Political candidates’ physical appearance and its effects on electoral success have been subject to a number of empirical studies (e.g., Jäckle et al., Reference Jäckle, Metz, Wenzelburger and König2020; Rosenberg et al., Reference Rosenberg, Bohan, McCafferty and Harris1986; Schubert et al., Reference Schubert, Curran and Strungaru2011; Todorov et al., Reference Todorov, Mandisodza, Gorem and Hall2005). Candidate appearance can often be used as an “information shortcut” for making voting decisions (Lau & Redlawsk, Reference Lau and Redlawsk2001). Several studies have highlighted a variety of influential candidate characteristics, including gender (Debus, Reference Debus2017; Huddy & Terkildsen, Reference Huddy and Terkildsen1993), ethnicity (Masters, Reference Masters1994; McDermott, Reference McDermott1998; Pietraszewski, Reference Pietraszewski2016; West, Reference West2017), competent-looking faces (Ballew & Todorov, Reference Ballew and Todorov2007; Todorov et al., Reference Todorov, Mandisodza, Gorem and Hall2005), and even physical height (Murray, Reference Murray2014).
Physical attractiveness—as an overall impression of candidates’ appearance—has been established to have a positive effect on vote shares in different national and electoral contexts (e.g., Jäckle et al., Reference Jäckle, Metz, Wenzelburger and König2020; Rosar et al., Reference Rosar, Klein and Beckers2012). It is important to note that physical attractiveness and its perception is interrelated with these candidate characteristics. In general, physical attractiveness has a “halo effect” on perceptions of a person (e.g., Verhulst et al., Reference Verhulst, Lodge and Lavine2010), whereby other personal characteristics are evaluated more favorably based on perceived attractiveness. As a consequence, attractive individuals receive preferential treatment in social situations, have a higher likelihood of professional success, and are even happier in life than unattractive individuals (Hamermesh, Reference Hamermesh2011; Hamermesh & Abrevaya, Reference Hamermesh and Abrevaya2013). Although these findings are only correlational, it is sometimes assumed that more attractive politicians have a “competition advantage” over their less attractive competitors (Rosar et al., Reference Rosar, Klein and Beckers2012).
Additionally, candidates’ emotional expressions can have an impact on impressions of them (e.g., Gabriel & Masch, Reference Gabriel and Masch2017; Stewart et al., Reference Stewart, Bucy and Mehu2015) and thereby affect electoral success. Here, emotional expressions function as social cues (e.g., Hareli & Hess, Reference Hareli and Hess2012; Van Kleef, Reference Van Kleef2016). According to the ethological framework of emotional expressions (e.g., Sullivan & Masters, Reference Sullivan and Masters1988), emotional displays can be grouped into three categories: happiness/reassurance, anger/threat, and fear/evasion. This classification has been shown to be highly applicable to the study of nonverbal communication among politicians and political leaders (e.g., Bucy & Grabe, Reference Bucy and Grabe2008; Stewart et al., Reference Stewart, Salter and Mehu2009). Emotional displays of happiness/reassurance are often seen as a sign of dominance, are suitable for incumbents (e.g., Sullivan & Masters, Reference Sullivan and Masters1988), and are commonly observed among front-runners during election campaigns (Bucy & Grabe, Reference Bucy and Grabe2008). Because of its attacking character, the emotional expression of anger is a suitable emotion for politicians of the opposition. While it is often connected with negativity and hostility, displays of anger can also be viewed as a sign of caring or empathy if the anger is directed toward a justifiable cause, such as social injustice (cf. Hess, Reference Hess and Parrott2014; Kinder, Reference Kinder, Lau and Sears1986; Masch, Reference Masch2020). Displays of fear/evasion, on the other hand, as well as displays of sadness, should be largely avoided for those who want to obtain power (Schubert, Reference Schubert, Schubert and Masters1991; Stewart et al., Reference Stewart, Salter and Mehu2009).
