, while more difficult, i.e. less consensually shared, items are indicated by positive difficulty values. The second dataset contains information at the participant level and reports the competence scores of each participant. For the potency and activity dimensions, participants’ competence is assessed in relation to the latent subculture to which they belong. Results are presented separately for each dimension. 4.2. Evaluation 4.2.1. Item difficulty By considering item difficulty as heterogeneous, we can identify items that are easier or harder to answer, i.e. items that are more or less consensually rated. shows the distribution of difficulty scores (center) and shows that they are normally distributed. On the left and right are concepts that are particularly easy (5th percentile) and difficult (95th percentile) respectively. The easiest concepts include many items related to professions and everyday interactions. On the other hand, almost all the difficult concepts seem to be related to negative concepts, and only very few cases can be attributed to semantic ambiguity.Footnote2 To systematically investigate how item difficulty varies in the affective space, a mixed-effects model with crossed random effects for stimuli set nested within the categories of social concepts was specified. The dependent variable, item difficulty, was regressed on the mean evaluation, potency, and activity ratings, as well as their interaction terms. presents the random effects, showing small variability across categories of social concepts and no variability across the stimuli set. The fixed effects () indicate that the more positive the concept, the easier it is to rate, and conversely, negative concepts are more difficult to rate (b = −.032, t(1,139) = −7.002, p < .001, 95% CI [−.04; −.02]). Concepts that are perceived as both positive and potent are rated more difficult (b = .018, t(1,139) = 4.291, p < .001, 95% CI [.01; .03]), while good and activeFootnote3 concepts are easier to rate (b = −.023, t(1,139) = −6.220, p < .001, 95% CI [−.03; −.02]). 4.2.2. Participants’ competence The second question is whether competence – i.e. the ability to produce consensual responses – varies as a function of social position. Competence scores are mostly normally distributed, but a few participants are clearly more able to provide consensual responses, skewing the distribution to the right (). The Supplementary Information (SI-5) provides additional graphs describing levels of competence by socio-demographic variables and helps to understand the regression model presented below. To examine whether social position affects the competence level, I computed a nested mixed-effects model in which participants’ competence was regressed on sociodemographic variables. Crossed random effects were defined with stimulus sets nested within the categories of social concepts. The results show no variability across categories of social concepts and some variability within the stimuli sets (). The fixed effects are reported in . Only coefficients significant at the p < .001 level are discussed. The results indicate a negative linear effect of age, with younger participants exhibiting greater competence (b = −.130, t(3,819) = −6.765, p < .001, 95% CI [−.17; −.09]). At the same time, a cubic effect of age is also significant (b = .057, t(3,819) = 3.436, p < .001, 95% CI [.02; .09]). All in all, the two coefficients combined reveals that – despite some oscillations – competence decreases with age (for a graphical visualization cf. SI-5). Moreover, male (b = .124, t(3,819) = 8.643, p < .001, 95% CI [.10; 0.15]) and German (b = .061, t(3,819) = 8.643, p < .001, 95% CI [.03; .09]) participants tend to provide more consensual answers. Furthermore, the positive linear effect of town size (b = .054, t(3,819) = 3.440, p < .001, 95% CI [.02; .08]) suggests that the larger the town in which the participant lives, the more competent he or she is. In contrast, years of education (b = −.015, t(3,819) = −5.557, p < .001, 95% CI [−.02; −.01]) have a negative effect on the tendency to provide consensual answers. 4.3. Potency 4.3.1. Latent subcultures and variations in truth keys In a single-culture context, such as in the case of evaluation, item difficulty captures differences in ratings as varying levels of consensuality around a social concept. For potency and activity, models were fitted to two latent subcultures. Thus, levels of competence are based on subgroups. However, cultural fragmentation around social concepts can still be assessed by examining which items lead to disagreement – indicated by large differences in the truth key – between the latent subgroups. (center) shows the distribution of differences in the truth key, which is clearly skewed to the right, suggesting that the two cultures identified in the dataset endorse relatively similar truth keys for many concepts, while largely disagreeing on only a few concepts. Concepts that lead to small (5th percentile) and large (95th percentile) differences between the two latent subcultures are plotted on the left and right, respectively. Again, a visual inspection of the 5th percentile suggests that – with a few exceptions (e.g. terrorist, orgy, escort club) – professional roles and everyday behavior lead to smaller differences in the truth key between the two latent subgroups. On the other hand, as with evaluation, the concepts that lead to the greatest differences between the two latent subcultures are mainly those with clearly negative connotations. To analytically address the question of which concepts lead to disagreement between the two latent subgroups, I computed a mixed-effects regression model with the EPA dimensions and their interaction terms as independent variables and the differences in the truth key between the two latent subcultures as the dependent variable. Random effects were defined for stimuli sets nested within categories of social concepts, indicating moderate variability at both levels. The residuals’ standard deviation is moderately high, suggesting some unexplained variability in the dependent variable (). The analysis of coefficients () shows that the more positive the concept, the smaller the difference between cultures (b = −.196, t(1,139) = −9.869, p < .001, 95% CI [−.23; −.16]). By contrast, active concepts result in larger differences (b = .230, t(1,139) = 6.699, p < .001, 95% CI [.16; .29]). Nonetheless, the interaction terms indicate that concepts perceived as good and potent lead to larger differences in truth keys (b = .199, t(1,139) = 10.593, p < .001, 95% CI [.16; .24]), while concepts perceived as good and active are judged more consensually between groups (b = −.163, t(1,139) = −10.044, p < .001, 95% CI [−.20; −.13]). 4.3.2. Participants’ competence Given the two-culture parameterization, competence must be assessed within the latent subgroup to which participants belong. As a first step, I computed 90 ordinal logistic regression models to investigate whether cultural membership is defined by social position, as captured by sociodemographic variables. The results (Supplementary Information – SI8) suggest that no sociodemographic variable consistently explains cultural membership across categories of social concepts and sets. Therefore, I focus the analysis on the competence levels assessed within each latent subculture. As in the case of evaluation, competence is right-skewed in both latent subcultures (). The Supplementary Information (SI-6) provides additional graphs demonstrating highly similar patterns of competence between groups across sociodemographic variables. A mixed-effects model was computed with latent subgroups nested within stimuli sets, which in turn were nested within categories of social concepts. As shown in , variability is higher within latent subgroups and relatively low between stimuli sets and social concepts. This makes sense considering the arbitrary labels (subgroup 1, subgroup 2) attributed to the latent subgroups. The results of the fixed effects analysis are presented in . The data indicates a negative linear effect of age (b = −.145, t(3,819) = −6.132, p < .001, 95% CI [−.19; −.10]). Additionally, men (b = .112, t(3,819) = 6.377, p < .001, 95% CI [.08; 0.15]) tend to produce more consensual answers. In contrast, German respondents provided less consensual answers (b = −.086, t(3,819) = −4.665, p < .001, 95% CI [−.12; −.05]). The number of years of education (b = −.022, t(3,819) = −6.325, p < .001, 95% CI [−.03; −.01]) has a negative effect. Those with less education tend to provide more consensual answers. Finally, individuals belonging to the middle income classes tend to provide less consensual responses, as evidenced by the quadratic positive effect describing a U-shaped effect (b = .066, t(3,819) = 3.689, p < .001, 95% CI [.03; .10]; for a graphical visualization cf. SI-6). 4.4. Activity 4.4.1. Latent subcultures and variations in truth keys As in the case of potency, I examined variations at item level by inspecting differences in the truth key. (center) shows that the distribution of difficulty scores is right-skewed, so that only a few concepts lead to large differences between the two latent subgroups. Other than in the case of evaluation and potency, no clear pattern seems to emerge when inspecting the concepts that lead to small (5th percentile) and large (95th percentile) differences between the two latent subcultures. To investigate how affective meanings relate to large differences in truth keys between the two latent subgroups, I computed a mixed-effects regression model with random effects for stimuli sets nested within categories of social concepts. reports the random effects, showing that despite some variability across sets, the intercepts are similar across categories of social concepts. The fixed effects analysis () suggests a similar pattern to that found in the evaluation and potency dimension, though the magnitude of coefficients is generally smaller. The more negative the concept, the larger the difference between latent subgroups (b = −.055, t(1,139) = −4.402, p < .001, 95% CI [−.08; −.03]). Furthermore, the evaluation dimension interacts with the potency dimension (b = .067, t(1,139) = 5.632, p < .001, 95% CI [.04; .09]). Good and potent concepts result in larger differences in truth key between latent subgroups. By contrast, evaluation and activity interact negatively (b = −.068, t(1,139) = −6.726, p < .001, 95% CI [.09; −.05]); thus, suggesting that positive and active concepts lead to greater differences. 