What’s in a voice? The direct and conditional effects of vocal characteristics on voting choices.
Master Thesis
Graduate School of Communication
Master’s programme Communication Science Name: Lonneke van Riele
Student ID: 12860263 Supervisor: Bert Bakker
Date of completion: 24-06-2021
Note: in this thesis, 3 studies were conducted and therefore the word limit was exceeded. This was discussed with and approved by the supervisor.
2 Abstract
Not just what politicians say, but also how they say it matters. Recent work by Banai et al.
2018 finds that voice pitch (F0) influences voting behaviour in both hypothetical and real-life elections. At this point it is unclear whether this effect is driven by dominance – as Banai et al. (2018) argue – or by attractiveness. Moreover, other vocal characteristics, like voice pitch variability (F0SD), that have received less attention may also influence electoral success. The aim of this study is threefold: (1) verify and replicate whether voices with lower F0 and F0SD are more successful in elections, (2) test the relationship between F0/F0SD and dominance and attractiveness, and (3) theorize and test whether dominance, attractiveness or both can explain the relationship between voices with lower F0 and F0SD and increased electoral success. In study 1, I verify and independently replicate – using results from 93 elections – the results by Banai et al. (2018), showing the preference in real-life elections for voices with lower F0 and F0SD is relatively robust. In study 2, I show voices with lower F0 and F0SD are more attractive, dominant and receive more votes. I conclude attractiveness and dominance explain part of the effect of low F0 and F0SD on election success. This conclusion is further supported by the internally more valid study 3, where similar results are found in a more controlled preregistered experiment bereft of the political context. Lastly, I show ideology may influence the way we perceive and interpret vocal characteristics like F0 and F0SD, as well as our voting choices. To conclude, this study adds to our understanding of how different vocal characteristics influence electoral success, shows the importance of assessing vocal characteristics besides F0, and shows the importance of taking into account individual characteristics as they influence both how voices are perceived and our voting choices.
3 Introduction
In 2016, Hillary Clinton, nicknamed “Shrillary”, was criticized and ridiculed for having a too high-pitched voice (Chvasta, 2020). Margret Thatcher infamously lowered her voice pitch during her time in office (Irwin, 2015). Not just the contents of what politicians communicate, but also the way in which they communicate has an effect on voting behaviour (Banai, Banai, & Bovan, 2017; Banai, Laustsen, Banai, & Bovan, 2018; Klofstad, 2016;
Klofstad, Anderson, & Peters, 2012). Most work on vocal characteristics focuses on
fundamental frequency or vocal pitch (F0). F0 describes how high or low the frequency of a voice is, and is dependent on the rate of vibration of the vocal cords (Titze & Martin, 1998).
Political candidates with lower-pitched voices get attributed positive personality traits, like intelligence, trustworthiness and honesty, more often (Tigue, Borak, O’Connor, Schandl, &
Feinberg, 2012). Males with lower-pitched voices are furthermore perceived as physically stronger (Klofstad et al., 2012), more attractive (Feinberg, Jones, Little, Burt, & Perrett, 2005), more competent (Klofstad et al. 2012; Klofstad, Anderson, & Nowicki, 2015) and more dominant (Puts, Hodges, Cárdenas, & Gaulin, 2007; Tigue et al., 2012).
Besides having an influence on how politicians are perceived, vocal characteristics have also been shown to influence the selection of leaders. In a recent influential study, Banai et al. (2018), for example, collected voice pitch data of the winners and losers of 69
presidential and parliamentary elections held across the world (together with aggregate national ideology for the countries in which the candidates were nominated). They find that winners of elections have, on average, lower voices than their opponents (Banai et al., 2018).
Similar results were found in other studies, both for real-life election outcome and in laboratory settings (Banai et al., 2017; Klofstad et al., 2012; Tigue et al., 2012).
That the voice pitch of politicians influences both how they are perceived and voting behaviour, both in laboratory settings and real life, seems clear. However, many questions
4 remain. It is, for example, still unclear which factors drive this preference for male leaders with lower-pitched voices. It has often been argued, for example by Banai et al. (2018), that dominance is the driving factor behind this preference. However, this preference may also be driven by attractiveness. Moreover, there are also other vocal characteristics besides F0 that have received less attention but may also influence electoral success. One of these
characteristics is voice pitch variability, or F0SD, which is the amount of variation that can be perceived in a voice. As argued by Banai et al. (2017), who find some initial evidence that voices with lower F0SD may be more successful in elections, F0SD could be a highly important factor in investigating the relationship between political candidates' voices and voters' decisions and should be researched more extensively. Lastly, it is unclear whether individual characteristics like ideology influence how we perceive voices and our voting choices.
This study aims to answer these questions. Firstly, this study will verify and replicate results found by Banai et al. (2018) by checking whether voices with lower F0 are indeed more successful in elections. Secondly, in this study the study by Banai et al. (2018) will be extended by testing the relationship between F0SD and electoral success. Furthermore, in this study the relationship between F0 and F0SD on the one hand, and vocal dominance and vocal attractiveness on the other hand will be tested, as well as to what extent dominance,
attractiveness or both can explain the association between voices with lower F0 and F0SD and increased electoral success. Lastly, this study will focus on how ideology influences the relationship between F0/F0SD and vocal dominance and vocal attractiveness, as well as how ideology influence the relationship between F0/F0SD and voting choices.
