Angry populists or concerned citizens? How linguistic emotion ascriptions shape affective, cognitive, and behavioural responses to political outgroups

ABSTRACT

Emotion expressions of outgroup members inform judgements and prompt affective responses in observers, shaping intergroup relations. However, in the context of political group conflicts, emotions are not always directly observed in face-to-face interactions. Instead, they are frequently linguistically ascribed to particular actors or groups. Examples of such emotion ascriptions are found, among others, in media reports and political campaign messaging. For instance, anger and fear are frequently evoked in connection with and ascribed to right-wing populist groups. Yet not much is known about the specific effects that ascriptions of discrete emotions to outgroups can have on intergroup relations. With this pre-registered study, we contribute to bridging this gap by analysing the effects of ascriptions of anger and fear to a right-wing populist outgroup. In an online survey experiment, administered to a sample of the German general population (N = 3500), we manipulated the emotions ascribed to these outgroups using brief vignettes. Our findings suggest that ascriptions of anger to right-wing populist outgroups increase reactive anger in observers, whereas ascriptions of fear reduce anger as well as contempt towards populists. Effects of ascribed emotions on stereotype content and action tendencies could not be identified.

KEYWORDS:

• Intergroup emotion
• emotion ascription
• political psychology
• stereotype content
• right-wing populism

Introduction

In many parts of the world, right-wing political populism has been on the rise for some time, while political polarisation and exclusionary identity politics intensify, posing threats to liberal democratic values and social solidarity (e.g. Cohen, 2019). To explain these dynamics, the role of discrete emotions for the mobilisation and maintenance of populist support has been widely investigated in theory and research (Demertzis, 2006; Salmela & von Scheve, 2017, 2018). Taking affective and emotional processes into account can improve our understanding of the emergence of these forms of intergroup conflict, because emotions are a crucial aspect of intergroup relations and thus of intergroup conflict and conflict resolution. First, ingroup members’ experience of emotions directed at outgroups fundamentally shapes their relationship with these outgroups (Halperin, 2016; Mackie et al., 2000). Additionally, some scholars have argued that outgroup members’ perceived emotions are important for intergroup relations as well, as they provide important social-emotional cues, which in turn affect ingroup members’ responses towards these outgroups (Mackie & Smith, 2015, p. 284). Regarding both intergroup relations in general and right-wing populist support in particular, this latter perspective has not been thoroughly investigated. For the case of populism, so far, most approaches have either focused on the emotional experience of supporters of populist parties (Rico et al., 2017) or on emotional cues as part of populist communications (Widmann, 2021; Wirz, 2018; Wirz et al., 2018).

Ascribed emotions

In this contribution, we argue that in many constellations of intergroup conflict, outgroup members’ emotions are rarely inferred from the vocal or facial expressions of an ongoing emotional experience. Instead, in many situations, they are rather inferred from ascriptions of emotions towards outgroup members, typically through language and other media. This is hardly surprising because in intergroup conflict, in- and outgroup members often try to avoid direct contact and acquire information about one another primarily through third-party sources. We therefore use the terms emotion ascription and ascribed emotions to designate situations in which ingroup members learn about outgroup members’ emotions (and vice versa) only through proxies, typically language and discourse, such as media reports, narratives, or anecdotal evidence. A simple example would be one person telling another one that a third person was angry. A more complex example might be the discursive ascription of offence in religious conflict (George, 2016). This proposed use of the term emotion ascription is not to be confused with the understanding of ascription as a cognitive labelling of an actual experiential state of the self or other (Stroud, 2018). Also, ascriptions as understood here should not be confused with causal attributions or ascriptions that predict individual emotional responses to salient events (Weiner, 1985; Weiner et al., 1982).

Importantly, the truth value of ascribed emotions is often difficult to determine because emotional cues used to infer emotions in social interactions, such as facial expressions, social sharing, vocalisations, or gestures, are absent or unavailable. This implies a considerable degree of uncertainty with respect to the emotions that are actually experienced by outgroup members. Emotion ascriptions may correctly represent the emotions experienced by outgroup members, but they may also be misleading, speculative, or simply incorrect. In particular, in situations of intergroup conflict, emotion ascriptions can be used as a strategy of political communication, for example, to discredit outgroup members or otherwise portray them in a disadvantageous position, or to boost the confidence and standing of the ingroup. This is not to say that emotions can always be correctly inferred from emotion expression in social interactions without effort and uncertainty (e.g. because of impression management and display rules), but inferences can draw upon a broader range of cues and modalities.

Irrespective of their truth value, we assume that ascribed emotions substantially affect intergroup behaviour and intergroup emotions. The emotions an ingroup experiences towards an outgroup and their attitudes and behaviours towards the outgroup are likely to hinge on the emotions that are ascribed to the outgroup. Given that the described forms of emotion ascriptions are ubiquitous in socio-political contexts, we argue that their potential effects are essential for intergroup relations.

In light of these considerations, research on political group conflicts in general and on the emotional intergroup dynamics involved in right-wing populism in particular, might benefit from an analytical perspective that takes emotion ascriptions as possible factors in the relations of political groups into account. To our knowledge, there is no single study systematically investigating the effect of emotion ascriptions in an intergroup context. In this study, we contribute to bridging this gap by analysing in which ways ascriptions of discrete emotions to specific political outgroups contribute to the emergence of outgroup-directed emotions, stereotypical cognitive evaluations, and behavioural action tendencies in recipients.¹

To pursue this analytical goal, we focus specifically on linguistic ascriptions, as these are central for political discourses and are less ambiguous and fluid than, for instance, ascriptions via selective audio-visual portrayals.