Research also suggests positive effects of politicians’ smiles on campaign posters (Horiuchi et al., Reference Horiuchi, Komatsu and Nakaya2012). Because of their prevalence in some countries, such as Japan and Germany, campaign posters might serve as mental representations of candidates when voters cast their ballot. In this sense, their presence on campaign posters is particularly relevant when candidate appearance is used by voters as an information shortcut (see Lau & Redlawsk, Reference Lau and Redlawsk2001). Research has also shown that the effect of smiling is dependent on the number of candidates in a voting district; a higher number of district candidates decreases the effect of smiling (Asano & Patterson, Reference Asano and Patterson2018). It should be noted, however, that a branch of research also considers differences in politicians’ smiles (i.e., posed, controlled, or smiles depicting enjoyment, amusement, or contempt) as important for their political success (Stewart et al., Reference Stewart, Bucy and Mehu2015). These subtle differences might not be easily perceived by everyone. Nonverbal differences between smiles attributable to amusement or contempt might be particularly prominent in televised debates and similar interactive formats, whereas campaign posters without verbal context provide a static framework, which makes a broader classification of smiles and happiness feasible and adequate. Furthermore, in countries such as Japan and Germany, district campaign posters typically show portrait pictures of candidates that follow cultural display rules of happiness or neutral expressions. Negative emotional displays such as anger or disgust are generally not displayed on such campaign posters.
While the effects of candidates’ physical attractiveness and candidates’ emotional displays have been studied independently, we aim to explore the relationship between physical attractiveness and facial expressions during election campaigns. Experimental studies have shown that facial attractiveness is predominantly dependent on facial symmetry, with more symmetrical faces perceived to be more attractive. In addition to facial symmetry, advantageous features extend to baby-facedness for women (Cunningham, Reference Cunningham1986) and facial dominance for men (Cunningham et al., Reference Cunningham, Barbee and Pike1990). For both genders, smiling has been linked to higher attractiveness ratings compared with neutral facial expressions (Cunningham, Reference Cunningham1986; Cunningham et al., Reference Cunningham, Barbee and Pike1990; Reis et al., Reference Reis, Wilson, Monestere, Bernstein, Clark, Seidl, Franco, Gioioso, Freeman and Radoane1990). A reciprocal relationship between displays of happiness and physical attractiveness has been discussed in the literature, as experiments show that attractiveness ratings are influenced by the intensity of smiling but also indicate that attractiveness has an impact on happiness ratings (Golle et al., Reference Golle, Mast and Lobmaier2014). Overall, it appears that displays of happiness increase the likelihood of being perceived as physically attractive.
Hence, perceptions of candidates’ physical attractiveness are likely to be positively affected by their positive emotional displays. In particular, displays of happiness on campaign posters, often measured by detecting smiles (e.g., Asano & Patterson, Reference Asano and Patterson2018; Horiuchi et al., Reference Horiuchi, Komatsu and Nakaya2012), might be favorable for attractive and unattractive politicians alike. Because voters might consciously or subconsciously be influenced by a variety of candidate characteristics when processing candidate appearance, the perception of physical attractiveness could be heightened by candidates’ happiness or lessened by negative emotional displays. Such findings would indicate that the perception of physical attractiveness is multidimensional.
We assume that the combined occurrence of happiness and physical attractiveness could lead to an interaction effect whereby happiness has even stronger effects on vote shares for more attractive candidates than for less attractive candidates. Based on the theoretical expectations, we derive three hypotheses:
H1: The more attractive candidates are, the higher their direct vote shares will be.
H2: The happier candidates appear, the higher their direct vote shares will be.
H3: Displays of happiness have stronger effects on direct vote shares for more attractive candidates than for less attractive candidates.
Design
This study focuses on the 2017 German federal election as a single case study with observational data. In Germany’s mixed-member proportional electoral system, voters can cast their vote (1) directly for a candidate and (2) for a party overall. The design is based on a full sample of all voting district candidates from the six most relevant political parties in Germany. This includes all 1,779 candidates from the parties that obtained at least one seat in parliament after the election: Christian Democrats (CDU/CSU), Social Democrats (SPD), Alternative for Germany (AfD), Free Democrats (FDP), the Greens, and the Left. The direct candidates ran for office in 299 electoral districts. In 13 out of 299 electoral districts, one party—the AfD—did not list any candidates because of its nascent history. In two additional electoral districts, the Greens and the Left did not list any candidates.