Voting is considered to be the foundation of a democracy (Lee et al., 2016). Especially in representative democracies, which are fully based upon having the public electing someone to represent them. Therefore, in order to have a well-functioning society, it is crucial for
5 individuals to be able to vote rationally for the politician that best represents their views and interests (Lee et al., 2016). However, as these findings show, our voting decisions may not just made rationally, but are also partly based on more shallow cues (like vocal
characteristics). As voting is such a crucial part of many societies, it is highly important that we gain a greater understanding of which different vocal cues influence our voting choices, as well as the mechanisms and individual characteristics that drive this influence.
Effect of vocal dominance and attractiveness on electoral success
‘Masculinity-related’ factors have been argued to be a driving mechanism behind the preference for leaders with lower pitched voices. As humans started living together in groups, it became necessary to solve and coordinate issues as a community (Petersen, 2015). In order to do so efficiently, it is assumed that political behaviour, like choosing a leader, emerged (van Vugt, 2006). As our ancestors relied on leaders that could coordinate ‘masculine’
activities like hunting and warfare, it is argued that humans to this day tend to choose leaders that have physical characteristics that are indicators of masculinity (Laustsen, & Petersen, &
Klofstad, 2015). An example of such a ‘masculine’ characteristic is a lower-pitched voice, as thicker, bigger and longer vocal cords, which are often found in men, vibrate at a lower frequency and are therefore perceived as lower-pitched.
As previously stated, this ‘masculine’, lower-pitched voice positively influences many different perceptions. However, the factor that has mainly received high amounts of attention as the explaining factor for this general preference for lower-pitched male politicians is vocal dominance. Indeed, lower-pitched, more masculine sounding males are perceived as more dominant (Puts, Gaulin, & Verdolini, 2006; Tigue et al., 2012). It has even been argued that low voice pitch has, partly, developed as a dominance cue among men, with dominant men lowering in voice pitch and less dominant men raising voice pitch in response to mate competition (Puts et al., 2006; Tigue et al., 2012). As dominance and masculinity are two
6 highly related characteristics (Wolff & Puts, 2010), similarly to masculinity, it is argued that dominant males are preferred when choosing a leader as they are more competent when dealing with historically important leader activities like warfare and are able to protect and prevail in conflict better than submissive leaders (Tigue et al., 2012).
However, is dominance truly the main explanatory factor behind this preference?
Besides dominance, attractiveness may also be the driving mechanism for the preference for lower-pitched politicians. Lower-pitched voices have also been found to be more attractive for men by both men and women (Jones, Feinberg, DeBruine, Little, & Vukovic, 2010;
Leaderbrand, Dekam, Morey, & Tuma, 2008). A well-established and studied phenomenon within the social psychology is the “halo-effect” (Alley & Hildebrant, 1988). Also known as the ”what’s beautiful is good” stereotype, it describes the tendency to attribute more
favourable characteristics to more attractive targets (Surawski, & Ossoff, 2006). This may explain why lower-pitched and thus more attractive voices are preferred when choosing between two political candidates.
It has been argued that vocal attractiveness and dominance are highly correlated measures. Researchers have suggested, for example, that the heterosexual men's vocal attractiveness ratings of other men are an just an index of dominance (Feinberg, DeBruine, Jones, & Little, 2008; Penton-Voak et al., 2001). However, prior research has shown that perceptions of attractiveness and dominance based on voice pitch are separable, with Jones et al. (2010) for example, showing that lowering men's voices does not have the same effects on men's perceptions of the attractiveness and dominance of other men.
These two (non-identical) concepts could theoretically both drive this voting
preference for lower-pitched male politicians in real-life elections. There are of course other factors that may be influenced by vocal characteristics and influence voting preference, like competence and age. However, dominance and attractiveness are the primary means by which
7 males win over mates across species (Andersson, 1994), making it likely that either one, or both, of these concepts is the main explaining factor for this phenomenon. Previous research has focused mainly on dominance as an explaining factor, often paying little attention to the higher attractiveness of lower-pitched voices and how this could influence voting preferences.
No previous research has investigated to what extent perceptions of dominance and
attractiveness can explain the relationship between F0 and voting behaviour. Therefore, it is unclear if increased perceptions of dominance is really what is driving this voting preference for lower-pitched male politicians in real-life elections, or whether it may actually be
increased perceptions of attractiveness. Therefore, in this study, the effect of F0 on attractiveness, dominance, and electoral success will be examined.
The effect of F0SD on electoral success
Although previous research into vocal characteristics and voting behaviour has mainly focused on F0 (Klofstad et al., 2012; Laustsen et al., 2015; Tigue et al., 2012), there are also other vocal characteristics (Banai et al., 2017), that thus far have been mainly overlooked, that may also have an influence on vocal attractiveness, vocal dominance and election success. An important vocal characteristic besides F0 is voice pitch variability, or F0SD, which captures the extent of high and low pitch in speech.