Anger and fear: two “populist” emotions

Furthermore, in the present study, we concentrate on the ascriptions of two emotions that seem especially relevant in the political sphere, namely anger and fear. Both emotions lend themselves to the analysis of emotion ascriptions as they are already well studied in terms of their underlying appraisal structures, while they still differ on relevant relational dimensions. Anger and fear both result from primary appraisals that are motivationally relevant and incongruent with regards to interests of the self, but while anger focuses on external accountability and other-blame, fear entails secondary appraisals of threat, uncertainty and low coping potential (Smith & Lazarus, 1993). Additionally, fear causes behavioural tendencies of avoidance, while anger is thought to entail offensive and activating action tendencies (Frijda, 1987). All of this makes these emotions particularly suitable objects of research in the case at hand.

The practical relevance of anger and fear for the issue of emotion ascriptions is further underlined by the fact that both are frequently evoked as means of political mobilisation, particularly by populist actors (Keller & Berger, 2017; Rico et al., 2017; Wahl-Jorgensen, 2018; Webster, 2020). Antagonistic affect, often taking the form of visible expressions of intergroup anger, is also described as being central to the construction of exclusionary group boundaries which are at the very heart of populist identity politics (Rico et al., 2017). Additionally, as we argue, these emotions are often casually or strategically (self-)ascribed to political groups. Over time they have even crystallised into common expressions like Wutbürger (angry citizens) or besorgte Bürger (concerned citizens).

In the following, we derive a set of testable hypotheses pertaining to the effects of anger and fear ascriptions. We begin by discussing the implications for recipients’ immediate affective responses in the form of intergroup or, specifically, outgroup-directed emotions. Second, we discuss how emotion ascriptions may affect stereotypical perceptions of outgroups. Finally, we focus on possible effects on behavioural action tendencies towards outgroups.

Outgroup-directed emotions

To our knowledge, there is no theoretical model directly addressing the emotional consequences of linguistically ascribed emotions for general audiences or individual recipients. In contrast, the effects of emotional expressions on observers in social interactions are comparably well understood, emotional contagion and mimicry (Hatfield et al., 1993) as well as social appraisals (Fischer & Manstead, 2001) and social functions of emotions (Keltner & Haidt, 1999) being key concepts to understand this link. In the following, we will therefore revisit these concepts to determine whether and how they can contribute to understanding the emotional consequences of ascribed emotions.

Emotional contagion and mimicry are central drivers of affective reactions to other people’s emotions and are usually understood as processes occurring in face-to-face interactions and relying on perceptions and mimicking of facial or vocal expressions or bodily postures (Hatfield et al., 2014). Of course, these processes cannot be readily generalised to linguistic emotion ascriptions, for which there are no facial expressions or vocalisations to mirror. However, there has been some initial support for the view that emotional mimicry might be first and foremost a communicative act that relies on knowledge about other people’s emotions as well as on contextual information and does not necessarily require the direct perception of emotion expressions (Hess et al., 2014). Importantly, emotional responses to others’ emotions are by no means always convergent, as the notions of mimicry or contagion may imply. Rather, in competitive ingroup-outgroup settings, divergent or counterempathetic emotional responses are likely to occur, as van der Schalk et al. (2011) point out. For the realm of political group conflicts, Bucy and Bradley (2004) and Gabriel and Masch (2017) have documented such processes of divergent or counter empathetic responses to genuine emotion expressions.

Apart from emotional contagion, there are additional mechanisms through which outgroup members’ emotion expressions may affect recipients’ emotions. Social functional accounts of emotions stress that emotion expressions help coordinate social interaction by revealing others’ emotions, intentions, and relational orientations towards observers (Keltner & Haidt, 1999, p. 511). The concept of social appraisal assumes that such emotion expressions can be themselves objects of appraisal for observers and thus shape their emotional response in social situations (Fischer & Manstead, 2001, p. 223). Building on these perspectives, the emotion as social information framework (van Kleef, 2009; van Kleef et al., 2010) distinguishes affective reactions and social inferences as two separate pathways through which emotion expressions can shape observers’ responses. Affective reactions here are largely based on relatively automatic processes of contagion, mimicry and afferent feedback while inferential processes shape behavioural responses through the information that emotional expressions convey about the emoter. Importantly, van Kleef (2009, p. 186) assumes that affective reactions can also be shaped by the intentional or relational information transported by emotion expressions, which is in line with the idea of social appraisals.

For the case at hand, in which outgroup members’ emotions are not directly perceivable as emotion expressions and recipients rely on cognitive processing of emotion ascriptions made by third parties to make inferences about outgroup properties, it seems reasonable to assume that the inferential pathway is relatively dominant. This is because higher degrees of information processing by the observers of emotion displays lead to a prevalence of the social inference pathway over the affective reaction pathway (van Kleef, 2009, p. 187). We have argued above that emotions ascribed to others and others’ emotional expressions both allow individuals to infer another person’s emotional state with varying levels of uncertainty. To the degree that certain types of social inferences or affective reactions are based on the subjective knowledge about others’ emotions, we can thus assume that perceived expressions of a certain emotion and the ascription of the same emotion should lead to similar responses. That is, of course, assuming that recipients of ascriptions believe at least to some degree that the ascribed emotion is a valid clue to the actual emotional experience of another person or group. Taken together, these findings and assumptions allow us to derive some initial expectations for the effects of ascribed emotions from existing theory and research on the effects of emotion expressions in social interaction.

Based on the reviewed frameworks, some specific hypotheses can be devised regarding the emotional consequences of ascriptions of anger and fear to outgroup members. It has to be noted that although we are interested in emotion-specific effects of ascribed anger and ascribed fear, the lack of existing research on the effects of emotion ascriptions does not allow us to rule out an additional general effect of any emotion ascription regardless of which discrete emotion is ascribed. Accordingly, the specificity of the effect of one ascribed emotion is hypothesised in quantitative terms in relation to the other emotion.