4.4.2. Participants’ competence As with potency, I investigated whether membership in latent subcultures in activity ratings is determined by sociodemographic variables. Similarly, the coefficients of the 90 ordinal logistic regression models do not reveal a consistent pattern (Supplementary Information – SI-9), so I turned my attention to participants’ competence scores. Within the two latent subcultures, competence is once again right-skewed (). In the Supplementary Information (SI-7), additional graphs illustrate the patterns of participants’ competence across socio-demographic variables. To analytically assess participants’ competence within groups, I computed a mixed-effects model with nested structure (subgroups : categories of social concept : stimuli set; see ). As in the case of potency, variability is low within stimuli sets and social concepts and high within subgroups. Among the fixed effects (), age shows the same combination of a negative linear (b = −.148, t(3,819) = −4.338, p < .001, 95% CI [−.21; −.08]) and positive cubic (b = .104, t(3,819) = 3.519, p < .001, 95% CI [.04;.16]) effect on competence. Again, competence declines in the older age cohorts (see Supplementary Information SI-7). Men are also more competent than women (b = .135, t(3,819) = 5.319, p < .001, 95% CI [.08; .18]). Lastly, the higher the education degree, the less culturally competent are the participants (b = −.028, t(3,819) = −5.726, p < .001, 95% CI [−.04; −.02]). 4.5. Summary of results Overall, results prove to be very consistent across categories of social concepts and dimensions, especially considering that data were collected across different languages and surveys. Considering item difficulty, negative concepts are judged less consensually. Yet the potency dimension interacts with the evaluation dimension, making good and potent as well as bad and impotent social concepts more difficult to judge. The evaluation dimension interacts also with the activity dimension: good and active, and correspondingly, bad and calm concepts are judged more consensually. Considering participants’ competence, age and years of education have a negative effect on competence. In addition, women tend to give answers that are less in line with the broader cultural consensus. 5. Discussion Motivated by recent methodological developments within the Cultural Consensus Theory (Anders & Batchelder, Citation2012, Citation2017), Affect Control Theory (ACT), and the sociology of culture (DiMaggio, Citation1997; Cerulo et al., Citation2021), this study aims to investigate, at the individual level, whether competence – defined as the ability to produce consensual responses – varies according to social position and the specific concept under consideration. At the aggregate level, the study examines the interplay between consensus and fragmentation within affective culture, with the goal of identifying the conditions under which fragmentation arises and poses a challenge to the prevailing consensus around the social order. At the aggregate level, the analysis confirmed the single-culture assumption for the evaluation dimension, but not for the potency and activity dimensions. For the latter two, the MC-GCM identified two distinct latent subgroups. Three key findings are highlighted: (I) membership in these latent subgroups is not explained by the sociodemographic variables considered; (II) differences in truth keys between the latent subgroups on the potency and activity dimensions mirror the pattern of difficulty observed within the evaluation dimension; and (III) this pattern is stable across categories of social concept. The consistency of the pattern across dimensions and categories of social concepts suggests that item-level variance is not randomly distributed and cannot be dismissed as mere noise. While individuals within the French and German affective culture share consensual affective meanings around most social concepts, systematic patterns of fragmentation still emerge. But what are the social concepts that are less likely to be shared consensually? The analysis indicates that affective meanings associated with negative concepts are judged less consensually than those associated with positive concepts across all three EPA dimensions. This finding aligns with cognitive research demonstrating an asymmetry between positive and negative information. Positive information tends to be more alike than negative information (Alves et al., Citation2017), is easier to classify, learn (Unkelbach et al., Citation2020), and is more easily mentally accessible (Vakoch & Wurm, Citation1997). Consequently, negative social concepts evoke more differentiated mental representations (Alves et al., Citation2017), leading – so the present finding – to greater variance in ratings. The pattern of consensus and fragmentation in the French and German affective culture reflects this information processing bias. From this perspective, the two significant interaction terms can be interpreted as modifiers to this fundamental pattern. Across