2 Theory

Ambady and Rosenthal (Reference Ambady and Rosenthal1992) show that many people engage in ‘thin-slicing’ when making decisions. They make sense of the world through thin moments, or narrow windows, of personal experience. This often means that they make conclusions based on minimal amounts of information. When it comes to interpersonal interactions, this often means that people make judgements about others – their values (Schmidt, Reference Schmidt2007), their economic status (Kraus and Keltner, Reference Kraus and Keltner2009), their psychology (Naumann et al., Reference Naumann, Vazire, Rentfrow and Gosling2009) – based on how they appear (Bar et al., Reference Bar, Neta and Linz2006).

With this tendency in mind, we should expect that face masks influence public perceptions of politicians for at least two reasons. One has to do with how facial coverings distort how people perceive the emotional states of those who wear them. Prior work shows that both the upper and lower halves of the face, especially the eyes and the mouth, convey emotions (Blais et al., Reference Blais, Roy, Fiset, Arguin and Gosselin2012; Eisenbarth and Alpers, Reference Eisenbarth and Alpers2011; Nusseck et al., Reference Nusseck, Cunningham, Wallraven and Bülthoff2008; Schurgin et al., Reference Schurgin, Nelson, Iida, Ohira, Chiao and Franconeri2014; Smith et al., Reference Smith, Cottrell, Gosselin and Schyns2005; Wegrzyn et al., Reference Wegrzyn, Vogt, Kireclioglu, Schneider and Kissler2017). People tend to recognise others’ happiness by reading mouth expressions, while they read others’ eyes for evidence of sadness, disgust, or fear (Wegrzyn et al., Reference Wegrzyn, Vogt, Kireclioglu, Schneider and Kissler2017). One consequence of this is that when people can only see the upper part of a person’s face, viewers are much more likely to perceive that the object of their sight is expressing negative emotions instead of positive emotions (Fischer et al., Reference Fischer, Gillebaart, Rotteveel, Becker and Vliek2012). In line with this, on the one hand, studies have shown that some facial coverings, particularly Islamic veils or headdresses, cause individuals to perceive their wearers as feeling or expressing more negative emotions than they actually do (Kret and de Gelder, Reference Kret and de Gelder2012; Kret and Fischer, Reference Kret and Fischer2018). On the other hand, some work has found a minimal or null effect of facial coverings and perceptions of emotional states. Calbi et al., (Reference Calbi, Langiulli, Ferroni, Montalti, Kolesnikov, Gallese and Alessandra Umiltà2021), for example, shows that people correctly recognised the individual facial expressions and emotions of Italian women, even when the lower part of their faces was covered with a mask or scarf.Footnote ² While the evidence is somewhat mixed on the effect of face coverings on public perceptions of wearers’ emotional states, it seems to suggest on balance that masks distort these perceptions in a negative way. This is important for our study because voters sometimes rely on perceptions of politician’s emotional states when deciding for whom to vote, particularly when they have little other information about the candidates, typically favouring more positive states to negative ones (Koo, Reference Koo2020).

A second reason why face masks might influence public perceptions relates to how they change the perceived attractiveness of their wearers. An expansive, cross-disciplinary literature on the ‘beauty effect’ (Biddle and Hamermesh, Reference Biddle and Hamermesh1998) suggests that how people see you plays a key role in a wide range of life outcomes, including economic success (Mulford et al., Reference Mulford, Orbell, Shatto and Stockard1998) and psychological well-being (Gupta, Etcoff, and Jaeger, Reference Gupta, Etcoff and Jaeger2016). Looks matter in politics, too. We know that candidates who appear more attractive are more likely to win elections (Atkinson, Reference Atkinson2009; Albright et al., Reference Albright, Dong, Fang, Malloy, Kenny, Winquist and Yu1997; Ballew and Todorov, Reference Ballew and Todorov2007; Banducci et al., Reference Banducci, Karp, Thrasher and Rallings2008; Berggren et al., Reference Berggren, Jordahl and Poutvaara2010; Hayes et al., Reference Hayes, Lawless and Baitinger2014; King and Leigh, Reference King and Leigh2009; Lawson et al., Reference Lawson, Lenz, Baker and Myers2010; Lenz and Lawson, Reference Lenz and Lawson2011; Praino et al., Reference Praino, Stockemer and Ratis2014; Rosenberg et al., Reference Rosenberg, Bohan, McCafferty and Harris1986; Rosar et al., Reference Rosar, Klein and Beckers2008; Stockemer and Praino, Reference Stockemer and Praino2015; Stockemer and Praino, Reference Stockemer and Praino2017; Todorov et al., Reference Todorov, Mandisodza, Goren and Hall2005), particularly when candidates have little information about them (Ahler et al., Reference Ahler, Citrin, Dougal and Lenz2017). While there is scant research on how facial masks influence perceptions of attractiveness, the little work done in the area (Miyazaki and Kawahara, Reference Miyazaki and Kawahara2016) supports the oft-heard public concern that the coverings make people seem less appealing to others.