Anger’s core appraisal is one of unfair injuries and norm-violations that the angry person wants to correct (Halperin, 2016, p. 51). Additionally, anger conveys a communicative message of strength and willingness to act (Halperin, 2016, p. 55). Inferring such threatening properties of outgroup members should result in a divergent emotional response of fear, as was also found by van der Schalk et al. (2011) for emotion expressions. Thus, we expect that ascriptions of anger to outgroup members increase outgroup-directed fear in recipients and that this effect is absent or significantly smaller for ascriptions of fear (Hypothesis 1.1). However, based on the affective reaction pathway of the emotion as social information framework (van Kleef, 2009), there is a possible alternative expectation: recipients may respond with reactive outgroup-directed anger to the anger ascribed to outgroup members because outgroup anger itself prompts appraisals of unfairness. Thus, we expect a positive and emotion-specific effect of ascriptions of anger on outgroup-directed anger while we would expect that effect to be absent or smaller for ascriptions of fear (Hypothesis 1.2). Shepherd et al. (2018) have actually found that outgroup threat by immigrants may lead to both, anger and fear directed at this outgroup. While their finding points to the possibility that both Hypothesis 1.1 and Hypothesis 1.2 may hold simultaneously, they find stronger positive effects of outgroup threat on fear and anxiety as compared to anger (Shepherd et al., 2018, p. 451). The question of which of the proposed effects is dominant may very well hinge on the question of whether outgroup anger is actually perceived as threatening or rather evaluated as unfair or immoral. Although this question is beyond the scope of this study, support for either of the proposed hypotheses may be a point of departure for further research.

Ascribed fear, as compared to ascribed anger, might be associated with sympathetic concern in a less competitive setting. However, for a strongly polarised ingroup-outgroup setting, we follow the suggestion by van der Schalk et al. (2011), according to which outgroup fear would be read as a submission signal and should elicit contempt or general aversion (p. 290). While this assumption is less well-established as compared to the effects of anger, the authors are able to demonstrate the proposed effect of outgroup fear using coded facial expressions (van der Schalk et al., 2011, p. 292). Although the generalisation of these findings to ascribed emotions and self-reported outgroup-directed emotions is not beyond doubt, we assume that both forms transport the same relational information and should thus result in the same appraisals with regards to the outgroup. Accordingly, we hypothesise that ascribed fear results in increased contempt towards the outgroup and that this effect is absent or smaller for ascribed anger (Hypothesis 2). In principle, we might focus on other aversive emotions like disgust as well. However, contempt is firstly more strictly related to human targets and their actions and secondly, while disgust implies revulsion by fear, contempt implies detachment and exclusion due to perceived inferiority of the other (Miceli & Castelfranchi, 2018). This perception of inferiority resonates well with the submission signalled by other’s fear as proposed by van der Schalk et al. (2011).

Stereotype content

Next, we ask in which ways the ascriptions of emotions to populist political groups shape stereotypical beliefs about these groups. Here, we refer to the stereotype content model, which assumes two basic dimensions of stereotype content, namely warmth and competence (Fiske, 2015; Fiske et al., 2002). Usually, stereotypes about social groups are seen 1) as relatively stable and are understood as predictors of group-related emotions (Cuddy et al., 2008), 2) as linked to emotional prejudice that is articulated through ascriptions of uniquely human emotions to outgroups (Leyens et al., 2000) and 3) as causes of behavioural tendencies towards outgroups (Fiske, 2015). Thus, emotions are usually understood as results or, at most, dimensions of stereotypes. However, if we assume that individuals make inferences concerning the state and self-perception of the outgroup based on outgroup emotions, as discussed before, such emotions could also inform, shape, and thus causally precede stereotypical cognitive perceptions of outgroups.

It was discussed before that ascribed anger is likely linked to inferences of an outgroup’s self-perception of strength. Accordingly, we expect that ascriptions of anger but not of fear to outgroup members cause evaluations of increased competence (Hypothesis 3). Inversely, as fear entails appraisals of low coping potential (Halperin, 2016, p. 77), we expect that ascribed fear but not ascribed anger results in decreased competence evaluations (Hypothesis 4). Evaluations regarding stereotypical warmth should be dependent on the relationality of the respective groups, in that competitive relationships to outgroups lead to decreased stereotypical warmth (Cuddy et al., 2008, pp. 94–95). Since we are interested in a polarised ingroup-outgroup setting that can already be assumed to be competitive, the question remains how ascribed anger and fear further impact relational evaluations in these settings.

Interpersonal fear is frequently portrayed as signalling distress and thus prompting social support by others (Fischer & Manstead, 2016, p. 428). However, as van Kleef et al. (2010, pp. 58, 61) point out, in competitive settings, such cries for help lack these affiliative properties and may even increase competition between individuals or prompt aversive tendencies. Similarly, expressions of anger are likely to reinforce competition as anger entails core relational themes of external blame (Smith & Lazarus, 1993, p. 238) and signals dominance and aggression (van Kleef et al., 2010, p. 60). Anger, as Fischer and Manstead (2016, p. 429) argue, serves a social distancing function and also implies self-elevation. Even if anger is not immediately directed at the individual in question, it is still likely to stress a competitive relationship. Generalising these characteristics to the intergroup context, we assume that ascribed anger and fear reinforce competitive relations, since we assume that they convey similar information and prompt appraisals comparable to actual anger and fear expressions. Thus, we expect a negative effect of both ascriptions of anger (Hypothesis 5) and fear (Hypothesis 6) on recipients’ evaluations of stereotypical warmth.

Action tendencies

Finally, the action tendencies that are fostered by ascribed emotions are perhaps the most relevant outcome with regards to the dynamics of intergroup conflict. Especially approach and avoidance tendencies are practically relevant because they not only predict offensive action but can also facilitate or inhibit intergroup contact that is central to reducing prejudice and intergroup conflicts (Paolini et al., 2018). However, the expected causal relationship is less straightforward here. There are at least two potential pathways through which ascribed emotions can affect behaviours. First, from intergroup emotion theory, we know that outgroup-directed anger is associated with approach tendencies towards outgroups (Mackie et al., 2000; van Zomeren et al., 2004; Yzerbyt et al., 2003). Outgroup-directed aversive emotions like fear, disgust or contempt are generally linked to avoidance behaviour (Devos et al., 2016). Second, stereotype content is thought to be predictive of behavioural tendencies; especially, evaluations of outgroup warmth predict approach versus avoidance tendencies (Cuddy et al., 2008).