Measures
The dependent variable consists of direct vote shares for candidates in their electoral districts. This is considered the first vote. Voters also cast a second vote that decides the proportional share of seats for each party in parliament. The personalized proportional vote system determines that candidates with the highest direct vote share enter parliament directly as district representatives (first-past-the-post system). The second vote is typically deemed to be the more important vote. This study focuses on the candidate vote, since it should be most affected by candidate appearance. This reasoning is also in line with previous research focusing mostly on direct votes for candidates and candidate appearance (e.g., Rosar et al., Reference Rosar, Klein and Beckers2008; Rosar et al., Reference Rosar, Klein and Beckers2012; Todorov et al., Reference Todorov, Mandisodza, Gorem and Hall2005). It is important to note that, traditionally, the CDU/CSU and the SPD have by far won most voting districts directly; however, some fluctuation can be observed over the last several years. The direct vote shares are measured in percentages, as reported by official statistics. All key confounding variables, such as party membership and age, are also derived from official and public data (The Federal Returning Officer, 2017).
To measure candidates’ physical attractiveness, we collected photographs of candidates (as seen on campaign posters) during the election campaign in the summer of 2017. The photographs were rescaled, and identifying information such as party logos, pins, and religious symbols were removed. The measurement of physical attractiveness is based on the assumption of an “attractiveness consensus” (Grammer et al., Reference Grammer, Fink, Møller and Thornhill2003; Henss, Reference Henss1992; Rosar et al., Reference Rosar, Klein and Beckers2008), according to which a small number of coders—even a dozen—is sufficient to reliably measure attractiveness (Henss, Reference Henss1992, p. 308). Twenty-four student coders rated the politicians’ physical attractiveness on a 7-point Likert scale ranging from unattractive (0) to attractive (6). Half the coders were male, and half were female; all were between the ages of 18 and 26 and received financial compensation for completing the task. The coders were not informed that the subjects of the portraits were politicians. To ensure that they could not easily identify the subjects of the study as politicians, we included well-known politicians only at the end of the questionnaire of the online rating task. According to the “Truth of Consensus Method” (e.g., Patzer, Reference Patzer1985), attractiveness mean scores were calculated for each candidate. The attractiveness ratings were statistically reliable (Cronbach’s alpha = .95). The overall attractiveness mean score was 1.80; the highest attractiveness rating was 5.33 and the lowest was 0.04.
The displayed expressions were not coded manually; instead, an algorithm for emotion recognition created by Microsoft Azure Cognitive Services was used to detect emotional expressions within each portrait. The algorithm is trained on a deep learning approach and decision rules, such as the Facial Action Coding System (e.g., Ekman, Reference Ekman1997; for further information, see Microsoft Cognitive Services 2020). The emotion recognition algorithm is used to classify expressions of eight distinct states: anger, contempt, disgust, fear, happiness, sadness, and surprise, as well as neutral appearances of candidates. The algorithm provides percentages for each emotional expression in one portrait (ranging from 0 to 1) that correspond to the probability of displaying the given emotion or a neutral expression. The probabilities sum to 1 across all classified emotional and neutral expressions for each picture.
While the emotion recognition algorithm classifies a number of discrete emotions simultaneously, a separate classification for facial features provides additional classifications for smiling. A closer inspection shows that the separate classification for smiling obtains the same results as the emotion recognition algorithm for happiness, indicating that happiness is most likely measured by movement of the lips (instead of true smiling or happiness). Therefore, displays of happiness are limited to the perception of smiles in this study. The algorithm is accessible via the Microsoft face API and has been used in recent studies in political communication (e.g., Boussalis et al., Reference Boussalis, Coan, Holman and Müller2021; Boussalis & Coan Reference Boussalis and Coan2021; Masch, Reference Masch2020). Most of these studies are based on the use of one algorithm. By comparing the Microsoft algorithm with other emotion recognition software—namely, OpenFace and FaceReader—it can be shown that the Microsoft algorithm performs similarly well and accurately with regard to displays of happiness (Masch et al., Reference Masch, Homan and Han2021).