Taken together, the literature suggests that individuals who wear face masks might be more likely to appear (a) negative and (b) unattractive. This logic leads us to our Mask Hypothesis.

Mask Hypothesis: The public will be less likely to positively evaluate politicians who wear face masks.

We can imagine that the effect of mask-wearing is different for men and women politicians. Women have been historically underrepresented at every level of political office. This is because they generally face many more difficulties running for, obtaining, and being re-elected to political office than men (Paxton et al., Reference Paxton, Kunovich and Hughes2007; Waylen, Reference Waylen1994; Yoon, Reference Yoon2001). For example, women candidates must provide substantial evidence of their qualifications, while men candidates often face little pressure to display their fitness for political leadership roles (Bauer, Reference Bauer2020). It can be difficult for women to establish those qualifications, though, since many people believe that women do not have either the temperament or capability to participate in politics, let alone run for office (Pateman, Reference Pateman1989). In addition, people generally assume that leadership is a masculine trait, discounting its presence or strength among women. Related to that, people tend to evaluate autocratic modes of leadership by women more negatively than the same behaviour by men (Eagly et al., Reference Eagly, Makhijani and Klonsky1992).

Women experience not just differential expectations about their qualifications but also about their emotions and appearance, two attributes that wearing masks might change. Regarding emotions, Koo (Reference Koo2020) shows that voters have higher expectations that women politicians appear positive than men and, consequently, smile more. If women politicians wear masks, we cannot (of course) see their smiles (except perhaps in their eyes), which means that they will be seen in a less positive emotional state than if they did not don coverings. The importance of portraying a positive emotional state is not lost on women politicians, who are less negative than men in how they present themselves in political debates (Boussalis et al., Reference Boussalis, Coan, Holman and Müller2021) and on social media (Barnes et al., Reference Barnes, Crabtree, Matsuo and Ono2021), among other venues.

In terms of appearance, an abundance of studies show that women face higher standards than men in nearly every aspect of life. This appears to be particularly true in politics, where women politicians often must be more attractive than men candidates (Olivola and Todorov, Reference Olivola and Todorov2010). For example, King and Leigh (Reference King and Leigh2009) show that, in the 2004 Australian election, ‘the average female candidate was at the 70th percentile of the beauty distribution, while the average male candidate was at the 43rd percentile of the beauty distribution’ (584). Considering that, and our earlier discussion about attractiveness, it seems clear that if wearing masks diminishes how pleasant women candidates appear, they will pay a greater price for this than their men counterparts. Based on this logic, we posit the Woman Mask Hypothesis.

Woman Mask Hypothesis: The public will be less likely to positively evaluate politicians who wear face masks. This negative effect will be stronger for women candidates than for men.

3 Research design

To test our two hypotheses, we fielded a survey experiment in Japan from August 27–28, 2020, using Lucid Marketplace, an online survey platform that is becoming an increasingly popular tool for social scientists to conduct research (Coppock and McClellan, Reference Coppock and McClellan2019). We consider Japan a hard case to test our theoretical expectations. Historically, face masks have been very common in Japanese daily life. We can imagine that this means the penalty for wearing them will be lower in Japan than in other places. In other words, when masks are seen as less common – in either place or time – the penalty for politicians wearing them will likely be higher.