If some of our previous hypotheses hold and ascribed fear actually leads to contempt (see Hypothesis 2) and decreased perceptions of stereotypical warmth (see Hypothesis 6), we can assume that fear ascriptions to outgroups also increase avoidance tendencies in recipients (Hypothesis 7). In the case of anger ascriptions to the outgroup, dependent on the induction of outgroup-directed fear or anger in recipients, two outcomes are conceivable that seem contradictory at first but may well be expected to manifest simultaneously. If Hypothesis 1 holds and ascribed anger causes fear in recipients, we would expect an increase in avoidance tendencies for ascribed anger as well (Hypothesis 8.1). This effect should also be reinforced by a decrease in perceived stereotypical warmth – as expected according to Hypothesis 6 – that would in turn increase avoidance tendencies.

It should be noted that some studies find support for intergroup fear as a motivating emotion for collective action rather than for withdrawal. Van Zomeren et al (2010) show that fear can motivate collective action to reduce the threat posed by climate change while Shepherd et al. (2018) find that fear, caused by symbolic and realistic threats posed by migrants, increased respondents’ willingness to engage in collective action to reduce the rights of the already disadvantaged outgroup. However, we argue that these forms of collective action do not directly translate to the approach tendencies we are interested in. In both studies, collective action was aimed at preventing an external threat without seeking direct confrontation with an outgroup that posed the risk. We, in contrast, are interested in more interactional approach or avoidance tendencies towards outgroup members themselves, as these are relevant for intergroup contact. This is especially the case, since there is no contested political issue researched in our study that could be addressed by collective action.

There is, however, one alternative expectation regarding the effects of ascribed anger on action tendencies that we have to consider. To the degree to which ascriptions of anger spark reactive outgroup-directed anger rather than outgroup-directed fear, that is if the expected elicitation of reactive anger (see Hypothesis 1.2) outweighs the expected elicitation of fear (see Hypothesis 1.1), the expected increase in avoidance tendencies caused by ascribed anger might not hold. In this case, anger ascriptions should actually increase approach tendencies and this increase should be absent or smaller for fear ascriptions (Hypothesis 8.2). This possible ambivalence is underlined by a finding by Kamans et al. (2011) according to which outgroup threat, which may – as we argue – be inferred from anger ascriptions, can lead to both avoidance tendencies prompted by outgroup-directed fear or approach tendencies prompted by outgroup-directed anger, depending on the power relations between the groups. For a detailed list of all hypotheses, please refer to the supplemental material.

The present study

Since complex emotional processes are not likely to be fully understood in isolation, they should be analysed in their situatedness in social environments (Griffiths & Scarantino, 2008). Arguably, social context is even more important for emotions that are based on social referencing between groups. In a similar vein, Bar-Tal et al. (2007, pp. 443–446) stress the importance of taking historical, cultural and socio-political contexts into account when analysing collective emotions in conflict settings. Thus, we chose to analyse the effects of emotion ascriptions in the specific context of contemporary political group conflicts surrounding populist parties and movements in German society. Accordingly, in order to test, the hypotheses formulated above by means of causal inference, we conducted an online vignette experiment in which we manipulated linguistic ascriptions of anger and fear to a particular political outgroup, namely supporters of the right-wing populist Alternative für Deutschland (AfD) and the protest movement Pegida. The study design has been peer-reviewed and pre-registered prior to data collection and analysis (https://osf.io/qujgr).

Studying emotion ascriptions in the specific context of political populism allows us to achieve a stronger ecological validity, to speak to the larger problem of political polarisation tendencies and additionally gives us a clearer a priori understanding of the actual intergroup relations, narratives, issue positions and interpretations that are at play. While the study’s focus reflects our interest in the political context in Germany, we have to acknowledge that broad generalisability to other context would require further research beyond the present study. Pre-existing attitudes towards the selected political groups may decrease the propensity of respondents to change their attitudes and emotions towards these groups as a response to experimental manipulation. However, constructing unknown or fictitious groups as references for the present study did not seem feasible because we are interested in political settings, in which groups are usually already known. Thus, we consider the proposed design in which pre-existing and widely known groups serve as references for ascribed emotions as a conservative test of the hypotheses in a relevant context. Accordingly, possible negative findings could not give sufficient evidence for rejecting the respective hypotheses for less politicised contexts. We are, however, encouraged by studies on the interpersonal level regarding emotion expressions by political leaders. While these find interactions of emotional stimuli and party-identification or pre-existing attitudes, these do affect but not suppress treatment effects (Gabriel & Masch, 2017; McHugo et al., 1985).

Methods

Participants

In order to test the formulated hypotheses in an actual socio-political setting, a survey sample of the general German population was obtained from an external survey provider ($N=3500$).² To achieve a higher degree of representativeness, non-crossed quotas pertaining to school education (three levels), gender and age were defined.³ The target population consisted of individuals between 18 and 74 years, who live in Germany and identify as male or female.⁴

While being an economic option, online surveys sometimes suffer from poor data quality due to different forms of careless responding (Shamon & Berning, 2020). To assure sufficient quality of the responses, we included an attention check item into the survey questionnaire. A total of 406 respondents who failed the attention check were screened out by the survey provider and were not counted towards the total sample size. Our attention check item follows recommendations by Shamon and Berning (2020), according to which item-level attention checks are preferred to those included into instruction texts. Adapting one of the proposed attention check items, we added one item to a randomised 7-point Likert scale, asking respondents to check the highest option for that item to indicate that they have read the item. The same attention check was successfully applied in a pilot test, as described below.

