What’s in a voice? The direct and conditional effects of vocal characteristics on voting choices.
by Lonneke van Riele
Thesis Supervisor: dr. Bert N. Bakker
Not just the contents of what politicians say, but also the voice with which they say it matters. Recent studies find that voice pitch (F0) and voice pitch variability (F0SD) influence voting behaviour in real-life elections (Banai et al., 2017; Banai et al., 2018), with winners of elections having, on average, lower voices than their opponents. Lower-pitched are perceived as more dominant (Puts et al., 2006), and the preference for more dominant leaders is often seen as the explaining factor for this general preference for lower-pitched male politicians. However, voices with lower F0 are also more attractive (Leaderbrand et al., 2008), and as we are more favourable to more attractive targets, attractiveness may also be the driving factor behind this preference. Using a combination of methods, I (1) verify and replicate whether voices with lower F0 and F0SD are more successful in elections, (2) test the relationship between F0/F0SD and attractiveness/dominance, and (3) theorize and test whether dominance, attractiveness or both can explain the association between voices with lower F0 and F0SD and increased electoral success.
Previous studies investigating the influence of F0 and F0SD on attractiveness and dominance have often used automated voice analysis with programmes like Praat (Boersma, & Weenink, 2013), which allows for computational analysis of speech. Although automated voice analysis can provide us with a lot of interesting information (like investigating how F0 and F0SD influence voting behaviour), it cannot provide us with explanations by answering the question why lower voiced politicians win elections more often. To overcome this issue, for this study, an innovative mix of automated voice analyses using Praat, manual coding, questionnaires and an experiment was used.
In study 1, the dataset used by Banai et al. (2018), consisting of voice data of the winners and losers of 69 different presidential and parliamentary elections, was extended with 24 more recent election waves. In order to test the mechanism behind the influence of F0 and F0SD on attractiveness and dominance in real-life elections, an attempt was made to have a group of 200 respondents (30 males, 170 females) code the perceived vocal dominance and vocal attractiveness of these voices on a scale of 1 to 7 in an online questionnaire. However, the intercoder reliability was low for all three factors. As it seemed scores of attractiveness and dominance were too dependent upon personal preference for the content coding approach to work, in study 2 we tried a new method. US-based respondents were asked to choose the most attractive and most dominant voice, as well as the voice they think would win in a national election, between pairs of voices taken from the dataset used in study 1.
Lastly, the goal of study 3 was to retest the effect of F0 and F0SD on vocal attractiveness, vocal dominance and voting preference in a more controlled way than in study 2. In order to do this, the F0 and F0SD of 6 voices were independently manipulated, leading to 4 different versions per voice (low/high F0, low/high F0SD). Participants were asked to indicate which voice they found most attractive, most dominant, and which they would vote for in a national election, thereby allowing for causal inferences on the effect of F0 and F0SD on vocal attractiveness, vocal dominance and voting preference instead of looking at correlation between these variables.
In study 1, the results by Banai et al. (2018) are replicated and verified, showing that the preference in real-life elections for voices with lower F0 and F0SD is relatively robust. In study 2, it is shown that voices with lower F0 and F0SD are more attractive, dominant and receive more votes. Thus, attractiveness and dominance explain part of the effect of low F0 and F0SD on election success. This conclusion is further reaffirmed by the internally more valid study 3 where the same results are found in a more controlled setting bereft of the political context. This study furthermore shows that a mixed design, combining automated voice analysis with studies where participants are asked to choose between two voices to uncover vocal characteristics like vocal dominance and vocal attractiveness in combination with a study with a more experimental approach may be the way forward.