The analysis mainly focuses on happiness as a metric variable and dichotomous variable according to this classification. The probability for happiness has also been recoded as a dummy variable of displaying happiness (1) for pictures with a probability higher than .8. A probability below .8 was coded as not happy (0). The cutoff point was chosen to reflect a high confidence that the images display happiness, and it further aligns with the skewed, almost bimodal, distribution of the variable. To obtain a classification, the pictures have to adhere to a certain pixel size (preferably at least 200 x 200 pixels). Unfortunately, this was not the case in one campaign photo, so that picture was manually coded at a later stage. The results do not change regardless of whether this specific observation is included in the analyses.
Procedure
To adequately reflect the structure of the data, linear hierarchical regression modeling is used, whereby 1,779 candidate observations on level 1 are clustered in 299 electoral districts on level 2. Furthermore, the statistical analysis considers a range of candidate characteristics, including age, gender, nobility titles, academic degrees, immigration background, party affiliation, and whether the candidate is a well-known public figure or previously was a member of parliament. Additionally, the number of candidates in an electoral district is considered in order to reflect the degree of competitiveness in a district. These factors have been deemed potentially confounding factors in previous research (e.g., Gassner et al. Reference Gassner, Masch, Rosar, Schöttle, Korte and Schoofs2019; Rosar et al. Reference Rosar, Klein and Beckers2008; Rosar et al. Reference Rosar, Klein and Beckers2012). The models further differentiate between candidates from East and West Germany for two political parties—the Left and the AfD—in order to reflect path dependencies of historical, political, and social variations that still lead to noticeable differences in electoral results. This is particularly the case for candidates of the two parties. The Left was established in 2007 by a merger between the Party of Democratic Socialism (PDS) and Labour and Social Justice – The Electoral Alternative (WASG). The PDS was the successor of the Socialist Unity Party of Germany (SED)—the ruling party of the German Democratic Republic—and as its descendant the Left has historically higher vote shares, including direct mandates, in East German voting districts. As a relatively new party, the AfD has gained a stronger foothold and support in East German states.
Results
Figure 1 displays the distributions of each classified emotion according to the Microsoft algorithm for perceived emotion recognition. The majority of pictures have a high likelihood of displaying happiness (smiles), while negative emotions such as anger, contempt, disgust, fear, and sadness show very low probabilities. These negative emotions as well as the emotion of surprise are therefore negligible when it comes to emotional expressions on campaign pictures in this case. Next to happiness, neutral expressions are the only other classification that can be empirically observed with high confidence. This overall pattern conveys a high face validity given the nature of the campaign posters.
Figure 1. Emotion classifications of candidates’ faces in the German federal election 2017. Figure displays the frequencies of classification results based on the Microsoft emotion detection algorithm for 1,778 portrait pictures of direct candidates.
A more detailed distribution of happiness in the candidate sample shows differences in physical attractiveness and happiness across candidates’ party affiliations. Candidates running for mainstream parties (CDU/CSU, SPD, FDP, the Greens) are not only perceived as more attractive physically than candidates of populist political parties (the Left, the AfD), they also display higher degrees of happiness (see Table 1). On average, the incumbents and likely front-runners of the SPD and CDU/CSU show the highest probabilities of happiness on their campaign posters compared with candidates of smaller parties. Candidates for the SPD, the FDP, and the Greens appear to be perceived as particularly physically attractive, while candidates of the Left and the AfD are perceived as noticeably less attractive on average.
Table 1. Descriptive statistics of physical attractiveness and happiness according to party membership.
Note: Table displays mean values and standard deviations in parentheses for attractiveness (ranging from 0 to 6) and happiness (ranging from 0 to 1). Two separate ANOVA show significant differences for these two variables across party membership (p < .001; two-tailed).
The findings indicate a positive effect of candidates’ physical attractiveness on direct vote shares (H1) when controlling for relevant candidate characteristics (see Model 1 in Table 2). The results further indicate a statistically significant effect of physical attractiveness on direct vote shares that remains when controlling for happiness (see Model 2 and Model 3 in Table 2). Both measurements of happiness show a positive effect on direct vote shares (H2). Additionally, a moderated relationship between happiness and physical attractiveness (H3) occurs when testing for an interaction between physical attractiveness and displays of happiness, whereby smiling candidates and more attractive candidates receive a higher degree of direct vote shares compared with others (see Models 4 and 5 in Table 2; see also Figure 2). Both interaction effects are statistically significant at the 10 percent level.
Table 2. Hierarchical regression of direct vote shares on physical attractiveness and happiness.