Measures

Outcome measures

After the experimental manipulation, respondents’ outgroup-directed emotions were measured using self-reported ratings for seven discrete emotions on 7-point Likert scales, of which only anger, fear and contempt ratings are relevant to the present study. The question wording read: to which degree do you experience the following emotions when you think about the supporters of these parties or movements? Scale items ranged from not at all (1) to strongly (7). As a measure for stereotype content, respondents completed three questionnaire items for perceived warmth and three items for perceived competence that were initially developed by Fiske et al. (1999, 2002). Due to temporal limitations, we utilised a shortened version of the original scale as proposed by Eckes (2002). The German translation was adopted from Asbrock (2010). However, since we are interested in respondents’ own impressions rather than their normative knowledge of stereotype content, we omitted the addition as viewed by society from the original wording and instead asked for respondents’ own perceptions. For stereotypical warmth, respondents were asked, how warm are supporters of AfD and Pegida? The same question was repeated for the adjectives likeable and good-hearted. Stereotypical competence was measured with the adjectives competent, competitive and independent in the same question wording. Answer options ranged from not at all (1) to very (7) and the different questions were presented in randomised order. Action tendencies were measured using three items for approach tendencies and three items for avoidance tendencies which were adapted from Mackie et al. (2000). For approach tendencies, respondents were asked, “Do you want to confront (oppose, contradict) supporters of AfD and Pegida?” The three avoidance items asked whether respondents wanted to avoid, keep distance or have to do with (inversely coded) outgroup-members using the same wording as before. Again, the order of items was randomised while answer options ranged from not at all (1) to very much (7). For all multiple-item measures, sum scores were calculated, applying listwise deletion of missing data.

Controls

Additionally, socio-demographic variables and further covariates including migration background, subjective social class assignment, voting preferences (and political left-right self-placement), gender, age and education were collected. Respondents were asked whether they are born on the territory of current-day’s Germany. Additionally, they were asked where each of their parents was born with answer options Germany and other. Respondents who were not born in Germany or have at least one parent that was not born in Germany are considered as having a migration background and dummy-coded accordingly. Social class was measured subjectively by asking respondents to which social stratum they would assign themselves, using an item from the German General Social Survey (Zentralarchiv für Empirische Sozialforschung (ZA) & Zentrum Für Umfragen, Methoden Und Analysen (ZUMA) E.V., 1997). Options were lower class, working class, middle class, upper middle class and upper class. Group identification with the supposed outgroup was approximated based on the voting preference. Voting preference was measured by asking respondents which party they would vote for, if there was a general election on the following Sunday.⁵ We dummy-coded those who planned to vote for the right-wing populist Alternative für Deutschland or would vote for that party if voting was mandatory to be identifiers with the right-wing populist outgroup. In case a more fine-grained measure for political attitudes was needed, we additionally included a political left-right self-placement item adapted from Breyer (2015). However, we argue that political placement on a left-right scale does not neatly translate to identification with a particular political group, which is why we opt for the voting preference item for the primary analyses. Education was measured using one question for school education, vocational education and tertiary education each, based on the ES-ISCED classification that is part of the European Social Survey (European Social Survey ERIC, 2019; Schneider, 2016). For reasons of parsimony, education was coded as a three-level aggregate variable, following the recommendation by Gesthuizen et al. (2011). In this procedure ISCED levels 0–2 are coded as low education, levels 3–4 as middle education and levels above 5 are coded as high education. For additional details regarding the operationalisation of all variables and the measurement instrument, please refer to the supplemental material.

Manipulation

In order to systematically manipulate ascribed emotions, we produced short textual vignettes. Since respondents should pick up an emotion ascription, as defined before, rather than a factual statement about the emotionality of the outgroup, the vignette texts mimicked newspaper headlines both in their wording and their graphic presentation. All vignettes referred to the political group, supporters of two German right-wing populist movements or parties (Alternative für Deutschland (AfD) and Pegida), to which emotions were ascribed. In the ascribed anger and ascribed fear conditions, simple emotion adjectives were used to introduce the emotion ascription. In the control condition, the emotion ascription was omitted, while the contextual reference to the political situation in Germany was kept for reasons of coherence and comparability.

For the ascribed anger and fear treatments, the German original wording of the vignettes translates to “Supporters of AfD and Pegida are angry due to the political situation in Germany” and “Supporters of AfD and Pegida are fearful due to the political situation in Germany” respectively. The control condition utilised a rather neutral German approximate synonym of “deal with” or “engage with” instead of the emotion ascription. The wording translates to, “Supporters of AfD and Pegida engage with the political situation in Germany”. For the original wording and the visual presentation, please refer to the supplemental material.

Manipulation check and pilot test

To assess the effect of the experimental manipulation, we devised a series of manipulation checks. In the first and most immediate manipulation check, respondents were asked how strongly discrete emotions were ascribed to the outgroup in the headline showed in the vignette. Emotion items included anger, fear, shame, guilt and hope and were presented in a randomised order. Answer options ranged from not at all (1) to very strongly (5). Two additional manipulation checks were developed that asked to which degree these emotions were typically ascribed to outgroup members (e.g. in public discourses or the media) and to which degree these outgroup members actually experienced the emotions in the perception of respondents. To test the efficiency of the manipulation, we pilot-tested the proposed vignettes followed by all three manipulation check items. The collected pilot data (Wunderlich et al., 2022a) was comprised of N = 602 respondents recruited by the survey provider. Quotes for gender and age, but not education were applied based on the same quota calculations that were used in the main study. Additionally, we included the item battery for the warmth and competence measures to pilot-test the attention check that was included into this battery. The attention check item resulted in an acceptable rejection rate of 7.34%; rejected respondents did not count towards the total sample size.

