Does appearance have an effect on vote share in the District Council elections?

13 Jun

Press Release

The Bauhinia Foundation Research Centre (the Centre) released on June 29th, 2016 a new edition of Occasional Paper “Candidate Appearance and Vote Share in Hong Kong”, which discusses the findings of the Centre’s trial study on District Council candidates and the impact of candidate appearance on vote share.

Overseas research have consistently demonstrated that candidates' appearances have an impact on election results, and the Centre wanted to find out whether this holds true in Hong Kong as well. At the same time, to potentially broaden its research inputs, the Centre has been considering survey methods beyond traditional telephone polls or face-to-face focus groups, namely online polls and surveys. To this end, the Centre devised a gamified online survey designed for mobile devices and used it to collect data on candidates' appearances.

The pilot survey was conducted from May 13th, 2016 to May 23rd, 2016. The survey invited participants to rate District Council candidates based on their official election photos, and the interface was modelled after the online dating app "Tinder.” Marketing efforts advertised the platform as the Centre's foray into online dating. Approximately 2,240 respondents participated in the survey, which consisted of two polls. The first poll invited participants to “like” or “dislike” ten District Council candidates based on their photos, while the second poll asked participants to select their preferred candidate from a list of photos belonging to candidates who competed in the same constituency.

The platform design differed significantly from the methodology used by telephone polling and academic studies conducted on candidate appearance in overseas contexts. For example, since the survey was conducted using online volunteers, the participants could not constitute a random and representative sample of the Hong Kong population. Also, the platform could not verify that users were accurately reporting their personal information or that they did not recognise the candidates they were evaluating. Ultimately, the sample was biased in favour of users who self-identified as young (aged 18 to 29) males and “localist” supporters. It is also likely that many candidates were recognised by participants.

Despite the fact that the sample was not representative of the Hong Kong electorate, the study found that candidates who were rated highly by our users in the first poll tended to win more vote share in the District Council elections. Candidates ranked in the top 20% by appearance rating had an average vote share of 45%, while candidates ranked in the bottom 20% won an average of 33% vote share. The disparity was found to be slightly greater for candidates ranked in the top 10%, who had an average vote share of 47%, and the bottom 10%, who had an average vote share of 30%. The average for all candidates was 42%.  Compared to all candidates, the effect of appearance ratings on vote share was more significant when only the top 20% and the bottom 20% of candidates by appearance rating were considered.

Participants had four options to evaluate candidates- “Super Dislike,” “Dislike,” “Like” and “Super Like”. These options corresponded to appearance scores from 1 to 4. Most candidate received average ratings concentrated around the mean of 1.86, with the ±1 standard deviation covering around 75% of the candidates. Average ratings ranged from a maximum of 3.06 to a minimum of 1.42. Only two candidates had an average rating higher than 3.00.

The study found that incumbency also had a positive and significant effect on vote share. In other words, District Councilors and relatively attractive candidates tended to win more votes than non-incumbents and unattractive candidates. Somewhat surprisingly, the effects of age and sex were insignificant in the first poll. Pro-establishment and pan-democrat candidates had a significant advantage in the elections over independent candidates. Candidates from “localist” parties had no advantage over independents.

In the second poll, the study found that participants “voted” for actual election winners at rates higher than winning odds. For two-candidate races, which made up the majority of constituencies in the election, 58.6% of the participants selected the winning candidate. In three-candidate races, the accuracy rate was 40.4%. Younger candidates had a slight advantage over older candidates in two-candidate elections in vote share.

If the survey findings are found to be accurate and applicable to Hong Kong elections, political parties seeking to maximise vote share should nominate candidates whose appearances (as judged by their photos) are rated highly based on voter surveys. At a minimum, parties should ensure that all of their candidates are photographed by professionals and image consultants who are able to maximise their visual appeal.

The major limitation stemming from the data set is that respondents do not proportionately represent Hong Kong's general population and certainly do not represent well the electorate who voted in the 2015 District Council elections. Despite this, when their evaluation of candidate appearances are applied to actual election results, a statistically significant relationship can be found.

Going forward, the Centre will continue to explore new means of conducting surveys and gathering data. Online polls, especially the informal type like this one, may be useful for collecting highly subjective or visual-based data. But for the Centre to ever use online polls as part of its mainstream research work, challenging issues such as sample size and representativeness must be resolved. 

Please refer to the Paper for details.