Decomposing the Relationship Between Candidates’ Facial Appearance and Electoral Success

Numerous studies show that candidates’ facial competence predicts electoral success. However, a handful of other studies suggest that candidates’ attractiveness is a stronger predictor of electoral success than facial competence. Furthermore, the overall relationship between inferences from candidates’ faces and electoral success is challenged in two ways: (i) non-facial factors in candidate photos such as clothing and hair style as well as (ii) parties’ nomination strategies are suggested as potential confounds. This study is based on original data about all 268 candidates running in three local elections in 2009 in Denmark and supports a two-component structure of the relationship between candidates’ facial appearance and their electoral success. Facial competence is found to mediate a positive relationship between candidates’ attractiveness and electoral success, but simultaneously facial competence also predicts electoral success over and above what can be accounted for by attractiveness. Importantly these relationships are found when seven different non-facial factors, parties’ nomination strategies and candidates’ age and gender are controlled for. This suggests that the two-component structure of the relationship between candidates’ facial appearance and electoral success is highly robust.

1. Brønderslev, Frederikshavn and Mariagerfjord municipalities, located in the northern part of Denmark.

2. Approval for using the pictures was given by Nordjyske Medier. The total number of candidates in each municipality is: Brønderslev (89 candidates), Frederikshavn (87) and Mariagerfjord (92).

3. The respondents do not constitute a representative sample of the Danish voters. However, prior research gives no reason to expect that using young respondents should create biased face-based trait inferences (Antonakis and Dalgas 2009). Still, one might expect young respondents to find younger candidates more attractive, which is also the case here (young candidates are on average rated as significantly more attractive than middle-aged candidates (p = 0.003); and middle-aged candidates are on average rated as more attractive than old candidates (p < 0.001) (see “Non-facial Elements, Ballot Position, Candidates’ Gender and Age” below for information on age categories). To minimize any bias stemming from this, the perceived age of the candidates is controlled for in all models.

4. Besides attractiveness and competence, respondents rated the following traits from the candidate photos: intelligence, responsibility, dominance, friendliness, and physical strength. From this list, intelligence and responsibility are of particular interest, since they correlate strongly with competence (r = 0.880 and r = 0.840, respectively). Attractiveness is only measured using one item. In order to place attractiveness and competence on an equal basis, the single competence item will be used in the analyses. However, the same substantial results are reached using a scale consisting of competence, intelligence, and responsibility.

5. Subjects were recruited from Rosborg Gymnasium, Campus Vejle and Risskov Gymnasium.

6. Number of votes for each candidate is available online: http://www.kmdvalg.dk/kv/2009/adk.htm.

7. Logging constitutes the standard procedure for solving skewness problems. Here, skewness is reduced from 3.89 to 0.24 (zero indicates no skewness). Furthermore, the distribution of error terms approximates the normal distribution better using the logged measure of electoral success; likewise, other assumptions for using OLS regression are better fulfilled using Electoral Success (the logged measure).

8. More specifically, the logarithmic values of Relative Success, log(Relative Success), are recoded to a 0–1 scale by subtracting the smallest observed value of log(Relative Success) from every other value and finally these values are divided by the range of log(Relative Success).

9. Age is operationalized as observers’ perception of candidates’ age and afterwards trichotomized following Berggren et al. (2010), who find that the same substantial relationship between candidates’ facial traits and their electoral success is reached using candidates’ real age and using observers’ perceptions of candidates’ age (Berggren et al. 2010, pp. 14).

10. It should be recognized that some previous work already includes control variables such as photo background, candidates’ ethnicity, gender etc. by for example only including candidate pairs who do not vary on these variables (e.g., Todorov et al. 2005). While this study is not the first to control for non-facial factors, it significantly extends the non-facial factors that are controlled for.

11. All variables (except for gender) were coded by non-respondent observers. The coding process proceeded in two steps: (i) in accordance with descriptions in a codebook, two or four observers coded all candidate photos with regard to certain variables. (ii) inter-coder reliability was calculated using Krippendorff’s α (Krippendorff 2004, pp. 221–243). α-values for glasses and background indicate total agreement among observers (i.e. α = 1), while α varies between 0.70 (hair color) and 0.93 (beard) for the rest of the control variables. Specific disagreements were discussed among the observers who, after agreement, clarified potential ambiguities in the codebook to avoid similar misunderstandings for the subsequent codings. The final codebook is available from the author upon request.

12. Data on candidates’ ballot positions: http://www.kmdvalg.dk/kv/2009/adk.htm.

13. Following Atkinson et al. (2009) reasoning, one might expect Danish parties to nominate candidates with high facial competence to the ballot’s top positions. However, predicting ballot position from facial competence (b = −7.867; p = 0.196) and attractiveness (b = 3.607; p = 0.534) as well as the same control variables as in Table 1 yields insignificant results.

14. Substantially, the same conclusion is reached when I do not include the control variables: Attractiveness predicts facial competence (b = 0.450, p < 0.001) and attractiveness does not predict Electoral Success when facial competence is introduced (b = 0.068; p = 0.493) while facial competence does (b = 0.290; p = 0.010).

15. Ballot position was only included as predictor of Electoral Success since facial competence and attractiveness must be expected to influence ballot position and not vice versa.

16. The SEM-analysis consists of models similar to Model C and Model D in Supplementary material 1. Furthermore, the indirect path is also significant without control for other variables (b = 0.130, p = 0.012).

17. The same substantial result is reached when non-facial factors, ballot position gender and age are excluded. R ² doubles from 0.020 to 0.044 (with adjusted R ² rising from 0.016 to 0.036) constituting a significant increase (F (1, 265) = 6.68, p = 0.010).

I would like to thank Karl Kjær Bang, Martin Bisgaard Christiansen, Shanto Iyengar, Anne Plougmann Knudsen, Ole Laustsen, Tor Falkesgaard Mortensen, Nicolai Ottosen, Michael Bang Petersen, Morten Pettersson, Rune Slothuus, participants in the research section on comparative politics at Department of Political Science and Government, Aarhus University, the editors and three anonymous reviewers for their help, comments and advice on earlier versions of this article.

Authors and Affiliations

1. Department of Political Science and Government, Aarhus University, Bartholins Allé 7, 8000, Århus, Denmark

   Lasse Laustsen

Corresponding author

Correspondence to Lasse Laustsen.

Reprints and permissions