Unfolding political attitudes through the face: facial expressions when reading emotion language of left- and right-wing political leaders

Spontaneous emotionally congruent facial responses (ECFR) to others’ emotional expressions can occur by simply observing others’ faces (i.e., smiling) or by reading emotion related words (i.e., to smile). The goal of the present study was to examine whether language describing political leaders’ emotions affects voters by inducing emotionally congruent facial reactions as a function of readers’ and politicians’ shared political orientation. Participants read sentences describing politicians’ emotional expressions, while their facial muscle activation was measured by means of electromyography (EMG). Results showed that reading sentences describing left and right-wing politicians “smiling” or “frowning” elicits ECFR for ingroup but not outgroup members. Remarkably, ECFR were sensitive to attitudes toward individual leaders beyond the ingroup vs. outgroup political divide. Through integrating behavioral and physiological methods we were able to consistently tap on a ‘favored political leader effect’ thus capturing political attitudes towards an individual politician at a given moment of time, at multiple levels (explicit responses and automatic ECFR) and across political party membership lines. Our findings highlight the role of verbal behavior of politicians in affecting voters’ facial expressions with important implications for social judgment and behavioral outcomes.

Seeing and reading about someone's smiling or frowning automatically elicits emotionally congruent facial responses (ECFR) in the observer or reader and this effect appears largely unconscious and difficult to suppress [1][2][3][4] . For instance, within 500 ms milliseconds after seeing a smiling face there is an activation of the zygomaticus major, the muscle that lifts up the corners of the mouth forming a smile. Similarly, when individuals see a frowning face they react with the activation of the corrugator supercilii, the muscle which draws the eyebrows together forming a frown. Scholars have explained such spontaneous facial congruent activity with the notion of 'facial mimicry' and 'embodied sensorimotor simulation': when individuals see a facial expression, they construct a simulation of the observed expression in (pre)motor and somatosensory brain networks which would reactivate related concepts, affective states, and autonomic and behavioral changes [5][6][7] . Embodied simulations could result in emotionally congruent facial muscle activity in the perceiver and contribute to emotion recognition, perspective taking and empathy 8,9 . Remarkably, a growing body of research has shown that these seemingly automatic processes are also context-dependent and sensibly modulated by social information (for reviews see [10][11][12]. Indeed, ECFRs are stronger towards members of one's own social group and reduced or even incongruent towards members of other social groups [13][14][15][16] . They are more likely for social targets eliciting positive attitudes 17 and people who like each other tend to have enhanced ECFR to one another's expressions, which in turn enhances one's feelings of liking and affiliation [18][19][20] . On the contrary, for disliked individuals and those we are in competition with, a complete lack of congruent facial reactions, or even incongruent reactions have been observed 21,22 . This body of evidence can be interpreted in light of neuroscience research showing an early processing of socially and emotionally relevant cues within a few milliseconds of exposure 12,[23][24][25] . Hence, ECFRs likely reflect the late peripheral correlates of an

Results
We measured ECFR using facial EMG recording in 28 left-wing and 25 right-wing young voters. Data were collected between January and October 2013, a few months prior to Matteo Renzi's confirmation at the head of the major left-wing party (end of 2013) and the subsequent unprecedented electoral support he rallied in the Italian elections for the European Parliament (March 2014).
Hypothesis 3 (intergroup differences in ECFR) was supported only for left-wing participants, who frowned more in response to negative emotion expressions of left-(ingroup) relative to right-wing politicians (outgroup), (p = 0.002). No significant difference in CS activity was found for positive emotion expressions attributed to ingroup relative to outgroup politicians, (p = 0.32).
With regard to Hypothesis 3, left-wing participants did not show significantly different zygomaticus activation when reading positive emotion expressions of ingroup relative to outgroup politicians, (p = 0.59). No significant difference emerged for negative emotion verbs attributed to ingroup and outgroup politicians, (p = 0.10). Partially confirming results were found for right-wing participants who showed a marginally significant difference in their smile response to positive emotion expressions of ingroup relative to outgroup politicians, (p = 0.085). No significant difference emerged between negative verbs attributed to ingroup and outgroup politicians, (p = 0.92).