all three EPA dimensions, concepts that are considered good and potent or bad and impotent become less consensual, while bad and potent or positive and impotent are more consensual. This negative relationship between evaluation and potency aligns with models of social cognition suggesting that goodness/warmth and potency/competence are negatively related (Fiske et al., Citation2002). Therefore, the association between positive evaluation and high potency may be cognitively more challenging and culturally more controversial within the population. Moreover, social position may significantly influence the perception of powerful roles, behaviors, and settings, with subordinate groups often experiencing power as illegitimate (Ambrasat et al., Citation2014; Fiske et al., Citation2002), thereby leading to less consensus. The greater consensus around negative and potent concepts compared to positive and potent concepts is not only consistent with social psychological research but may also be interpreted to be the basis for the “common-enemy” effect. As Jaegher and Kris (Citation2021) points out, the presence of a powerful common enemy tends to increase cooperation. The present findings support this idea, showing that members of a given culture more easily reach consensus on negative and potent concepts than on positive and potent ones. This suggests that within a given culture fostering cooperation around a common enemy (E-P+) is easier than building unity around a shared vision of leadership (E+P+). The ease of agreement on negative and powerful entities may therefore play a key role in facilitating collective action. The second significant interaction term suggests that concepts perceived as good and active, or bad and calm, tend to be rated more consensually than those perceived as good and calm, or bad and active. This pattern may be influenced by the composition of the “French-German EPA dictionary,” which contains a higher proportion of urban rather than natural settings, identities, and behaviors. Research shows that natural environments, along with the behaviors and identities associated with them, are often linked to fundamental affective states such as relaxation and calmness (Russell, Citation2003, ; Ulrich, Citation1983), serving an important psychological function as restorative environments (Hartig et al., Citation1991). In contrast, positively connoted urban and built environments are more likely to be perceived as lively and vibrant. As a result, the consensus around active and positive social concepts may stem from the overrepresentation of urban concepts in the dictionary. Conversely, the lower agreement on positive and calm social concepts could be due to the relative underrepresentation of nature-related concepts. At the individual level, the results provide valuable insights into participants’ cultural competence, which can be used to understand whether adherence to the cultural norm varies as a function of social position. Overall, the results suggest that sociodemographic characteristics alone explain only a small portion of the variance in cultural competence. In this regard, our findings align with previous research on affective enculturation (Heise, Citation2007). However, by using a representative sample and observing consistent results across six distinct French and German surveys and the three EPA dimensions, we can cautiously generalize about cultural fragmentation across different social groups. Young and middle-aged men with relatively few years of education were found to be the most culturally competent. In contrast, older women with many years of education provided ratings that tended to diverge from the majority. The higher competence of younger participants is particularly noteworthy, given that participants over 50 years old constituted the largest groups in the representative sample. This finding suggests that consensus is not simply a function of cohort size.Footnote4 The greater competence observed among younger participants – especially those under 50 – aligns with previous research, which indicates that individuals with larger social networks are more culturally competent and more capable of influencing the culture around them (Thomas & Heise, Citation1995). While data on participants’ social networks were not collected, evidence from the European Social Survey (ESS, Citation2020) indicates that individuals under 50 in French and German societies meet more frequently with friends, relatives, or colleagues (Supplementary Information – SI-10). Following our results, women are also less culturally competent, i.e. tend to provide less consensually shared answers. This result contradicts Heise analysis of enculturation, which found that female students were more likely than male students to provide the consensual answer because “females are more attuned than males to affective matters” (Heise, Citation2007, p. 164). Yet, the tendency of women to provide less consensual answers resonate well with empirical studies suggesting that women are less likely to endorse stereotypes (i.e. Barreto & Doyle, Citation2023; Carter et al., Citation2006). Expectation States Theory explicitly links the tendency to assume