The studies that have focused on F0SD find evidence that low F0SD (a more
monotone voice) is associated with masculinity and dominance (Hodges-Simeon, Gaulin, &
Puts, 2010). High F0SD (a more sing-song character voice), has been argued to be associated with friendliness and attractiveness (Hodges-Simeon et al., 2010). Adults, for example, are likely to heighten their F0SD when speaking to children (Trainor, Austin, & Desjardins, 2000). However, conclusive evidence about the relationship between F0SD on the one hand, and dominance and attractiveness on the other hand, is missing. Most studies do indeed find moderate to high F0SD being positively related to social attractiveness ratings (Ray, Ray, &
8 Zahn, 1991; Zuckerman & Miyake, 1993). However, Riding, Lonsdale, & Brown (2006) find a trend towards increasing attractiveness with decreasing F0SD, although statistically
insignificant.
As previously stated, F0SD is a vocal characteristic that has been researched relatively little in political contexts. However, Banai et al. (2017), who did look at F0SD in political contexts, find evidence that election winners, on average, also have lower F0SD (Banai et al., 2017). They also find an interaction effect between F0 and F0SD, where male presidential candidates with lower F0 benefit from having a higher F0SD and candidates with high F0 benefit from having a lower F0SD. If a higher F0SD does indeed imply greater attractiveness, these results would imply that masculinity and dominance do not fully explain the preference for lower-pitched politicians, but vocal attractiveness may be an explaining factor as well.
As F0SD is a vocal characteristic that has been shown to likely influence election outcomes, it is an interesting factor to investigate further (Banai et al., 2017). However, as no study thus far has actively manipulated the F0SD of voices, no causal relationship between F0SD and election outcome has been established yet. Furthermore, manipulating F0SD would allow for a more controlled way of examining the relationship between F0SD on dominance and attractiveness. As previously stated, a lower F0 appears to be related to both more attractive and more dominant voices, but a low F0SD has been theorized to be related to a more dominant, but not a more attractive voice. Therefore, gaining knowledge about the relationship between F0SD and attractiveness and dominance on the one hand and F0SD and electoral success on the other hand, may also help us understand how attractiveness and dominance influence electoral success. Therefore, in this study, the effect of F0SD on attractiveness, dominance, and electoral success will be examined.
9 Ideology as a moderator of the relationship between F0 and F0SD on attractiveness, dominance and voting choices
Previous research focussing on the relationship between F0 and vocal dominance and attractiveness (Puts et al., 2006; Tigue et al., 2012) has mainly focused on average dominance and attractiveness. However, to what extent dominance and attractiveness can explain the preference for lower-voiced politicians may be highly dependent upon other factors. It is, for example, still unclear to what extent different types of audiences react to some of these vocal characteristics. One of the factors that is likely to influence how people perceive the
attractiveness and dominance of a voice is the ideology of the voter (Laustsen et al., 2015).
The influence of ideology on the perception and importance of different vocal characteristics, like vocal attractiveness and vocal dominance, has not been examined before.
The theory of trait ownership (Hayes, 2005) states that political candidates own traits, and that the public comes to expect candidates of a certain party to be associated with those traits. In the US, for example, Republicans and conservatives are associated with stronger, moral, more masculine traits whereas Democrats and liberals are perceived as more compassionate, friendly, empathetic and warm (Hayes, 2005; Laustsen & Petersen, 2016).
Research shows that people are able to identify these traits and are more likely to re-elect politicians whose traits match the stereotype related to their preferred party and/or ideological position (Samochowiec, Wänke, & Fiedler, 2010; Laustsen & Petersen, 2016).
Some evidence that ideology possibly influences the relationship between vocal characteristics and voting preference is found by Laustsen, Petersen and Klofstad (2015).
They find evidence that political candidates with lower-pitched voices are voted for more often among conservative Republicans than among liberal Democrats. Taking this into account, it seems likely that conservatives and liberals are equally likely to vote for voices with low F0, as they are associated with both more dominant and more attractive, whereas
10 conservatives are more likely to vote for voices with low F0SD (if it is indeed associated with a more dominant voice) and liberals are more likely to vote for voices with high F0SD (if it is indeed associated with a more attractive voice).
Furthermore, sensitivity to vocal dominance has been shown to be heightened when the ability to protect and prevail in conflict is at its most important, for example during wartime (Tigue et al., 2012). As conservatives have been shown to be more likely to view the world as a competitive and dangerous place than liberals (Duckitt & Sibley, 2010; Hibbing, Smith, & Alford, 2013), this may mean that conservatives are not only more likely to elect those with dominant voices, but are also more sensitive to dominance cues in voices. For liberals, who are more likely to see the world as a place of cooperative harmony, this may mean an increased sensitivity to attractiveness cues in voices (Duckitt & Sibley, 2010;
Hibbing et al., 2013). This may mean conservatives are more likely to score voices with low F0/low F0SD as more dominant than liberals, whereas liberals are more likely to score voices with low F0/high F0SD as more attractive than conservatives.
The present study
The present study will focus specifically on how different vocal characteristics (F0, F0SD, vocal dominance and vocal attractiveness) affect both each other and voting
preferences in real-life and hypothetical election scenarios. Furthermore, this study will focus on how ideology (measured on both personal and national-level) influences the perception of vocal attractiveness and dominance, as well as electoral choices.