Physiological data (ECFR): supplementary analysis.
The fact that the predicted intergroup differences in ECFR (Hypothesis 3) were supported only for left-wing but not right-wing participants was somehow unexpected. Considering the results on political attitudes towards individual politicians and voting intention scores which surprisingly showed a consistently positive evaluation of left-wing leader Renzi among right-wing participants, we reasoned that such favorable attitudes could be reflected in the lack of significant differences in ECFR to ingroup and outgroup politicians among right-wing participants. We thus conducted further participant political orientation × political party × valence × linguistic category ANOVAs in order to explore this possibility. In these analyses left-wing politician Renzi was removed from the left-wing political group. We expected that dissociating participants' responses to Renzi from the analysis should boost ingroup and outgroup differences in the facial muscle response of right-wing participants.
In line with Hypothesis 3, the CS activation of left-wing participants was higher when reading negative emotion expressions of the ingroup politician compared to outgroup politicians, (p = 0.004). The same effect was www.nature.com/scientificreports www.nature.com/scientificreports/ found for right-wing participants, although the difference was only marginally significant, (p = 0.078). No significant difference in CS activity was found for positive verbs attributed to ingroup and outgroup politicians for both left, (p = 0.52), and right-wing participants, (p = 0.93). See Supplementary results, S2 for further analysis comparing participants' facial reactions to Renzi with those of other right-wing politicians by dissociating the data of the other left-wing politician, Bersani.
The participant political orientation × valence of emotion expression × political group interaction was significant, (F(1, 52) = 5.46, p = 0.023, η 2 = 0.095). In line with Hypothesis 2, when left-wing participants read emotion verbs attributed to the ingroup politician, their ZM activity was higher in response to positive (0.11 ± 0.03) than negative (0.03 ± 0.02) emotion verbs, (p = 0.011). However, their ZM response did not show difference for positive emotion verbs attributed to the ingroup politician than outgroup politicians (0.12 ± 0.05), (p = 0.815). As expected, no significant difference was observed when left-wing participants read positive and negative (0.13 ± 0.05) emotion verbs attributed to outgroup politicians, (p = 0.81). Furthermore, left-wing participants displayed an incongruent facial reaction as they smiled more when reading negative emotion verbs attributed to outgroup politicians than the left-wing politician, (p = 0.013).
The ZM response of right-wing participants was higher when reading positive (0.12 ± 0.05) than negative (0.06 ± 0.05) emotion verbs of ingroup politicians, although the difference was marginally significant, (p = 0.071). No difference was found between positive (0.04 ± 0.03) and negative (0.03 ± 0.02) emotion verbs attributed to outgroup politicians (p = 0.73). As expected, right-wing participants showed greater activation of the ZM when reading positive emotion verbs attributed to ingroup relative to outgroup politicians, (p = 0.028). No significant difference in the zygomaticus response was found for negative emotion verbs attributed to ingroup vs. outgroup politicians, (p = 0.37).
Overall, the results obtained when the responses to Renzi were removed from the left-wing political group fully confirmed Hypothesis 2 for both left-and right-wing participants, who displayed congruent facial reactions in response to ingroup politicians only. Results of right-wing participants also supported Hypothesis 3 by showing that their facial responses to emotional verbs of politicians belonging to the left and right political orientation were different. Conversely, left-wing participants, as expected from this specific analysis, showed a similar activation of the smile muscle in response to positive expression verbs of target politicians independent of their political orientation. Interestingly, though, they also smiled in response to negative emotion expressions of right-wing politicians, which is clearly an incongruent facial reaction notably linked to a counter-empathic response 14   www.nature.com/scientificreports www.nature.com/scientificreports/