Results from the pilot study generally provided support for the experimental manipulation as a series of linear regression models suggests. When regressing the dummy coded treatment condition on the perceived anger ascription in the vignette, only the ascribed anger condition had a significant positive effect over the control condition, ($b=1.11,t\left(557\right)=7.11,p<0.001$). Inversely, only the ascribed fear condition had a significant positive effect on perceived fear ascriptions in the vignette over the control condition, ($b=1.40,t\left(556\right)=9.62,p<0.001$). In both cases, the regression coefficients indicate that the effects on the mean on a 5-point Likert scale are not only highly significant but also quite substantial in size. The second, more abstract manipulation check asked respondents which emotions are generally ascribed to the reference groups in the public discourse. Here, we found a significant positive effect of the ascribed anger treatment (reference: control) on perceived typical anger ascriptions ($b=0.43,t\left(548\right)=3.31,p=0.001$). Perceptions of typically ascribed fear were not significantly affected by the treatments. For the last manipulation check, in which we asked for respondents’ perceptions on the actual emotional experience of the reference group, the results were similar: there was a significant positive effect of the ascribed anger treatment (reference: control) on perceived anger experiences of the reference group ($b=0.32,t\left(545\right)=2.49,p=0.013$). Again, there are no effects of the ascribed fear treatment on either perceived experienced anger or fear in the reference group. Generally, these results are encouraging as respondents evidently picked up the emotion ascription from the vignettes. For ascribed anger, but not for ascribed fear, this effect even extended to reported perceptions of typical ascriptions in the public discourse and reported perceptions of actual emotion experiences. A detailed description of the questionnaire and the complete regression results are included in the supplemental material of this article.

Given that the pilot test already provided good support for the feasibility of the experimental manipulation and in order to limit the decay of treatment effects over time, only the first manipulation check question was included into the main study, directly following the experimental manipulation. Here, we expected similar statistically significant treatment effects as in the pilot test, on which the primary outcomes of this study are contingent.

Procedure

After providing informed consent to the processing of their anonymised data, respondents completed an online survey of approximately 20 min duration that is outlined in detail in the supplemental material.⁶ After a brief set of socio-demographic questions that was used for controlling the survey quotas, respondents answered questions pertaining to political self-placement and voting preferences. These questions were posed before the manipulation to avoid priming effects. Next, respondents were prompted to read the following text carefully and one of the three experimental vignettes outlined in the preceding section was randomly assigned and displayed. Following the manipulation, respondents completed the manipulation check and outcome measures. Last, respondents completed the remaining set of attitudinal and socio-demographic questions, including measures for social class self-placement and migration background. The collected data set (Wunderlich et al., 2022b) is publicly available online (https://osf.io/p6cfm/).

Data analysis

After data collection, erroneous data and univariate outliers on dependent variables were excluded. Multiple OLS regression models including all relevant control variables and dummy variables for the treatment conditions were fit to the data for each outcome variable. Also, an interaction term between the treatment variable and the group identification variable was added to each model to limit the estimated treatment effects to those respondents who perceive the right-wing populists as an outgroup. Subsequently, necessary assumptions for OLS regression were tested. Details on these procedures can be found in the pre-registered data analysis plan (see the supplemental material).

Results

Preliminary analyses

The online survey yielded a total sample of $N=3502$ respondents. 50.66% of respondents were female, and the mean age was 46.29 years ($SD=15.11$). After the exclusion of potentially erroneous data and univariate outliers according to the pre-registered procedures, a final sample of $N=3229$ remained for further analysis.

All measurement scales applied in the online survey reached good degrees of internal consistency (warmth: $\alpha =0.95$, competence: $\alpha =0.88$, avoidance: $\alpha =0.87$ and approach: $\alpha =0.78$). However, the distributions of some of the dependent variables show irregularities, in that they are strongly polarised towards one or both of the extremes. The distributions of outgroup-directed anger and contempt are bi-modal with many respondents on the floor and ceiling of the distribution, while warmth, competence and avoidance are skewed with either a high floor or ceiling percentage, as detailed in Table 1.

Table 1. Distributions of dependent variables.

	Min	Max	Mean	SD	Median	Floor %	Ceiling %	Skewness
Anger	1.00	7.00	4.23	2.17	4.00	20.04	21.71	−0.22
Fear	1.00	7.00	3.40	2.00	3.00	28.07	8.98	0.26
Contempt	1.00	7.00	4.37	2.21	5.00	19.42	25.18	−0.31
Warmth	3.00	21.00	7.01	4.73	5.00	37.41	1.89	1.15
Competence	3.00	21.00	8.67	4.60	8.00	18.09	2.11	0.65
Avoidance	3.00	21.00	16.06	5.33	18.00	2.04	35.34	−0.81
Approach	3.00	21.00	11.99	5.13	12.00	8.80	6.88	−0.06

In terms of experimental manipulation, the manipulation check provides strong support for the effectiveness of the vignettes. The experimental manipulation predicted a significant proportion of the variance of the manipulation check items for perceived ascribed anger $\left(F\right(2,3226)=210.4,p<0.001,R^2=0.12)$ and perceived ascribed fear $\left(F\right(2,3226)=195.5,p<0.001,R^2=0.11)$. Respondents in the ascribed anger condition perceived stronger ascriptions of anger in the vignettes as compared to those in the control condition ($b=1.36,t\left(3226\right)=15.15,p<0.001$). Similarly, respondents in the ascribed fear condition detected higher levels of fear ascriptions in the vignette as compared to those in the control condition ($b=1.43,t\left(3226\right)=16.15,p<0.001$). Interestingly, respondents in the opposite treatment condition reported significantly less ascribed fear ($b=-0.19,t\left(3226\right)=-2.11,p=0.035$) or ascribed anger ($b=-0.42,t\left(3226\right)=-4.7,p<0.001$) as compared to control. First, this suggests that respondents did not merely detect any emotion ascription but were able to discriminate between anger and fear. Second, it could be the case that respondents in the neutral condition still suspected some ascribed emotionality in the neutral vignette but not so much in the opposite ascribed emotion conditions. Overall, together with the pilot test, the manipulation check underlines the effectiveness of the experimental manipulation.

Main analyses

In line with the pre-registered data analysis plan, assumptions of OLS regression were tested prior to interpreting the results of the respective multiple regression models for each outcome variable and the following corrections were applied: first, a squared term for the control variable age was included into all models except for the model for outgroup-directed fear and competence evaluations to account for the curvilinear relationship between age and the outcome measures. Second, multivariate outliers with a Cook’s Distance larger than $4/n$were excluded from the regression models. Third, robust standard errors were estimated to account for the existing heteroskedasticity of the regression residuals. Last, Bonferroni-corrected p-values were calculated to account for the problem of multiple independent statistical tests within a single sample. Below, we report coefficients and p-values based on robust standard errors prior to the Bonferroni correction. An annotated script with all steps of the analysis and the detailed output can be reviewed online (https://osf.io/7yxs8/).