In addition, both politicians and members of the public in Japan generally understand the importance of wearing masks. We would expect that when masks are viewed as less necessary, which might be different from when they are less common (i.e. they can be both very necessary and not all that common, as in the early days of COVID in the United States), the penalty for politicians wearing them will also likely be higher.

That said, we do wish to note that the gender dynamics in Japan might make finding a difference in effects across men and women more likely than in other places. As has been extensively documented at this point (Crabtree and Muroga, Reference Crabtree and Muroga2021; Crabtree and Mcclean, Reference Crabtree and Mcclean2024; Muroga and Crabtree, Reference Muroga and Crabtree2020), women professionals, including elected office holders and candidates, face different expectations than men in Japan. This might matter in the context of our experiment because women in Japan tend to be evaluated more for their looks than men. Face tends to obscure appearance. To the extent that this is true, we might expect women politicians to suffer more for this in Japan than elsewhere (Golder et al., Reference Golder, Crabtree and Dhima2019).

We paid each respondent $1.50 to complete the survey, which took respondents about 7.5 minutes to complete (median = 447 seconds). This means that we paid each respondent roughly $12 an hour, or about 150% of the Japanese minimum wage at the time of this study. We think that this amount should have incentivised careful work, in addition to providing an ethical reward.

To collect a reasonably representative sample of adults across Japan, we constructed our sample using demographic quotas with respect to respondents’ age (18–29, 30–39, 40–49, 50–59, 60–74), sex (male, female), and region (Chugoku, Chubu, Hokkaido, Kinki, Kanto, Kyushu, Shikoku, Tohoku).Footnote ³ Shortly after respondents agreed to participate in our survey, we asked them to answer an attention-check question about the weather and filtered out those who responded inaccurately.Footnote ⁴ The total number of respondents who answered this question correctly and then completed the survey is $1508$ .Footnote ⁵

Prior to presenting respondents with our survey experiment, we asked them 3 sets of questions. These include (1) standard socio-demographic questions (gender, race/ethnicity, education, age, place of current residence, and income), (2) a small battery of questions about their attractiveness, and (3) several questions designed to assess their political knowledge. We asked these 3 sets of questions in random order to reduce the possibility that respondents’ answers to them affected the way in which they read the treatment materials and answered our outcome questions.

After the main experimental component, which we explain next, we asked an open-ended question where respondents were prompted to tell us what they thought about the role of beauty in politics. We then asked them to identify the country from which the politicians shown in the survey experiment came. We use these two questions as additional attention checks. Crucially, our results are substantively similar if we only examine responses from individuals who successfully pass these checks. These findings should be interpreted with caution, though, since respondent answers are recorded post-treatment, and conditioning on them could bias our estimates (Montgomery et al., Reference Montgomery, Nyhan and Torres2018). Finally, we debriefed respondents.

As mentioned above, our survey contains an experiment. In it, we told respondents that we wanted their opinion about a couple of politicians. We then asked them to evaluate two of these four currently serving Japanese politicians with national reputations: Shinzo Abe, the Prime Minister at the time;Footnote ⁶ Satsuki Katayama, a Member of the House of Councillors; Yuriko Koike, the Governor of Tokyo; and Shinjirō Koizumi, the Minister of the Environment.Footnote ⁷

We randomly assigned one man politician and one woman politician to participants. We selected a set of politicians for each gender to minimise the possibility that our results are driven by the particular politicians we selected for the experiment, described next. To minimise any possible order effects, we randomly assigned whether respondents first evaluated a man or a woman politician. We also randomised which of the two politicians they viewed within each gender category, so as to ensure that our findings were not driven by public reaction to particular politicians. Each evaluation task was presented on a separate screen to focus respondent attention and minimise spillover. We asked respondents to evaluate two politicians, instead of one, to maximise statistical power.