Three different studies will be conducted. Studies 1 and 2 will be based on the research by Banai et al. (2018). The dataset used by Banai et al. (2018) is the most extensive dataset on vocal data regarding elections, including worldwide presidential and parliamentary elections from 2004-2015 and national ideology data. The aim of these studies is threefold.
Firstly, the original results by Banai et al. (2018) will be verified by reanalysing the data. It is
11 important to check formerly reported results by independently reanalysing the data (Nuijten, Bakker, Maassen, & Wicherts, 2018). After all, if the original results cannot be reproduced, the need to replicate is diminished. Secondly, a new, more recent wave of elections will be added to the dataset by Banai et al. (2018). Adding new, more recent data will allow for both replicating and verifying the robustness and reliability of the results of Banai et al. (2018), and enables the elevation of results from single observations to scientific evidence if the same results are obtained again and again (Zwaan, Etz, Lucas, & Donnellan, 2018). If replication appears impossible, we may argue that the claims made by Banai et al. (2018) are in fact unreliable or should be interpreted with caution. The first study extended the study by Banai et al. (2018) by checking to what extent F0SD influences electoral outcomes.
Furthermore, the second study will extend the study by Banai et al. (2018) by measuring the attractiveness and dominance of the voices in the Banai et al. (2018) dataset and testing to what extent vocal dominance and vocal attractiveness can explain why we tend to prefer male leaders with lower F0 and F0SD. Study 2 will also check how national
ideology influences the effect of dominance and attractiveness on electoral outcome. Lastly, study 3 was designed to experimentally and in a more controlled manner test how F0 and F0SD affect vocal dominance and attractiveness.
As the internal validity of study 2 is relatively low, study 3 was designed to
experimentally retest the hypotheses tested in study 2. The main goal of study 3 is therefore testing how dominance and attractiveness influence electoral choices. Furthermore, study 3 tests to what extent personal ideology influences the scoring of vocal attractiveness, vocal dominance and voting preference. For the full research design, the strategy of the different studies and the hypotheses, see table 1.
12 Table 1
The strategy and tested hypotheses in each of the studies present in this paper
Study: Goal: Test hypotheses: Sample Pre-
registered?
1 Verification, replication and extension of Banai et al.
(2018)
H1: results by Banai et al. (2018) will be verified and replicated, with voices with lower F0 being more successful real-life elections.
H2: voices with lower F0SD will be more successful in real-life elections.
Voice samples of the winners and losers of 93 presidential and
parliamentary elections from dataset by Banai et al. (2018) and the extension of this dataset
No
2 Testing to what extent vocal attractiveness and dominance can explain the results found by Banai et al. (2018). Also, test how national ideology influences the relationship between dominance, attractiveness and electoral success. Lastly, checking correlation between F0/F0SD and attractiveness/dominance in real-life elections,
H3: more attractive voices will be more successful in real-life elections.
H4: more dominant voices will be more successful in real-life elections.
and vocal dominance influence voting preference in real-life elections H5a: the effect of dominance on electoral success is higher in countries that are more conservative.
H5b: the effect of attractiveness on electoral success is higher in countries that are more liberal.
Voice samples of the winners and losers of 93 presidential and
parliamentary elections from dataset by Banai et al. (2018) and the extension of this dataset
Yes
3 Reaffirm results found in study 2 in a more controlled way bereft of the political context, and testing how personal ideology influences the scoring of vocal
attractiveness, vocal dominance and voting preference.
H6: Voices with lower F0 will be scored higher in attractiveness and dominance, and will receive more votes than voices with higher F0.
H7: voices with lower F0SD will be scored higher in dominance and lower in attractiveness and receive more votes than voices with higher F0SD H8: Conservatives are more likely to score voices with low F0/low F0SD as more dominant than Liberals, whereas liberals are more likely to score voices with low F0/high F0SD as more attractive than conservatives.
H9: Conservatives and liberals are equally likely to vote for voices with low F0, whereas conservatives are more likely to vote for voices with low F0SD and liberals are more likely to vote for voices with high F0SD.
Six male voices
manipulated in either F0 (high and low) and F0SD (high and low).
Yes
Study 1: verification and replication of Banai et al. (2018) Method
Similarly to Banai et al., (2018), voice data for different male prime ministers and presidential candidates was collected together with the national-level ideology of the country the candidates were running in and the election outcome. As Banai et al. (2018) restricted their sample to male political candidates, because of “the anatomical differences between
13 male and female vocal cords and the small amount of women present among the candidates with the largest vote shares” (p. 4), the focus throughout this study will be on male politicians as well.
Measures
National-Level Ideology. The data on national-level ideology was collected from the World Values Survey (WVS), which is a survey on changing values and their impact on social and political life. It is distributed across different countries worldwide every five years.
In the WVS, respondents are asked to self-report their political ideology from left to right.
Similarly to Banai et al. (2018), both the fifth and the sixth wave of the WVS were used in this study. In order to extend the study by Banai et al. (2018), and to test whether their findings are robust when adding new and more recent data, the seventh wave of the WVS (elections between 2017 and 2021) was also used in this study.