first order beliefs about status differences to third order beliefs about what people consensually think. Since women – according to the present results – are more likely to experience a mismatch between first and third order beliefs, they are less likely to attribute social validity to the latter beliefs and, accordingly, to endorse stereotyped views about status differences (Ridgeway & Correll, Citation2006). Another variable consistently related to cultural competence was education. Interestingly, participants with many years of education were found to be less culturally competent. This finding contradicts prima facie earlier analyses by Rossi and Berk (Citation1985), which showed that higher levels of formal education were associated with the ability to provide answers closer to the central tendency. However, Rossi and Berk’s research focused primarily on cognitive culture (Heise, Citation2007). In the context of affective culture, higher education often promotes values such as independence, uniqueness (Snibbe & Markus, Citation2005), and critical thinking (Brabeck, Citation1983). These academic values can foster habits and perspectives that differ from those of other social classes and the broader majority (Stephens et al., Citation2012). As a result, highly educated individuals are more likely to develop distinctive perspectives in affective judgment, leading to lower cultural competence in mainstream cultural contexts. Overall, the results paint an intriguing picture. Participants largely agree with each other especially when considering positive social concepts. Yet cultural norms become systematically less consensual when it comes to defining negatively connotated social concepts as well as positive and potent – i.e. leadership – and positive and calm – i.e. restorative – concepts. In these cases – within the French and German language culture – young and middle-aged men with relatively few years of education may intuitively and automatically participate in cultural processes, making their interactions within the social order smoother and more aligned with prevailing norms, this way experiencing less alienation (Heise et al., Citation2015). However, this proficiency in navigating the dominant cultural landscape may also contribute to a greater resistance to cultural change, as established norms and values become ingrained and are defended more staunchly (Schröder et al., Citation2016). 6. Limitations Three limitations to the results must be mentioned. Firstly, Bayesian models require that hyperparameters are fitted to the data. In the case of an incomplete block design – common in the ACT tradition – researchers may need to strike a balance between maintaining a consistent definition of hyperparameters across the stimuli set and fitting the model to a single set. Researchers must explicitly acknowledge this trade-off and carefully consider its consequences. Secondly, sociodemographic effects on enculturation and response styles are difficult to disentangle. Age and education are related to response styles (Greenleaf, Citation1992). The present analysis is not able to distinguish between these two classes of effects. Thirdly, the results are specific to French and German language cultures and cannot be generalized to illustrate enculturation and social order in general. This cultural specificity limits the broader applicability of the findings. Future research should aim to replicate these findings in other cultural contexts to enhance their generalizability and robustness. 7. Conclusions The cognitive shift in the sociology of culture, combined with recent methodological developments in Cultural Consensus Theory (CCT) and access to representative datasets, paves the way for the Affect Control Theory (ACT) community to refine its understanding of affective culture. This study contributes to that effort by demonstrating that while affective meanings are widely shared within language cultures, patterns of fragmentation are systematically structured rather than randomly distributed. Our findings reveal that consensus around affective meanings varies primarily along the evaluation dimension, which interacts with potency and activity, echoing established biases in social cognition and cooperation. Additionally, consensus is not uniform across individuals: social positions influence the extent to which one aligns with the dominant affective meanings, thereby identifying subgroups that are more or less likely to conform to majority-held cultural norms. In sum, this research offers two key contributions: methodologically, it pushes forward ACT research by employing a Bayesian framework that accounts for variation across social concepts and participants; theoretically, it highlights the complex interplay between consensus and fragmentation, shedding light on how affective meanings sustain the social order and legitimize status hierarchies. Ethics statement The surveys used in this manuscript involve human participants and were approved by the ethical board of Université Paris Cité (N° 2021-5). Informed consent In the data collection of the study, participants provided their informed consent at the beginning of the online