Prior to asking respondents to evaluate a politician, respondents were then assigned with equal probability to see a photograph of the politician either wearing a mask (the Mask treatment group) or not (the Control group). In other words, there were four possible treatment combinations that respondents could see across the two politicians they viewed – ‘Mask/Mask’, ‘Mask/No Mask’, ‘No Mask/Mask’, and ‘No Mask/No Mask’ – each assigned with equal probability.Footnote ⁸ To keep as many things constant across our treatments as possible, we selected the photographs to be as similar as possible in size, colour, and other attributes. One difficulty in doing so, though, is that Japan’s widely promulgated social distancing rules have changed the conditions (e.g. lighting, locations) in which politicians appear in public.Footnote ⁹ Figure 1 shows the photos we used.Footnote ¹⁰ In addition to showing a photograph of the politician, we also provided respondents with their name and official position. We did this to make sure that respondents had some common baseline information about the politicians, though this was perhaps not necessary given that each official is highly visible in national media.

Figure 1.

Politician photos and treatments.

Note: Respondents assigned to the Control condition for a politician received the image on the right; respondents assigned to the Mask condition for a politician received the image on the left.

We decided to use photos of real politicians, instead of artificially generated ones or photos of models, to enhance the realness of our survey instrument and the generalisability of our results. We note, though, that this poses a hard test for our hypotheses. Typically, researchers avoid using real political figures in these kinds of studies because respondents already have attitudes towards them, and the potential change in those attitudes caused by the relatively subtle manipulations in the experiment is therefore muted.Footnote ¹¹

3.2 Statistical analysis

To test our two hypotheses formally, we take the data from our $3016$ respondent evaluations ( $1508$ respondents $ \times 2$ evaluation tasks) and estimate a set of ordinary least square (OLS) models. We use the following equation to evaluate our Mask Hypothesis.

(1) $${Y_i} = {b_0} + {b_1} \cdot Mas{k_i} + {u_i}$$

The baseline category for this model (i.e. Mask ${{\rm{\;}}_i} = 0$ ) is the Control or ‘No Mask’ condition.

We use the following equation to evaluate our Woman Politician Mask Hypothesis.

(2) $${Y_i} = {b_0} + {b_1} \cdot Mas{k_i} + {b_2} \cdot Katayama/Koik{e_i} + {b_3} \cdot Mask \times Katayama/Koik{e_i} + {u_i}$$

The baseline category for this model (i.e. Mask ${{\rm{\;}}_i} = 0$ and Katayama/Koike ${{\rm{\;}}_i} = 0$ ) is Control and Abe/Koizumi.

The dependent variable ( ${Y_i}$ ) in each set of OLS models is the respondent’s ( $i$ ) answer to our 6 different outcome questions. As discussed above, we ask respondents to answer our primary outcome question, Support on a 5-point scale and our 5 other outcome questions – Attractive, Competent, Smart, Strong, and Trustworthy – on a ten-point scale. To ensure comparability across outcomes, we rescale Support to be on a 10-point scale. In our models, we treat all outcome measures as continuous. The models also include an error term ( ${u_i}$ ). To account for the fact that respondents rated two politicians, we cluster standard errors by respondents.

In additional models, we include a vector ( ${\bf{X}}$ ) of pre-treatment covariates capturing respondent differences in age, gender, education, income, region, perceptions of self-attractiveness, and political knowledge. Our results are substantively similar if we omit these pre-treatment covariates from our models.

Our Mask Hypothesis suggests that respondents in Mask, as compared to those in Control, are less likely to express support for a politician’s job performance. Therefore, we expect that in the model notated in Equation 1, ${b_1} x003C; 0$ . Our Woman Candidate Mask Hypothesis suggests that the public will be less likely to be satisfied with the job performance of women politicians when they wear masks. Thus, we expect that in the model notated in Equation 2 ${b_3} x003C; 0$ .

In several exploratory analyses, we make subgroups of respondents based on pre-treatment covariates that include respondent age, gender, education, income, prefecture, respondent evaluations of their own attractiveness, and respondent political beliefs. We selected these pre-treatment covariates because they are presumably predictive of vote choice and electoral support. We add a binary moderator (e.g. whether a respondent is a man or woman, whether a respondent makes more than the median income) and its interactions with the treatment to the model and investigate whether the treatment effects are significantly different between the two subgroups. We dichotomise the moderators to address the possibility of non-linear interactive effects.