Outgroup-directed emotions

Deviating from our expectation, the regression predicting respondents’ outgroup-directed fear ($F\left(14,3051\right)=84.4,p<0.001,R^2=0.14$) did not yield statistically significant results for ascriptions of anger ($b=-0.12,t\left(3051\right)=-1.31,p=0.19$) as well as ascriptions of fear ($b=0.02,t\left(3051\right)=0.246,p=0.806$) to the outgroup. Thus, we could not find any support for our Hypothesis 1.1. according to which anger ascribed to the outgroup should have led to outgroup-directed fear. However, we also hypothesised an alternative effect according to which ascribed anger might cause reactive out-group directed anger in respondents (Hypothesis 1.2). In fact, the regression model for outgroup-directed anger ($F\left(15,3041\right)=256.12,p<0.001,R^2=0.24$) yields a modest, yet significant positive effect of ascribed anger ($b=0.22,t\left(3041\right)=2.39,p=0.02$) which supports this hypothesis. Although we did not hypothesise or expect any effect of fear ascriptions on outgroup-directed anger, surprisingly, the same regression model shows a significant negative effect of fear ascriptions to the outgroup ($b=-0.23,t\left(3041\right)=-2.4,p=0.016$). This indicates that ascriptions of fear to the outgroup in fact decrease the anger that respondents report to experience when thinking of that outgroup.

According to Hypothesis 2, ascribed fear should have increased contempt directed towards the outgroup. However, our findings contradict this expectation in that ascribed fear exhibits a negative rather than positive effect ($b=-0.22,t\left(3029\right)=-2.35,p=0.019$) in the regression predicting outgroup-directed contempt ($F\left(14,3029\right)=206.11,p<0.001,R^2=0.3$). As expected, ascribed anger had no effect on outgroup-directed contempt ($b=0.04,t\left(3029\right)=0.481,p=0.63$). A major limitation to these results is that the statistical significance of all effects of interest does not hold after the Bonferroni correction for the experiment-wise error-rate. The implications of this will be discussed under limitations and corrected p-values are reported in the supplemental material.

Stereotype content

We did not find any statistically significant effects in support of our hypotheses concerning the effects of emotion ascriptions on evaluations of stereotypical warmth or competence. There were no effects of anger ($b=0.01,t\left(2985\right)=0.029,p=0.977$) and fear ($b=-0.07,t\left(2985\right)=-0.399,p=0.69$) on stereotypical competence. Also, stereotypical warmth was not affected by ascribed anger ($b=0.03,t\left(2662\right)=0.207,p=0.836$) or fear ($b=0.02,t\left(2662\right)=0.172,p=0.864$).

Action tendencies

Similarly, we did not find significant effects of emotion ascriptions on action tendencies towards the outgroup. Neither anger ($b=0.117,t\left(2796\right)=0.649,p=0.516$) nor fear ($b=0.09,t\left(2796\right)=0.5,p=0.617$) ascribed to the outgroup predicted avoidance behaviours and approach behaviours were also affected by neither anger ($b=-0.07,t\left(3030\right)=-0.328,p=0.744$) nor fear ($b=-0.26,t\left(3030\right)=-1.155,p=0.248$). Our hypotheses regarding the treatment effects on action tendencies implied mediation paths via outgroup-directed emotions and stereotypical evaluations. Despite the absence of direct treatment effects, correlations among dependent variables (Table 2), are in line with this reasoning. However, the unexpected negative effect of ascribed fear on contempt suggests a suppression rather than a mediation of a potential treatment effect on avoidance. In contrast, the positive effect of ascribed anger on reactive anger, raises the question of why no direct effect on approach tendencies could be inferred, despite the correlation of both dependent variables. Both observations suggest that the causal framework underlying Hypotheses 7 and 8.2 may be underspecified.

Table 2. Zero-level correlations of dependent variables.

	Anger	Fear	Contempt	Warmth	Competence	Avoidance
Anger	1	.	.	.	.	.
Fear	0.60	1	.	.	.	.
Contempt	0.78	0.54	1	.	.	.
Warmth	−0.50	−0.33	−0.59	1	.	.
Competence	−0.44	−0.25	−0.52	0.85	1	.
Avoidance	0.58	0.42	0.67	−0.78	−0.67	1
Approach	0.52	0.36	0.55	−0.40	−0.37	0.54

Please note that we relied on offensive rather than affiliative conceptualizations and measures of approach tendencies, which explains the negative correlation with warmth.

Discussion

The present study set out to research how emotions ascribed to outgroups could affect responses towards these groups analogously to directly perceived emotion expressions of outgroup members. As our manipulation check showed, respondents do in fact pick up ascriptions of discrete emotions from short descriptions, as they are frequently presented in newspaper headlines. Results from our study provide initial insights into the effects of these emotion ascriptions on responses towards the outgroups in question but are limited and warrant further scrutiny.