The elections closest to the date of the WVS survey were included for each of the countries in each of the three different waves. As in the research by Banai et al. (2018), both parliamentary and presidential elections were included in this study. For countries where executive power is divided between a prime minister and a president, both elections were included in the sample. For the electoral information of all added elections, see appendix A.
Acoustic analysis. For each of the parliamentary and presidential elections included in this study, the voices of the two main candidates of the election were analysed from a
YouTube video of the candidate. For each of the candidates, three randomly extracted
different 5-second long (uninterrupted) voice samples were acoustically analysed. Using Praat software, voice pitch (F0) and pitch variability (F0SD) were measured for each of the
contestants. The full list of included political candidates and corresponding YouTube URLs from the WVS seventh wave can be found in appendix A.
14 Election-specific information. Furthermore, some election-specific variables were gathered. Firstly, for each of the elections it was determined whether it was a presidential or parliamentary election. Furthermore, for each of the elections, it was recorded who won the election and the percentage of votes each of the candidates got.
Results study 1
Firstly, the dataset used by Banai et al. (2018), existing of 138 voices divided across 69 elections, was extended with the seventh wave of the WVS survey, existing of 48 voices across 24 elections. The average F0 and F0SD for both won and lost elections can be found in appendix B. There was a slight difference in voice intensity with losers scoring higher in intensity than winners (using the standardized mean intensity variable: t = 2.301, p = .023.
This was similar to Banai et al. (2018), who found t = 2.101, p = .038. Therefore, intensity was accounted throughout the analyses.
Voice pitch and election success
Across all 93 elections, an overall statistically significant negative relationship between voice pitch (F0) and electoral outcome was found (b = -.414, p = .006, table 2, column 1). Similarly, a statistically significant negative relationship between F0 and percentage of votes received was found (b = -2.65, p = .010, table 2, column 4). So, candidates with lower-pitched voices have both better chances of winning elections and generally receive more votes than those with higher-pitched voices. Candidates with a voice pitch 1 standard deviation above the sample mean (holding intensity its mean) have a 39.68%
predicted probability of winning an election. This increases to a predicted probability of 60.08% for candidates with a voice pitch 1 standard deviation below the sample mean. This effect is slightly lower than the effect found by Banai et al. (2018), who find a 34.5%
predicted probability for a candidate with a voice pitch 1 standard deviation above the sample
15 mean to win an election and a 64.9% predicted probability of a candidate with a voice pitch 1 standard deviation below the sample mean to win the election.
Table 2 shows the coefficients for wave 5 and 6 of the WVS (the original dataset by Banai et al. (2018)), and wave 7 of the WVS (the added data) separately. As can be seen, there was no statistically significant relationship between F0 and electoral outcome (b = .013, p = .962, table 1, column 3) in wave 7, whereas Banai et al. (2018) did find a statistically significant relationship between F0 and electoral outcome (b = -.627, p = .001, table 1,
column 2). Even more so, the effect of voice pitch and electoral outcome for elections in wave 7 was positive, contrary to the findings of Banai et al. (2018). This also explains why the effect of voice pitch on electoral outcome in the full, new dataset (waves 5 ,6, and 7) is smaller than the effect of voice pitch found by Banai et al. (2018). Furthermore, there was no statistically significant relationship between F0 and percentage of votes received (b = -3.12, p
= .139, table 1, column 6) in wave 7, whereas Banai et al. (2018) did find a statistically significant relationship (b = -2.866, p = .017, table 1, column 5). However, the relationship between F0 and percentage of votes received in wave 7 was negative, as also found by Banai et al. (2018). Although the overall of effect F0 on percentage of votes received across all waves was slightly smaller than the effect found by Banai et al. (2018), the effect did become more statistically significant.
16 Table 2
Main Effects of F0 on electoral outcome and percentage of received votes
Outcome Electoral Outcome Percentage of Votes
Waves 5,6,7 5,6 7 5,6,7 5,6 7
F0 -.414**
(.149)
-.627***
(1.170)
.013 (.280)
-2.65**
(1.01)
-2.866*
(1.170)
-3.12 (2.04) Voice
intensity
-.283*
(.139)
-.326*
(.159)
-.285 (.303)
-1.39 (1.16)
-1.825 (1.387)
-.134 (1.98) Constant -.005
(.044)
-.014 (.069)
.001 (.010)
38.62***
(1.03)
37.577***
(1.240)
41.63 (1.68)
1R2/pseudo- R2
0.047 .079 .014 0.04 .049 0.04
N 186 138 48 186 138 48
*p < .05. **p < .01. ***p < .001.
1All models with electoral outcome as response variable are logistic regressions (pseudo-R2), all models with percentage of votes as response variable are OLS (R2).
As for Banai et al. (2018), models to explore the nature of the relationship between F0 and national ideology were tested on both outcome variables (electoral outcome and
percentage of votes)1. The full models can be found in appendix D3. Similarly to Banai et al.