questionnaire. Supplemental material Consensus and Fragmentation in Affective Culture_Supplementary Information.pdf Download PDF (9 MB)Consensus and Fragmentation in Affective Culture_Supplementary Information.pdfDisclosure statement No potential conflict of interest was reported by the author(s). Data availability statement All data and analysis scripts are openly available at https://osf.io/8974f/?view_only=01d9d99a1dc44386ae7dcc718e5d9f35. Supplementary material Supplemental data for this article can be accessed online at https://doi.org/10.1080/0022250X.2026.2659375 Additional information Funding Notes 1 It is important to note that, moving forward, the terms “culture” and “subculture” will follow the CCT terminology and refer to latent subgroups within the sample. Accordingly, cultural context is not defined by membership in established categories such as nationality or class. Rather, the cultural context to which a participant belongs is determined based on the patterns found in the data. 2 Cases of semantic ambiguity that may have led to variation in ratings include the verb “hit,” which was presented in German as “treffen.” This term can mean both “hit” and “meet someone.” Additionally, the potentially uncommon identity of “female playboy” and the behavior “chill somewhere,” which is primarily used in juvenile language, may have contributed to misunderstandings. Please note that these are English-language labels used for presentation purposes. Participants rated the concepts using French and German stimuli, which were validated by native-speakers. 3 A more accurate interpretation of the interaction term between Evaluation (E) and Potency (P) reveals that a positive coefficient sign indicates that an item is more difficult to rate when the evaluation and potency dimensions align – that is, when both are either positive or both are negative. Conversely, the negative sign of the interaction term between Evaluation (E) and Activity (A) suggests that an item is easier to rate when these two dimensions have the same signs. To simplify the discussion of these results, I will primarily interpret the findings from the perspective of positive signs, e.g. good and potent. This focus allows for a more straightforward interpretation and clearer comparisons across different dimensions. 4 Similarly, participants whose mother tongue was not exclusively French or German demonstrated higher competence on two out of the three dimensions. This finding highlights the robustness of the method. Larger groups are not necessarily more competent; instead, smaller groups that consistently provide answers aligning with the truth key identified in the overall sample are considered more competent. This result underscores that cultural competence is not solely a function of group size but rather the alignment with the dominant cultural interpretations within the sample. References - Alves, H., Koch, A., & Unkelbach, C. (2017). Why good is more alike than bad: Processing implications. Trends in Cognitive Sciences, 21(2), 69–79. https://doi.org/10.1016/j.tics.2016.12.006 - Ambrasat, J., Scheve, C. V., Conrad, M., Gesche, S., & Schröder, T. (2014). Consensus and stratification in the affective meaning of human sociality. Proceedings of the National Academy of Sciences of the United States of America, 111(22), 8001–8006. https://doi.org/10.1073/pnas.1313321111 - Ambrasat, J. & von Scheve, C. (2022). Affective meanings and social relations: Identities and positions in the social space. Emotions and Society, 4(2), 161–180. https://doi.org/10.1332/263169021X16340311969719 - Anders, R. (2017). Package ‘cctpack’. https://cran.r-project.org/web/packages/CCTpack/CCTpack.pdf(open in a new window) - Anders, R., Alario, F. X., & Batchelder, W. H. (2017). Consensus analysis for populations with latent subgroups: Applying multicultural consensus theory and model-based clustering with CCTpack. Cross-Cultural Research, 52(3), 274–308. https://doi.org/10.1177/1069397117727500 - Anders, R., & Batchelder, W. H. (2012). Cultural consensus theory for multiple consensus truths. Journal of Mathematical Psychology, 56(6), 452–469. https://doi.org/10.1016/j.jmp.2013.01.004 - Anders, R., & Batchelder, W. H. (2015). Cultural consensus theory for the ordinal data case. Psychometrika, 80, 151–181. https://doi.org/10.1007/s11336-013-9382-9 - Barreto, M., & Doyle, D. M. (2023). Benevolent and hostile sexism in a shifting global context. Nature Reviews Psychology, 2(2), 98–111. https://doi.org/10.1038/s44159-022-00136-x - Batchelder, W. H., & Anders, R. (2012). Cultural consensus theory: Comparing different concepts of cultural truth. Journal of Mathematical Psychology, 56(5), 316–332. https://doi.org/10.1016/j.jmp.2012.06.002 - Batchelder William, H., & Romney, K. A. (1986). The statistical analysis of a general Condorcet model for dichotomous choice situations. In B. Grofman & G. Owen (Eds.). Information pooling and group decision making: Proceedings of the second University of California, Irvine conference on political economy (pp. 103–112). Greenwich, CT: JAI Press. - Berger, P. L., & Luckmann, T. ( [1966] 1991). The social construction of reality: A treatise in the sociology of