We also tested the two predictions using a continuous measure of nomination success. Specifically, we create a continuous measure of nomination success by dividing a candidate’s position on the ballot by the total number of nominated candidates on the list. In this way, the continuous variable still takes the differences between being positioned as for instance number five in small and large parties into account. Finally, this measure is recoded to a 0-1 scale on which “0” reflects minimal nomination success (being positioned low on the ballot), whereas “1” reflects maximum nomination success (top ballot positions). Table S.I.7 below reports models corresponding to the models reported for the dichotomous dependent variables in the main text and in the Online Supporting Information S.I.3-S.I.6. However, because the dependent measure is now continuous instead of dichotomous we apply linear regression instead of logistic regression. Importantly, the overall pattern remains unchanged when this continuous variable is employed as our operationalization of nomination success. First, facial competence remains a positive and significant predictor of nomination success (Model 1: b = 0.204, p < 0.001)), and this holds regardless of candidates’ ideology as illustrated by the insignificant interaction between facial competence and candidate ideology in Model 2 (b = -0.029, p = 0.623). Second, although less clear and only approaching significance the pattern for facial dominance still follows the same pattern as for the analyses based on the different dichotomous operationalizations of nomination success: Facial dominance is negatively related to nomination success among liberal candidates (b = -0.164, p = 0.111) and although it falls just short of conventional levels of significance the coefficient is sizably comparable to the coefficient for facial competence. In contrast, no such relationship is found for facial dominance and nomination success among conservative candidates (b = -0.041, p = 0.561). Finally, although substantial sizable this difference in predictive power of facial dominance for liberal and conservative candidates is not statistically significant (b = 0.123, p = 0.358). In sum, despite that the analyses based on the continuous measure of nomination success only significantly supports the facial competence prediction, the pattern for facial dominance still resembles the clear results obtained across the analyses based on the dichotomous measures of nomination success.
Table S.I.7: Prediction of nomination success measured continuously (a given candidate’s position divided by the total number of list candidates). Model 1 reports effects for facial competence, facial dominance and the interaction between facial dominance and candidate ideology. Model 2 further includes the interaction between facial competence and candidate ideology. Models report unstandardized OLS regression coefficients with standard errors in parentheses.
S.I.8. Robustness analyses for facial competence using all 268 candidates
As reported in the main text only 257 of the 268 candidates were running for one of the main Danish parties with a clear ideological leaning. Therefore, the analyses in the main text are based only on these 257 candidates, because the effects of facial competence and facial dominance are investigated simultaneously in the same model. Below we estimate the relationship between facial competence and nomination success using all of the available 268 candidates, while only estimating a main effect of facial dominance. As is clear from table S.I.8, a significant effect of facial competence on nomination success is also found when all 268 candidates are included in the model.
Table S.I.8: Prediction of candidate ballot position from facial traits across all 268 candidates. Model 1 reports effects for facial competence (and facial dominance). Model reports unstandardized logit regression coefficients with standard errors in parentheses.
1 Likewise, using the American National Election Studies from 1984 to 2008 Laustsen (2016) finds that conservative voters weigh evaluations of “strong leadership” in presidential candidates more than liberal voters. Liberals, on the other hand, put a greater weight on traits related to warmth (e.g. compassion and empathy). That is, regardless of whether candidate impressions are based on unknown faces and voices or on evaluations of real-world American politicians, conservatives and liberals are found to prefer different traits in political leaders.
2 Moreover, a recent study finds that especially conservative politicians benefit from being attractive (Berggren, Jordahl & Poutvaara, 2017). Given that male attractiveness is closely linked to masculinity and facial dominance (for a review see Rhodes, 2006) this finding is in line with the results linking conservative predispositions to preferences for dominant leader and candidate faces.
3 Importantly, Danish intra-party nominations in relation to local elections should not be seen equivalent to for instance American primary elections as they are much smaller. Depending on the exact procedures as well as the size of a given local party organization, nomination races (at least partly) carry features of both committee voting and mass elections (with the participation of the members) - but the result always reflect local partisans’ candidate preferences.
4 Categorized by sex and ideological block affiliation simultaneously the 257 candidates representing main parties are distributed in the following way: 34 liberal females; 84 liberal males; 33 conservative females; and 106 conservative males.