(2018), the interaction between F0 and national ideology was statistically insignificant in terms of predicting both electoral outcome (b = −.006, p = .980, table D3.1., model 1) and percentage of votes (b = −.762, p = .691, table D3.1). Thus, the general preference for lower- pitched voices holds in countries both with more left-wing and more right-wing ideologies.
Voice pitch variability and election success
Across all 93 elections, an overall statistically significant negative relationship
between voice pitch variability and electoral outcome was found (b = -.500, p = .002, table 3, column 1). Furthermore, an overall statistically significant negative relationship between voice pitch variability and percentage of votes was found (b = -.3.24, p = .001, table 3, column 4). So, candidates with less pitch variability (lower standard deviations) have both better chances of winning elections and generally receive more votes than those with more
1 Banai et al. (2018) also tested the effect of election type, as well as the nature of the relationship between voice pitch, election type and national ideology on both outcome variables (electoral outcome and percentage of votes). For full models and results, see appendices D1, D2, D3, D4.
17 pitch variability. Candidates with a voice pitch variability 1 standard deviation above the sample mean (holding intensity its mean) have a 37.56% chance of winning an election. This increases to a probability of 62.07% for candidates with a voice pitch variability 1 standard deviation below the sample mean. Although Banai et al. (2018) did not take into account F0SD in their paper, the coefficients for wave 5 and 6 of the WVS (the original dataset by Banai et al. (2018)), and wave 7 of the WVS (the added data) were separated (see table 3). As can be seen, there was a negative but not statistically significant relationship between F0SD and electoral outcome (b = -.004, p = .987, table 3, column 3) in wave 7, whereas Banai et al.
(2018) did find a negative and statistically significant relationship between F0SD and electoral outcome (b = -.689, p = .001, table 3, column 2). Furthermore, there was no statistically significant relationship between F0SD and percentage of votes received (b = - 1.91, p = .366, table 3, column 6) in wave 7, whereas Banai et al. (2018) did find a statistically significant relationship (b = -3.86, p = .001, table 3, column 5). However, the relationship between F0SD and percentage of votes received in wave 7 was negative, as also found by Banai et al. (2018). Although the overall of effect F0SD on percentage of votes received across all waves was slightly smaller than the effect found by Banai et al. (2018), the effect did become more statistically significant.
18 Table 3
Full regression models for main effects of F0SD on electoral outcome (Model 1) and percentage of received votes (Model 2)
Outcomes Electoral outcome Percentage of votes
Waves 5,6,7 5,6 7 5,6,7 5,6 7
F0SD -.500**
(.158)
-.689**
(.200)
-.004 (.281)
-3.24***
(.971)
-3.86**
(1.14)
-1.91 (2.07) Voice intensity -.248x
(.143)
-.260 (.161)
-.279 (.295)
-1.13 (1.11)
-1.41 (1.32)
-.646 (2.07) Constant -.008
(.048)
-.019 (.074)
.001 (.041)
38.62***
(1.03)
37.58***
(1.23)
41.63***
(1.71)
1R2/pseudo-R2 0.058 .089 .014 0.053 .073 .021
N 186 138 48 186 138 48
*p < .05. **p < .01. ***p < .001.
1All models with electoral outcome as response variable are logistic regressions (pseudo-R2), all models with percentage of votes as response variable are OLS (R2).
The relationship between F0SD and national ideology was tested on both outcome variables (electoral outcome and percentage of votes)2. The full models can be found in appendix D4. The interaction between F0SD and national ideology was statistically insignificant in terms of predicting both electoral outcome (b = .085, p = .742, table D4.1., model 1) and percentage of votes (b = -1.37, p = .513, table D4.1., model 2). Thus, the general preference voices with less pitch variability holds in countries both with more left-wing and more right-wing ideologies.
Can dominance and attractiveness explain the association between F0 and vote choice?
To test to what extent vocal dominance and vocal attractiveness can help explain why we tend to prefer male leaders with lower-pitched voices, an attempt was made in study 1 to have a group of 200 respondents (30 males, 170 females) coding perceived vocal dominance and vocal attractiveness on a scale of 1 to 7 in an online questionnaire conducted in march 2021. Furthermore, participants were also asked to guess the age of the speaker. However, the intercoder reliability (Krippendorff’s Alpha) was way too low for all three factors
(attractiveness: α=0.266, dominance: α=0.162, age: α =0.214). Furthermore, contrary to what
2 In this study, the effect of election type, as well as the nature of the relationship between F0SD, election type and national ideology on both outcome variables (electoral outcome and percentage of votes). For full models and results, see appendices D1, D2, D3, D4.
19 has consistently been found in previous research, voices with lower F0 were coded as being less dominant than voices with higher F0. Therefore, these results were deemed unreliable.
For more information about the method and results when I do use these (unreliable) measures see appendix C.
Study 2: the effect of attractiveness and dominance on electoral success
Method Study 2
The coding of perceived vocal dominance and vocal attractiveness in study 1 was not successful. The goal of study 2 is to test to what extent vocal attractiveness and vocal
dominance influence voting preference in real-life elections and to what extent national ideology influences this relationship. A questionnaire with a between-respondents design was used to examine determine the vocal dominance and vocal attractiveness scores of 93 the politicians’ voice fragments used in study 1 (all voices collected by Banai et al. 2018 and the extension of this dataset). Furthermore, this study was preregistered before data collection started on the first of June 2021 (for full preregistration see
https://doi.org/10.17605/OSF.IO/W8R2P).