General Discussion
The present study uniquely highlighted ECFR to emotion language of political leaders and a modulating role of political orientation and attitudes towards target politicians. Importantly, through the integration of behavioral and physiological measures we were able to capture political attitudes consistently across different measures and at varying degrees of awareness. By measuring automatic facial muscle activation, we found that reading verbal stimuli that describe political leaders' emotions produces ECFR in the readers, such that positive expressions elicit "smiles" whereas negative expressions elicit "frowns" (Hypothesis 1). Remarkably, our findings highlighted a strong ingroup selectivity of ECFR (Hypothesis 2), as reading about politicians' positive and negative emotions yielded ECFR in readers for ingroup but not for outgroup politicians. Furthermore, we found intergroup differences in ECFR (Hypothesis 3) for left-wing subject, who showed a greater frowning response when reading negative emotion expressions of politicians belonging to the left compared to the right political orientation. Conversely, on a first analysis, right-wing participants showed an undifferentiated smiling and frowning response to positive and negative emotion expressions of right-and left-wing politicians. Taken together, our findings support the idea that ECFRs are context-dependent and sensibly modulated by social information [23][24][25][26][27] . However, to better grasp the meaning of these results, we considered findings from explicit measures on political attitudes and voting intention, which consistently indicated that right-wing participants held favorable attitudes towards left-wing politician Renzi despite they clearly identified him as a left-wing political leader. So much so that when it came to their political attitudes and inclination to vote, they did not differentiate between ingroup (right-wing) politicians and left-wing leader Renzi. On this basis, we reasoned that favorable attitudes towards Renzi among right-wing participants could have obscured differences in their facial response to ingroup and outgroup politicians' expressions. Supplementary analysis where we compared facial reactions to emotion language of right-and left-wing politicians by dissociating Renzi from the left-wing group revealed that language-based ECFR of right-wing participants were indeed enhanced for emotion expressions of ingroup but not outgroup politicians. Most likely, positive attitudes towards Renzi concealed the hypothesized intergroup differences in facial simulation of right-wing participants. To put it differently, our findings revealed a "favorite political leader effect" according to which attitudes toward an individual political leader, more than mere left vs. right political orientation, affected embodied simulation of emotion language.
The present study builds upon and expands previous research showing that psychophysiological markers of sensory, emotional and motor reactivity are sensitive to social influences 12,[23][24][25][26][27]31 . Our findings support the view that integration of contextual and social information occurs very rapidly 12,[23][24][25][26][27]31 and affects the magnitude of a relatively late phenomenon like ECFRs to linguistic stimuli. Moreover, our study complements applied research on the social moderators of ECFR, political attitudes and voting behavior. First, we highlighted that voters' ECFR to language describing politicians' emotion expressions could signal political attitudes that go beyond explicitly expressed political orientation. Our results revealed that political attitudes sensibly modulated ECFRs, as shown by the fact that political leaders singularly favored by left-and right-wing electorates triggered similar facial effects. Renzi's appeal as a political leader throughout 2013 was steadily growing thanks to his ability to connect to voters beyond the party divide [32][33][34][35] . Our data were collected only a couple of months prior to his confirmation at the head of the major left-wing party (end of 2013) and the subsequent unprecedented electoral support (more than 40%, the best result ever reached by his political party) he rallied in the Italian elections for the European Parliament (March, 2014). By measuring automatic ECFR, we could therefore gauge on the general tide of positive attitudes that was actually building up in favor of Renzi, who emerged in that particular moment and in that particular context as the political leader able to move audiences beyond the logic of left and right, building a wide consensus that flowed into actual electoral support.
Moreover, we showed for the first time that language describing politicians' emotions and the political context in which such stimuli are embedded affect spontaneous subtle non-verbal behaviors such as ECFR 10 . Going beyond previous evidence on the social and political factors affecting ECFR 11 , we clearly found facial sensitivity to political leaders' emotion expressions even when such information is conveyed through linguistic material: what people read about politicians, being in newspaper or in social media, has the power to affect their facial reactions the same as when viewing politicians' happy or angry faces. This can have important implications for eliciting congruent social judgments 4,36 and for behavioral outcomes especially as they relate with voting behavior. Recent research has shown that automated facial expressions analysis could be used to predict with good accuracy voting behaviors 37 . Our findings suggest that ECFR to written media and reports on candidates' emotional behaviors could also serve as indices of embodied political attitudes with potential implications for predicting voting decisions 38 .
One potential limit of the present study is the use of a self-reported one-dimensional scale of political orientation. There is a considerable debate on the validity of such a measure to capture the political ideology of electorates (for a review, see 39 ). Nevertheless, it should be noticed that the traditional self-identification measure on a single scale performs quite well as a predictor of political behavior (e.g., in predicting the vote 33,40,41 ). Furthermore, a great body of research has demonstrated the utility of unidimensional ideological axes (leftright) as the main political orientation instrument used by individuals to navigate in the Italian political context, at least until the moment of data collection 34,42 . Thus, we could consider such a measure as valid for the purpose of the present study. Future research should consider whether our evidence could be confirmed using more complex measures of political orientation and ideology.
Another potential limit of the present study is related to the fact that we could not provide a clear-cut explanation of why left-wing participants did not differentiate their smile response to positive expressions of ingroup and outgroup politicians. This result is in line with studies on the effects of group membership on facial mimicry 13,15 , suggesting that smiling in response to expressions of opponents incurs low social costs in the political intergroup context. Interestingly, we found an incongruent facial activation pattern instead, that is smiles in response to negative expressions of outgroup politicians. Given that ZM activation is also observed when individuals report www.nature.com/scientificreports www.nature.com/scientificreports/ feeling negative emotions, it is possible that such a pattern reflects an instance of 'schadenfreude' or counter empathic response 21,43 probably due to morally tainted attitudes of left-wing participants towards particularly controversial Italian right-wing politician Berlusconi, who is especially despised among Italian left-wing voters 44 . This explanation is in line with the higher liking of left-wing participants of negative verbal expressions of outgroup politicians (see Supplementary results, S1). Future studies should test this possibility and also examine whether participants' smiles express genuine enjoyment or represent a reaction related with negative emotions by measuring the simultaneous activation of the ZM and the orbicularis oculi (OO) muscle, which is supposed to be activated only when a smile is associated with truly felt enjoyment 6,45 .
Lastly, our sample was not balanced in terms of gender and we cannot exclude that gender may have had an impact on language based ECFR. It should be noted however that previous facial EMG research 42 has not evidenced significant differences in terms of gender (e.g., for instance, it has been found that levels of facial muscle activation are more pronounced in females compare to males, but no significant differences are reported in terms of facial pattern congruency between male and female subjects). Future research should investigate language triggered ECFR using a more gender balanced sample.