knowledge. Penguin UK. - Brabeck, M. M. (1983). Critical thinking skills and reflective judgment development: Redefining the aims of higher education. Journal of Applied Developmental Psychology, 4(1), 23–34. https://doi.org/10.1016/0193-3973(83)90056-4 - Carter, J. D., Hall, J. A., Carney, D. R., & Rosip, J. C. (2006). Individual differences in the acceptance of stereotyping. Journal of Research in Personality, 40(6), 1103–1118. https://doi.org/10.1016/j.jrp.2005.11.005 - Centola, D., & Macy, M. (2007). Complex contagions and the weakness of long ties. The American Journal of Sociology, 113(3), 702–734. https://doi.org/10.1086/521848 - Centola, D., Willer, R., & Macy, M. (2005). The emperor’s dilemma: A computational model of self-enforcing norms. The American Journal of Sociology, 110, 1009–1040. https://doi.org/10.1086/427321 - Cerulo, K. A., Leschziner, V., & Shepherd, H. (2021). Rethinking culture and cognition. Annual Review of Sociology, 47(1), 63–85. https://doi.org/10.1146/annurev-soc-072320-095202 - Dametto, D., Vieira, L., Blaison, C., & Schröder, T. (2024). French German EPA dictionary [Data set]. PsychArchives. https://doi.org/10.23668/psycharchives.14172 - Dametto, D., Vieira, L., Nöel, T., Schröder, T., Aelenei, C., & Blaison, C. (2024). The affective and linguistic roots of gender stereotypes. Does grammatical gender influence the affective meanings attached to social categories in French and German? Pre-Print. https://doi.org/10.31234/osf.io/h945z - Dametto, D., Vieira, L., Schröder, T., & Blaison, C. (2023). Charlemagne’s legacy: A consensus analysis of affective meanings in French and German culture. Social Psychology Quarterly, 87(4), 479–500. https://doi.org/10.1177/01902725231205855 - De Jaegher, K. (2021). Common-enemy effects: Multidisciplinary antecedents and economic perspectives. Journal of Economic Surveys, 35(1), 3–33. https://doi.org/10.1111/joes.12394 - DiMaggio, P. (1997). Culture and cognition. Annual Review of Sociology, 23(1), 263–287. https://doi.org/10.1146/annurev.soc.23.1.263 - Fiske, S., Cuddy, A., Glick, P., & Jun, X. (2002). A model of (often mixed) stereotype content: Competence and warmth respectively follow from perceived status and competition. Journal of Personality & Social Psychology, 82(6), 878–902. https://doi.org/10.1037/0022-3514.82.6.878 - Greenleaf, E. A. (1992). Measuring extreme response style. Public Opinion Quarterly, 56(3), 328–351. https://doi.org/10.1086/269326 - Hartig, T., Mang, M., & Evans, G. W. (1991). Restorative effects of natural environment experiences. Environment & Behavior, 23(1), 3–26. https://doi.org/10.1177/0013916591231001 - Heise, D. R. (2007). Expressive order: Confirming sentiments in social actions. Springer. - Heise, D. R. (2010). Surveying cultures: Discovering shared conceptions and sentiments. John Wiley & Sons. https://doi.org/10.1002/9780470575789 - Heise, D. R., MacKinnon, N. J., & Wolfgang, S. (2015). Identities, roles, and social institutions. In E. J. Lawler, S. R. Thye, & J. Yoon (Eds.), Order on the edge of chaos (pp. 165–188). Cambridge University Press. https://doi.org/10.1017/CBO9781139924627.010 - Hunt, P. M. (2012). Examining the affective meanings of interaction settings in the jamband music subculture. The Journal of Public and Professional Sociology, 4(1). https://digitalcommons.kennesaw.edu/jpps/vol4/iss1/5(open in a new window) - Lawler, E. J., Thye, S. R., & Yoon, J. (2015). Social psychology og social order. In E. J. Lawler, S. R. Thye, & J. Yoon (Eds.), Order on the edge of chaos (pp. 1–17). Cambridge University Press. https://doi.org/10.1017/CBO9781139924627.002 - Lizardo, O. (2017). Improving cultural analysis: Considering personal culture in its declarative and nondeclarative modes. American Sociological Review, 82(1), 88–115. https://doi.org/10.1177/0003122416675175 - Morgan, J. H., Rogers, K. B., & Hu, M. (2016). Distinguishing normative processes from noise: A comparison of four approaches to modeling impressions of social events. Social Psychology Quarterly, 79(4), 311–332. https://doi.org/10.1177/0190272516666794 - Mostafavi, M., Porter, M. D., & Robinson, D. T. (2024). Contextual embeddings in sociological research: Expanding the analysis of sentiment and social dynamics. Sociological Methodology, 55(1), 25–58. https://doi.org/10.1177/0081175024126072 - Osgood, C. E., Suci, G. J., & Tannenbaum, P. H. (1957). The measurement of meaning. University of Illinois Press. - Osgood, C. E., William, H., May, S. M. M., & Miron, M. S. (1975). Cross-cultural universals of affective meaning. University of Illinois Press. - Plummer, M. (2003). Jags: A program for analysis of Bayesian graphical models using Gibbs sampling. Proceedings of the 3rd International Workshop on Distributed Statistical Computing, 124(125.10), 1–10. - Quinn, J. F., Freeland, R. S., Rogers, K. B., Hoey, J., & Smith-Lovin, L. (2023). How cultural meanings of occupations in the U.S. changed during the COVID-19 pandemic. The American Behavioral Scientist, 67(1), 125–147. https://doi.org/10.1177/00027642211066041 - Ridgeway, C. L., & Correll, S. J. (2006). Consensus and the creation of status beliefs. Social Forces, 85(1), 431–453. https://doi.org/10.1353/sof.2006.0139 - Rogers, K. B. (2019). Sources of consensus and variegation in cultural affective meanings. Social Currents, 6(3), 219–238. https://doi.org/10.1177/2329496518805688 - Rogers, K. B. (2020). The problem of order: Understanding how culture predicts social action. Sociology Compass, 14(7). https://doi.org/10.1111/soc4.12800 - Romney, K. A., Batchelder, W. H., & Weller, S. C. (1987). Recent applications of cultural consensus theory. The American Behavioral Scientist, 31(2), 163–177. https://doi.org/10.1177/00027648703100200 - Romney, K. A., Weller, S. C., & Batchelder, W. H. (1986). Culture as consensus: A theory of culture and informant accuracy. American Anthropologist, 88(2), 313–338. https://doi.org/10.1525/aa.1986.88.2.02a00020 - Rossi, P. H., & Berk, R. A. (1985). Varieties of normative consensus. American Sociological Review, 50, 333–347. https://doi.org/10.2307/2095543 - Round, ESS 10: European Social Survey Round 10 Data. (2020). Data file edition 3.0. Sikt - Norwegian agency for shared services in education and research, Norway - data archive and distributor of ESS data for ESS ERIC. https://doi.org/10.21338/NSD-ESS10-2020 - Russell, J. A. (2003). Core affect and the psychological construction of emotion. Psychological Review, 110(1), 145. https://doi.org/10.1037/0033-295x.110.1.145 - Schneider, A. (1989). Mean affective ratings of 804 concepts by West German students in 1989. University of Georgia. UGA Affect Control Theory Website. http://research.franklin.uga.edu/act/(open in a new window) - Scholl, W. (2013). The socio-emotional basis of human interaction and communication: How we construct our social world. Social Science Information, 52(1), 3–33. https://doi.org/10.1177/0539018412466607 - Schröder, T. (2011). A model of language-based impression formation and attribution among Germans. Journal of Language & Social Psychology, 30(1), 82–102. https://doi.org/10.1177/0261927X10387103 - Schröder, T., Hoey, J., & Rogers, K. B. (2016). Modeling dynamic identities and uncertainty in social interactions: Bayesian affect control theory. American Sociological Review, 81(4), 828–855. https://doi.org/10.1177/0003122416650963 - Sewell, A. A., & Heise, D. R. (2010). Racial differences in sentiments: Exploring variant cultures. International Journal of Intercultural Relations, 34(4), 40012. https://doi.org/10.1016/j.ijintrel.2010.03.001 - Shepherd, H. (2011). The cultural context of cognition: What the implicit association test tells us about how culture works. Sociological Forum, 26(1), 121–143. https://doi.org/10.1111/j.1573-7861.2010.01227.x - Smith-Lovin, L. (1987). The affective control of events within settings. Journal of Mathematical Sociology, 13(1–2), 35–70. https://doi.org/10.1080/0022250X.1987.9990026 - Smith-Lovin, L., & Douglass, W. (1992). An affect control analysis of two religious subcultures. In V. Gekas & D. Frank (Eds.), Social perspectives on emotion (Vol. 1, pp. 217–247). JAI Press. - Smith-Lovin, L., Robinson, D. T., Cannon, B. C., Clark, J. K., Freeland, R., Morgan, J. H., & Rogers, K. B. (2016). Mean affective ratings of 929 identities, 814 behaviors, and 660 modifiers by University of Georgia and Duke University undergraduates and by community members in Durham, NC, in 2012-2014. University of Georgia. UGA Affect Control Theory Website. http://research.franklin.uga.edu/act/(open in a new window) - Snibbe, A. C., & Markus, H. R. (2005). You can’t always get what you want: Educational attainment, agency, and choice. Journal of Personality & Social Psychology, 88(4), 703. https://doi.org/10.1037/0022-3514.88.4.703 - Stephens, N. M., Townsend, S. S. M., Markus, H. R., & Phillips, T. L. (2012). A cultural mismatch: Independent cultural norms produce greater increases in cortisol and more negative emotions among first-generation college students. Journal of Experimental Social Psychology, 48(6), 1389–1393. https://doi.org/10.1016/j.jesp.2012.07.008 - Stets, J. E., & Burke, P. (2000). Identity theory and social identity theory. Social Psychology Quarterly, 63(3), 224–237. https://doi.org/10.2307/2695870 - Thomas, L., & Heise, D. R. (1995). Mining error variance and hitting pay-dirt: Discovering systematic variation in social sentiments. The Sociological Quarterly, 36(2), 425–439. https://doi.org/10.1111/j.1533-8525.1995.tb00446.x - Ulrich, R. S. (1983). Aesthetic and affective response to natural environment. In Altman, I., and Wohlwill, J.F. (Eds.), Behavior and the natural environment (pp. 85–125). Boston: Springer US. https://doi.org/10.1007/978-1-4613-3539-9_4 - Unkelbach, C., Alves, H., & Koch, A. (2020). Negativity bias, positivity bias, and valence asymmetries: Explaining the differential processing of positive and negative information. Advances in Experimental Social Psychology, 62, 115–187. https://doi.org/10.1016/bs.aesp.2020.04.005 - Vakoch, D. A., & Wurm, L. H. (1997). Emotional connotation in speech perception: Semantic associations in the general lexicon. Cognition & Emotion, 11(4), 337–349. https://doi.org/10.1080/026999397379827 - Weller, S. C. (2007). Cultural consensus theory: Applications and frequently asked questions. Field Methods, 19(4), 339–368. https://doi.org/10.1177/1525822X07303502