To begin with, we could only uncover statistically significant effects of emotion ascriptions on emotional responses towards the outgroup, not on stereotypical cognitions or action tendencies. There may be several explanations for this: first, the relatively small treatment effects may have further deteriorated over time within the survey, while respondents moved from the questions pertaining to outgroup-directed fear, anger and contempt to the item batteries for stereotype content and action tendencies. Future studies could vary the ordering of questions and apply stronger treatments to account for this. Importantly, emotion ascriptions in real-world settings likely occur repeatedly in different contexts like mediated discourses or private conversations, which is in stark contrast to our brief experimental manipulation. Repeated and potentially multi-modal ascriptions could be more effective in gradually transforming preconceptions of outgroup emotionality and could thus also be more consequential for real-life outcomes than a single headline. Longer vignettes with repeated ascriptions or treatments like group discussions entailing emotion ascriptions could be a first step to approximate this in experimental research. Second, the high floor or ceiling percentages described above may have resulted in floor or ceiling effects for some of the dependent variables in question, severely limiting their variance and thus restricting the ability to detect covariances between ascribed emotions and the responses to outgroups. Qualitatively, we could interpret these floor and ceiling effects as a result of strong pre-existing dispositions that respondents exhibit with regards to the populist right-wing groups AfD and Pegida that were researched in this study. These lasting stereotypical evaluations or behavioural tendencies towards outgroup members could remain largely unaffected by new information on outgroup emotions. In contrast, outgroup-directed emotions may be more volatile and less polarised which would make them more likely to be affected by short-term impressions of the outgroup’s emotions or corresponding ascriptions. Last, our expectations regarding effects on action tendencies are likely not taking the full complexity of causal mechanisms into account, as described above.

With regards to the effects of ascribed emotions on outgroup-directed emotions, we find, against our expectations, no significant effect of ascribed anger on outgroup-directed fear (Hypothesis 1.1.), which suggests that respondents do not infer strong threatening properties from the ascribed anger. Yet the uncovered positive effect of ascribed anger on outgroup-directed anger supports our alternative hypothesis (Hypothesis 1.2) according to which respondents’ appraisals of the right-wing populists' anger as in itself unfair or unjust leads to reactive outgroup-directed anger. Taken together, the findings suggest that this appraisal of ascribed anger as unfair was likely stronger than any potentially perceived threat, resulting in reactive anger rather than fear of the outgroup. The fact that average competence evaluations of the outgroup were relatively low in general may corroborate this interpretation since an appraisal of threat by respondents would presuppose at least some ability of the outgroup to effectively act on their ascribed anger. Furthermore, respondents may simply not feel personally threatened by right-wing populists’ anger but may instead assume that the outgroup’s anger is directed at others, like certain minority groups they are not themselves part of, and thus poses a rather abstract political threat. In this scenario, reactive anger on behalf of such threatened third groups could again be a likely reaction. Thus, if respondents perceived of themselves as in a position of relative competence or strength as compared to the right-wing populist outgroup or not as direct targets of the outgroup’s ascribed anger, reactive anger seems to be a more plausible response to ascribed anger than out-group-directed fear, which would be in line with our empirical findings.

The fact that ascribed fear was found to decrease anger towards the outgroup remains open for interpretation as we did not specify any hypotheses regarding this effect. One explanation could be that a fearful outgroup might be perceived as more defensive and thus less likely to offend or attack others, making appraisals of unfairness that underly outgroup-directed anger less likely. A view according to which a fearful outgroup is generally evaluated in more positive terms is further supported by our finding that fear decreases contempt, even though we originally hypothesised that the weakness and submission associated with ascribed fear should result in detachment, aversion and thus increased contempt. In a more neutral intergroup setting ascribed fear may very well spark aversive emotions as suggested by the reviewed literature. In the case at hand, however, right-wing populist’s fear may give respondents an opportunity to empathise with an outgroup that is otherwise often portrayed as angry and may thus foster compassion or pity rather than aversion. This could be a strategical reason why right-wing populists often aim to portray themselves as concerned citizens rather than Wutbürger (angry citizens). These findings strongly suggest that social context does matter for the intergroup effects of ascribed emotions, arguably even more so than for relatively automatic emotional contagion or mimicry. Clearly, further research is thus needed, first, to study the effects of ascribed emotions in more neutral intergroup settings and, second, to study intergroup effects of emotion expressions as well as ascriptions by taking into account a wider range of relational settings, discrete emotions and potential intergroup outcomes.

Limitations

There are two major limitations to the present study. First, the suspected floor and ceiling effects could mask potentially existing effects of ascribed emotions on stereotypical evaluations and action tendencies. Less sensitive measurements for the outcomes of interest or experimental paradigms relying on less polarising outgroups could be applied to either discover or fully rule out such effects. Second, the fact that the correction for multiple comparisons using the Bonferroni method renders all effects discussed above statistically insignificant points towards the potential of falsely rejecting true null-hypotheses (type-I-error) because the probability of at least one false discovery per experiment increases once multiple independent hypothesis tests are conducted. However, the conservative Bonferroni correction creates another problem, namely an increased probability to make type-II-errors, in which null-hypotheses are kept despite the presence of effects. Since the present study is pioneering in the research on the effects of ascribed emotions and since we are less interested in the global null hypothesis that all tested effects are insignificant, we decided to qualitatively interpret the individual uncorrected effects, despite the risk of false discoveries. Reproduction studies would be needed to assess the reliability of these results.

Data availability statement

The data that support the findings of this study are openly available in the OSF repository at http://doi.org/10.17605/OSF.IO/U2XZR (Pilot Data) and https://doi.org/10.17605/OSF.IO/P6CFM (Main Dataset).

Disclosure statement

No potential conflict of interest was reported by the authors.

Correction Statement

This article has been republished with minor changes. These changes do not impact the academic content of the article.

Additional information

Funding

This research was co-funded by a grant from the Research Pool of the Department of Political and Social Sciences at the Freie Universität Berlin, Germany.

Notes

1 We use the term recipients to denote those members of audiences or publics to whom an emotion ascription that is made to a third party is communicated.

2 See section “Power Analyses” of the data analysis plan in the supplemental material for details on the sample size determination.

3 The quota targets for age and gender were derived from the most recent Eurostat data at the time of data collection (European Commission, 2021). The quota target for school education was derived from the weighted ESS 2018 (European Social Survey ERIC, 2019).

4 The somewhat problematic sample restriction to binary gender categories is a limitation introduced by the survey provider.

5 This is a common measurement procedure applied, for example, by Forschungsgruppe Wahlen, Mannheim (2019), and several commercial polling institutes like Infratest dimap (2021).

6 The survey questionnaire as reported in the supplemental material contains some additional questions that are not used in the present study but are included to provide a complete picture of the instrument.