conclusion
This research is the first to show that reading about emotion expressions of ingroup political leaders elicits greater ECFR than reading about expressions of outgroup politicians. We also showed that, beyond shared political orientation, such facial reactivity reflects political attitudes toward individual leaders which could flow into actual electoral support. Going beyond previous research, and through the integration of behavioral and psychophysiological methods we were able to consistently tap on a 'favored political leader effect' whereby political attitudes towards an individual politician were sensibly captured at a given moment of time, at multiple levels of awareness (explicit responses and automatic facial reactions) and across political party membership lines. ECFR to language describing politicians' emotions proved a highly sensitive and informative measure, as it was able to reveal, at a given time the political attitudes and voting intentions of the electorate. It should also be noted that the effect we found was closely related to a well-defined moment in time, depicting the glory of a political leader: sic transit gloria mundi, very likely so do ECFRs toward formerly glorious leaders.

Method
Participants. Fifty-four (43 females; 22 ± 1.42) undergraduate Italian students of the University of Bologna participated in the experiment for academic credits. Of these, 28 reported being of left-wing and 25 of rightwing political orientation. Participants were selected on the basis of a pre-test in which we asked 170 students to indicate their political orientation with the following item "In politics people talk of left and right. Using this card, where would you place yourself on this scale, where 1 indicates the left, 4 indicates center and 7 indicates the right?" (see 33,46 ). This item, is routinely administered in multinational surveys (e.g., European Social Survey and European Values Survey), is as a valid and reliable measure of left-right political orientation 40,46 and has been the basis of several political psychological studies in the Italian context 35,42,47 . According to recent studies, self-placement on a left -right dimension has been the primary criterium utilized by the Italian electorate to perceive, categorize, and evaluate candidates, parties and issues in political terms 34,42,48 . Participants who reported clear-cut political attitudes in the pre-test (N = 62), that is those who ticked below '3' (left-wing political orientation), and those above '5' (right-wing political orientation) were solicited to take part in the study. Among them, 54 (87%) agreed to participate. Prior to the start of the experiment, they read the ethical approval statement and filled out the informed consent form. Data were collected between January and October 2013.
Linguistic stimuli. The stimulus material consisted in 12 subject-verb sentences where the subject was either a left-or right-wing Italian politician and the verb referred to either positive or negative emotion expressions (e.g., 'Renzi smiles' , 'Berlusconi frowns'). Politicians were selected through a pretest in which 30 undergraduate students were asked to write down the names of the first right-and left-wing national politicians who came to their mind. Four politician names were selected on the basis of the highest frequencies of citation. Bersani (24 citations) and Renzi (21 citations) who belonged to the main Italian left-wing political party (Democratic Party), and Berlusconi (26 citations) and Alfano (23 citations), belonging to the main right-wing political party (People of Freedom) and representing the political opposition at the time. It should be noted that both parties were born and defined as moderate (or center-left and center-right) parties, so much so that more extreme left and right parties do exist in the Italian context. The verbs used in the sentences were derived from Fino et al. 2 and matched for word length and frequency in language use (see 2 ). Half of them were action verbs and directly refer to the action of smiling or frowning, and the other half were state verbs, referring to more abstract emotional states 49 (see Supplementary material).