5 Denmark does not have an institutional review board for research outside biomedical research (see http://www.nvk.dk/english). In accordance with the national scientific guidelines, formal ethical approval for the present research was not obtained. Specifically, the Committee Regulations §14, #2 exempts "survey research and interview research that does not include human biological material". At the same time, it is relevant that some of the students in the current study were under 18 years of age. To ensure the safe participation of all students, we sought and obtained formal approval for conducting every part of the study by the high school teachers of the participating students. High school teachers were approached via email and informed about the aim of the project and then decided if they would include the survey as part of their teaching. In addition, participating students always received a thorough debriefing after participation.
6 Liberal parties are Enhedslisten, SF, Socialdemokraterne, and Radikale Venstre; while conservative parties are Kristendemokraterne, Venstre, Konservative, Dansk Folkeparti and Liberal Alliance. These blocks correspond to the traditional alliance formation in Danish politics (for example, from 2001 to 2011, Venstre and Konservative held government with support from Dansk Folkeparti and Liberal Alliance (after 2007); while Socialdemokraterne, Radikale Venstre and SF (only supporting party after 2014) were in government from 2011 to 2015 supported by Enhedslisten).
7 Testing the inter-coder reliability of candidates’ perceived age using Krippendorff’s α (Krippendorff, 2004: pp.221-243), we get a score of α = 0.75. Rater disagreements were discussed and agreements were reached based on common procedures for the coding.
8 In analyses of electoral and nomination success conducted in the US campaign budgets is an often used control variable (e.g. Jacobson, 1980). In the present analyses, however, information on campaign spending is not available. Importantly, there are reasons to doubt that campaign spending constitutes an important control variable for estimating the relationship between candidates’ facial appearances and nomination success in the 2009 Danish local elections. First, parties need to nominate and rank their candidates well ahead of Election Day and, in the Danish context, the campaign budgets for individual candidates are to a significant extent provided by the parties themselves. This makes budgets a product of rather than a cause of nominations and intra-party rankings. In this regard, it is also relevant that campaign spending is focused on the final weeks leading up to a local election and is used to target voters rather than fellow party members. For these reasons, it remains highly unlikely that campaign spending affects how candidates are ranked on the ballot. Second, because incumbency is already included as a control variable in the analysis, we also indirectly control for campaign budgets because incumbents are often positioned as top candidates who receive larger shares of the parties’ total budgets. Finally, existing research on the role of campaign budgets in the 2009 Danish local elections shows that the majority of campaign budgets fall within the modest range of $1 to $3,333 and that budgets (in line with these modest sizes) only played a minor role in affecting the electoral outcome of the 2009 Danish local elections (Hansen & Hoff, 2010: pp.14, 22). For these reasons it remains highly unlikely that campaign spending could bias the estimated relationships between candidates’ facial appearances and intra-party nominations.
9 Furthermore, candidate sex does not moderate the effect of facial competence (b = -0.74, p = 0.812). The full model is reported in Online Supporting Information S.I.1.
10 Previous research suggests that the effect of facial dominance is moderated by the sex of the candidate (Laustsen & Petersen, 2016). Here we find that candidate sex does not moderate the effect of facial dominance (b = 0.25, p = 0.918), nor is a three-way interaction between facial dominance, candidate ideology and candidate sex significant (b = 6.50, p = 0.255). The full model is reported in Online Supporting Information S.I.2.
11 Moreover, this is not caused by differences between female and male conservative candidates since candidate sex does not further moderate the interaction between facial dominance and ideology (see Online Supporting Information S.I.2)
12 In addition, as reported above only 257 of the 268 represented one of the main Danish parties with a clear ideological leaning for which reason main models (estimating the effects of facial competence and dominance simultaneously) are based on these 257 candidates. However, the relationship between facial competence and nomination success replicate when the full sample of 268 candidates is employed (b = 2.50; p < 0.001) (see Online Supporting Information S.I.8 for full models).
13 Whereas Armstrong et al. (2010) does investigate the relationship between candidates’ facial competence and their success in American primary elections, American primary elections are arguably more “open” than the nomination processes in Danish parties analyzed here. Moreover, Armstrong et al. (2010) focuses entirely on facial competence leaving out the possibility that different types of candidates—and different facial traits—might appeal to voters with opposing ideological outlooks.
14 Although trait impressions and visual appearance is completely absent in the report, candidates did report a significant use of portrait posters in their campaigns suggesting that voters know how the candidates look.