Note: Table 2 displays unstandardized regression coefficients. Standard errors in parentheses.
† p < .1; * p < .05; ** p < .01; *** p < .001 (two-tailed).
Attractiveness * Happiness (metric), tolerance = .10, VIF = 19.34; Attractiveness * Happiness (dichotomous), tolerance = .05, VIF = 10.26.
Figure 2. Interaction between physical attractiveness and happiness on direct vote shares. Figure displays the predicted direct vote shares based on a marginal effect plot with a 95% confidence interval according to happiness (dichotomous) and physical attractiveness.
Discussion
In line with previous research, displaying smiles on campaign posters has a positive impact on political success. The findings indicate that not only is displaying happiness favorable for candidates overall, it is especially beneficial for attractive politicians. The results show that physical attractiveness has a consistent positive effect on vote shares that increases slightly when candidates display happiness. While controlling for displays of happiness, the effects of attractiveness persist, underlining the importance of candidates’ physical appearance for their electoral success.
Previous research has put forward the idea that “beauty is beastly” and may come with a price, whereby a high level of attractiveness can be disadvantageous, especially for women (Heilman & Saruwatari, Reference Heilman and Saruwatari1979; Lizotte & Meggers-Wright, Reference Lizotte and Meggers-Wright2019). This effect can be emphasized for attractive female politicians who are smiling because of gendered “display rules” of emotions and paternalistic stereotypes (Eagly & Johannesen-Schmidt, Reference Eagly and Johannesen-Schmidt2001; Renner & Masch, Reference Renner and Masch2019). Therefore, both attractiveness and smiles could lessen perceptions of competence and therefore decrease vote shares. Hence, moderating effects of gender have been tested in this analysis, but the results do not show any significant effects (see Table A in the Appendix). However, it has to be noted that the sample size of very attractiveness female candidates is fairly small, and therefore this study might be limited in detecting the “beauty is beastly” effect.
In addition, it is necessary to consider ethnicity and race when studying the effects of candidate appearance (e.g., West, Reference West2017), since voters typically attribute cues to it, such as representation and advocacy for a minority. However, one limitation of this study lies in its sample size, as it does not allow for detailed testing of moderating effects for race and ethnicity. The analysis for political candidates in Germany only considered immigration background as a confounding variable. This is due to the fact that the sample of German candidates in 2017 is not diverse enough to study politicians’ ethnic backgrounds and race in a quantitative manner. Gender, ethnicity, and race are relevant candidate characteristics that should be explored more closely in further research, especially with respect to stereotypical perceptions. One candidate characteristic that does not seem to moderate the effects of attractiveness is age. An interaction between attractiveness and age is not significant based on these data. By analyzing the effects of attractiveness for each age group as specified in Table 2, a consistent effect emerges across age groups (see Tables B, C, D, E, and F in the appendix).
In candidate-centered politics, voting decisions are inherently based on choices between candidates. As a result, perceptions of candidates during election campaigns are often formed in relative terms—that is, candidates are evaluated in comparison with their opponents (Rahn et al., Reference Rahn, Aldrich, Borgida, Sullivan, Ferejohn and Kuklinski1990, p. 155). Hence, the findings in Table 2 were replicated with relational measures of physical attractiveness and happiness for each candidate compared with other candidates in their electoral districts, as suggested by Rosar et al. (Reference Rosar, Klein and Beckers2008). The results indicate that the findings are robust when using a relational measure of attractiveness (see Table G in the Appendix), since relative physical attractiveness and happiness are associated with higher vote shares. A relational measure of happiness is not linked to higher vote shares (see Tables H and I in the Appendix). Therefore, in the case of single-shot campaign posters, displays of happiness are likely to be beneficial for candidates regardless of their opponent’s behavior. However, it can be assumed that relational effects of emotional expressions emerge more strongly when the range of displayed emotions increases—that is, when anger or disgust is displayed alongside happiness and neutral expressions in other political settings.