Procedure and measures.
Participants were asked to participate in a study on language of politics. In order to conceal the real purpose of the study they were told that skin conductance levels would be recorded via sensors placed on the face during a reading task. The computer task consisted in sequential presentation of verbal material on a monitor and subsequent evaluation of stimulus likeability by right hand clicking on a 5-point scale (1 = I don't like it at all; 5 = I like it very much). The scale appeared on the monitor after each stimulus sentence and disappeared upon response (see Supplementary results S1 for results on liking of politicians' linguistic expressions).
After the computer task, participants completed a questionnaire including manipulation checks, political attitudes and voting intention. They were first asked to indicate their political orientation on the same item used in the pre-test. As manipulation checks, participants rated the valence of each stimulus sentence (1 = very positive; 7 = very negative), indicated the political affiliation of each politician by choosing between two options (left or right), and rated how much each target politician represented the political wing they had indicated in the www.nature.com/scientificreports www.nature.com/scientificreports/ previous question (1 = not at all; 7 = very much). Then, we assessed participants attitudes towards target politicians by means of a series of items aimed to tap into the affective and cognitive components of attitudes which have been shown to operate as partially distinct determinants of political candidate evaluation and subsequent voting behavior 50,51 . Items referred to the cognitive and affective components of participants' political attitudes were measured by asking "How much does target politician represent your political ideas?", "How much do you like target politician?", "Would you be happy if target politician won in the coming elections?". Given the strong correlations between the answers (Bersani's α 0.87; Berlusconi's α = 0.94; Alfano's α = 0.91; Renzi's α = 0.82), they were averaged into four variables representing participants attitudes toward each politician. The behavioral component of the attitudes was measured by asking participants on their voting intention for each target politician in the coming election (1 = not at all; 7 = very much). When asked about their ideas regarding the purpose of the experiment, no participant revealed awareness of the hypotheses, and none suspected that facial muscular activation was being measured. All participants reported normal or corrected-to-normal vision.
Data acquisition and preprocessing. Facial muscle activity was measured using miniature Ag/AgCl (10/4 mm) surface electrodes attached over the left CS (brow) and ZM (cheek) muscle regions according to Fridlund and Cacioppo's guidelines 52 . The EMG raw signal was measured with a Biopack MP36 (Systems Inc.), was pass filteredn online with a 20-250 Hz passband and a 50 Hz notch filter and was full-wave rectified offline. Stimuli were presented in a random order with E-prime software (Psychology Software Tools, Pittsburgh, PA) on a monitor located approximately 1 m from the participant. Each trial was composed of a central fixation cross (duration 2000 ms), followed by the presentation of the target stimulus (3000 ms), an evaluation scale (average duration ~4 seconds) and blank screen (1000 ms). The inter-stimulus interval was ~10 seconds allowing sufficient time for facial muscles to relax after stimulus presentation. For each participant three blocks of stimuli were presented with each block containing all the 12 emotion expression verbs and 6 neutral verbs attributed to all politicians in random order, resulting in 72 stimuli presentations altogether. In order to familiarize with the task, participants completed 10 practice trials at the beginning of the experiment.
EMG values in millivolt (mV) were expressed as difference in activity from the pre-stimulus level, defined as baseline (for a similar procedure see 13,17,53 ). For each trial and muscle, baseline EMG level was calculated as the mean signal computed during the 1000 ms preceding the sentence stimulus. Then, facial EMG response to target stimuli was computed as the mean EMG signal for 3000 ms of stimulus exposure, which was then baseline corrected. Before statistical analysis, EMG data were collapsed over the 18 trials with the same emotional expression and linguistic category for left-and right-wing political party. Data analysis and statistics. All statistical analyses were conducted using SPSS software (version 21.0).
Group comparisons were performed through conventional t tests (or Wilcoxon Rank Sum for non-normal distributions) and repeated measures analysis of variance (ANOVA). We carried out 2 (participant political orientation: left-wing, right-wing) × 2 (political party: left-wing, right-wing) × 2 (valence of emotion expression: positive, negative) × 2 (linguistic category: action verbs, state verbs) ANOVAs with repeated measures on the last three factors on EMG data for the CS and ZM muscles separately. The Greenhouse-Geisser correction for degrees of freedom was applied when the sphericity assumption was violated. For cases where repeated measures ANOVA revealed a significant interaction effect, Bonferroni post hoc analyses were used to examine the specific comparisons and significance levels.
Ethical approval. All participants signed written informed consent. The experiment was conducted in accordance with relevant guidelines and regulations. The study protocol was approved by the Institutional Review Board of University of Bologna.

Data availability
The datasets generated during and/or analyzed during the current study are available in Microsoft Excel format (Supplementary Dataset 1).