Participants
The total amount of respondents in this study was 175. This sample size was determined based on the budget available for this study. Of those, 169 responses remained after responses not adhering to the exclusion criteria3 had been deleted (people who finished the survey within 7 minutes). Participants (N=169) were 51 females and 117 males who received a payment of $2 via MTurk in exchange for participation. The survey was accessible
3 See https://doi.org/10.17605/OSF.IO/W8R2P
20 to American MTurk workers above 18 years old with a number of HITs approved above 10.000, and a HIT approval rate above 98%.4
Instruments
All voices included in the extended database from study 1 were paired by election and presented to participants in pairs. Each participant was randomly presented six out of 93 pairs of voices (one of the winner and one of the loser of an election) For each of these pairs of voices, participants are asked to indicate which of the two voices they find most attractive, most dominant and to predict which of the two participants they think would win the election.
Procedure
Participants were first shown an introductory phase, where the study was explained.
Furthermore, participants were also explained the criteria responses had to comply with in order to be valid. These criteria were the following: all questions had to be answered, the sound check has to be filled out correctly, participants had to spend more than seven minutes on the survey and participants had to give informed consent. Afterwards, respondents were directed to the sound check, where the sound of a cat meowing was played. Participants were asked to indicate which animal they heard. For those who indicated the wrong animal, or indicated they could not turn on their sound at the moment, the questionnaire was stopped.
Next, respondents were asked general questions (gender, age category) and given a short introduction to the upcoming task (see appendix G2), after which they were presented with the 10 pairs of voices. After they had completed this task, they moved on to study 3.
4 For age and ideology distribution of respondents, see appendix E1
21 Method of analysis
Each participants listened to six sets of voices consisting of the loser and winner of an election randomly taken from the 93 elections used in the analyses of study 1. For instance, if a participant was presented with election number 68 (the 2019 Andorra parliamentary
election), a participant would get to hear the voices of Xavier Espot Zamora (the winner of this election) and Pere Lopez Agras (the loser of this election). For each of these voice pairs, participants were asked to choose the most attractive and the most dominant voice.
Furthermore, participants were asked to choose which of the two they thought would win in a national election. This way I got 1014 ratings for 186 voices of politicians. For each voice, the proportion of times the voice was chosen by participants as being more attractive, more dominant and more likely to win an election was calculated.
Study 2 also included the national ideology scores for each of the countries the elections were held in as found in the WVS study (see study 1). Linear and OLS regressions were used to investigate the relationship between vocal attractiveness and vocal dominance on the one hand and electoral outcome, percentage of votes received and the predicted chance of winning on the other hand. Furthermore, electoral outcome and percentage of votes received were regressed on vocal dominance, vocal attractiveness, national-level ideology, and the interaction between the two variables.
Results study 2
Firstly, the correlations between F0/F0SD and vocal attractiveness and vocal dominance were calculated (see appendix E2). The results indicate a negative relationship between F0 and dominance. This direction is similar to what has previously been found in research. The relationship between F0/F0SD and vocal attractiveness and dominance was more extensively tested in study 3.
22 Vocal dominance and electoral success.
Across all 93 elections, no statistically significant relationship between vocal
dominance and electoral outcome was found (b = -.130, p = .532, table 4, column 1). Besides not being statistically significant, the direction of this relationship was opposite of the
expected direction. Furthermore, no overall statistically significant relationship between vocal dominance and electoral outcome was found (b = -1.09, p = .460, table 4, column 3).
Although not statistically significant, the direction of this relationship was also opposite of the expected direction. There was, however, a statistically significant relationship between vocal dominance and predicted chance of winning (b = .187, p < .001, table 4, column 5). So, candidates with a more dominant voice are predicted to win elections more often than candidates with less dominant voices. Candidates with a vocal dominance 1 standard deviation above the sample mean (holding intensity its mean) have a 68.66% chance of winning an election. This decreases to a probability of 31.34% for candidates with a voice pitch variability 1 standard deviation below the sample mean. These outcomes did not differ between men and women (see appendix, table E3.2). Therefore, gender was not accounted for in further analyses.
The nature of the relationship between vocal dominance and national ideology was tested on both outcome variables (electoral outcome and percentage of votes). The interaction between vocal dominance and national ideology was statistically insignificant in terms of predicting both electoral outcome (b = .142, p = .692, table 4, column 1) and percentage of votes (b = .423, p = .778, table 4, column 2). Thus, the lack of preference for more dominant voices holds in countries both with more left-wing and more right-wing ideologies.