Based on the nature of campaign posters in Germany, this study is limited to the effects of happiness compared with other emotional expressions such as anger and disgust. Effects on vote shares could vary greatly with regard to different emotional expressions. The perception of emotional expressions is highly context-dependent (Barrett et al., Reference Barrett, Mesquita and Gendron2011; Hess et al., Reference Hess, Dietrich, Kafetsios, Elkabetz and Hareli2020), and it has been shown that politicians’ emotional expressions are dependent on displays of other politicians (Masch, Reference Masch2020, p. 237). Therefore, a variety of emotional expressions and their interplay with each other and with candidate characteristics, such as physical attractiveness, should be considered in the future.
Overall, it seems worthwhile to expand measurements of attractiveness to capture physical attributes of attractiveness beyond mere facial features and distinguish them from each other; ideally, this could be undertaken in experimental settings. As measurements of physical attractiveness are predominantly based on a consensus of attractiveness, it is necessary to clarify this measure of facial attractiveness conceptually. Our reasoning for this is as follows: the effects of physical attractiveness are continuously present in research on candidate appearance and electoral success, but there seem to be differences between levels of physical attractiveness according to party affiliation, whereby politicians of mainstream parties (CDU/CSU and SPD) and especially of niche parties with a high-income voting clientele (FDP and the Greens) are rated as more attractive than politicians of fringe or even populist parties (the Left and the AfD).
This finding could be caused by several factors. First, it could be that mainstream parties have a higher degree of professionalism when campaigning (e.g., professional photo shoots), which may enhance professional appearance and candidate expressions on campaign posters. Second, more attractive politicians might self-select into more successful and mainstream political parties. Because of lifelong advantages well into adulthood, attractiveness could function as a social advantage in candidates’ biographies. Based on these individual experiences, attractive individuals may be satisfied with the status quo (Peterson & Palmer, Reference Peterson and Palmer2017). As a result, they may have more incentives to maintain the social, political, and economic status quo and therefore participate in mainstream political parties. However, politicians who did not experience those advantages based on their looks might be inclined to challenge the status quo, and consequently engage in populist and antiestablishment political parties. Third, belonging to antiestablishment parties could also encourage politicians to signal otherness by looking and dressing differently than the elites: untraditional and “roguishly.” Such codes contradict the idea of a single beauty norm within a society and, therefore, also contradict the measurement of an attractiveness consensus. Lastly, the perception of physical attractiveness measured by the consensus could also reflect social class attributes, wealth, and status since candidates’ physical attractiveness clearly corresponds to (median) income levels of their parties’ voting clientele.
Given that physical attractiveness combines many visible candidate characteristics, and its perception could even be influenced by emotional displays (e.g., heightened by happiness), it is necessary to study the perception of physical attractiveness and its consequences more closely in an experimental manner. Observational studies are limited to a certain number of candidate characteristics and their potential interactions without enabling researchers to analyze their causal relationship. As the presented models already indicate some concerns regarding potential issues of multicollinearity, observational studies cannot simply be expanded to include a larger number of candidate attributes. This is particularly true, given that ethnicity, gender, and race influence whether citizens run for office. Changing the design to thoughtfully planned experiments could show which specific visible attributes increase favorability, all else being equal.
Conclusion
The findings of the present study show that candidate appearance is crucial for electoral success. They also highlight the need to study factors that determine physical attractiveness in light of distinct facial features such as dominance (Laustsen & Petersen, Reference Laustsen and Petersen2020) and competence (Todorov et al., Reference Todorov, Mandisodza, Gorem and Hall2005), as well as their interaction with visible attributes of social status and wealth that are part of any immediate overall impression. It is further of relevance to consider the effects of physical attractiveness ratings with regard to specific emotional expressions. Positive effects of physical attractiveness on vote shares remain present when controlling for displays of happiness. There is also some support for the assumption that more attractive candidates profit the most from smiling on campaign posters. While the results support all three hypotheses, the analysis is only based on observational data and the investigated links are therefore strictly correlational. The reported associations between physical attractiveness, displays of happiness, and vote shares necessitate further experimental research for the purpose of investigating the underlying relationship between varying visible candidate attributes, their interactions, and vote shares. Future experiments are required in order to analyze the causal relationship between physical attractiveness and emotional expressions and to disentangle the causal mechanisms between politicians’ emotional expressions and their physical attractiveness on their electoral success.
Supplementary Materials
To view supplementary material for this article, please visit http://dx.doi.org/10.1017/pls.2021.17.
Open access