23 Table 4
Main Effects of vocal dominance on electoral outcome and percentage of received votes (Model 1) and interaction of vocal dominance and national ideology on electoral outcome and percentage of received votes (Model 2)
Outcomes Electoral Outcome Percentage of Votes Predicted chance of winning
Model 1 2 1 2 1
Vocal dominance
-.130 (.208)
-.944 (2.07) -1.09 (1.47) -3.52 (12.46)
.187 (.013)***
Ideology - .034 (.047) - -.646
(1.85)
- Vocal
dominance x ideology
- .142 (.358) - .423 (2.24) -
Voice intensity
-.346**
(.132)
-.355 (.132)**
-1.89 (1.13) -1.89 (1.15)
.012 (.011) Constant .001
(.027)
-.194 (.262) 38.62***
(1.05)
42.28 (10.42)***
.5*** (.001)
1R2/pseudo- R2
0.023 .025 0.019 .019 0.56
N 186 186 186 186 186
*p < .05. **p < .01. ***p < .001.
1All models with electoral outcome as response variable are logistic regressions (pseudo-R2), all models with percentage of votes as response variable are OLS (R2).
Vocal attractiveness and electoral success.
Across all 93 elections, no statistically significant relationship between vocal attractiveness and electoral outcome was found (b = .239, p = .253, table 5, column 1).
However, although not statistically significant, the direction was found in the expected direction, with candidates with a vocal attractiveness 1 standard deviation below the sample mean (holding intensity its mean) having a 44.12% chance of winning an election, which increases to 55.99% chance for candidates with a vocal attractiveness 1 standard deviation above the sample mean. Furthermore, no overall statistically significant relationship between vocal attractiveness and percentage of votes received was found (b = -.900, p = .446, table 5, column 3). Though not statistically significant, the direction of this relationship was opposite of the expected direction. However, this was only the case for males (b = -.757, p = .531, see appendix Table E3.1), as the direction of this relationship was positive but not statistically significant for females (b = .052, p = .965, see appendix Table E3.1). There was a statistically significant relationship between vocal attractiveness and being predicted as the winner of an
24 election (b = .147, p < .001, table 5, column 5). So, candidates with a more attractive voice are predicted to win elections more often than candidates with less attractive voices.
Candidates with a vocal attractiveness 1 standard deviation above the sample mean (holding intensity its mean) have a 64.71% chance of winning an election. This decreases to a
probability of 35.29% for candidates with a voice pitch variability 1 standard deviation below the sample mean. These outcomes did not differ between men and women (see appendix, table E3.1). Therefore, gender was not accounted for in further analyses.
The nature of the relationship between vocal attractiveness and national ideology was tested on both outcome variables (electoral outcome and percentage of votes). The interaction between vocal dominance and national ideology was statistically insignificant in terms of predicting both electoral outcome (b = .033 p = .936, table 5, column 2) and percentage of votes (b = -2.53, p = .325, table 5, column 4). Thus, the lack of preference more attractive voices holds in countries both with more left-wing and more right-wing ideologies.
Table 5
Main effects of vocal attractiveness on electoral outcome and percentage of received votes (Model 1) and interaction of vocal attractiveness and national ideology on electoral outcome and percentage of received votes (Model 2)
Outcomes Electoral Outcome Percentage of Votes Predicted chance of winning
Model 1 2 1 2 1
Vocal
attractiveness .239 (.209)
.051 (2.33) -.900 (1.17) 13.35 (14.21)
.147 (.019)***
Ideology - .032 (.047) - -.659
(1.85)
- Vocal
attractiveness x ideology
- .033 (.411) - -2.53
(2.55)
-
Voice intensity
-.344**
(.135)
-.347 (.135)**
-1.92 (1.14) -1.76 (1.17)
.017 (.013) Constant .002
(.027)
-.180 (.259)
38.62***
(1.05)
42.36***
(10.4)
.5*** (.001)
1R2/pseudo- R2
0.030 .030 0.017 0.024 0.35
N 186 186 186 186 186
*p < .05. **p < .01. ***p < .001.
1All models with electoral outcome as response variable are logistic regressions (pseudo-R2), all models with percentage of votes as response variable are OLS (R2).
25 The influence of both attractiveness and dominance on electoral success.
To further investigate to what extent attractiveness and dominance influence electoral success, a multiple regression was run to predict election outcome, percentage of votes and predicted chance of winning from vocal attractiveness and vocal dominance. Furthermore, the interaction between ideology and attractiveness/dominance was added to the model to check if the interactions between ideology and dominance/attractiveness were the same in direction and size when adding them together and controlling for the coherence between these two variables. Also, the interaction between attractiveness and dominance was added to the model to check for the correlation between attractiveness and dominance.56
A binomial logistic regression was run to understand the effects of these variables on the electoral outcome (see Table 6, column 1 and 2). The results of the regression indicated that the model explained 3.9% of the variance. Furthermore, a multiple regression was run to understand the effects of these variables on the percentage of votes received (see Table 6, column 3 and 4). The results of the regression indicated that the model explained 3% of the variance and that the model was not a statistically significant predictor of percentage of votes received, F(7, 178) = .78, p = .61. Lastly, a multiple regression was run to understand the effects of these variables on the predicted chance of winning (see Table 6, column 5 and 6).
The results of the regression indicated that the model explained 71.8% of the variance and that the model was a significant predictor of predicted chance of winning the election, F(7, 178) = 64.7, p < .001. Attractiveness and dominance did not statistically significantly contribute to the model (vocal attractiveness: p = .167, vocal dominance: p = .218).
5The models in table 6 were not preregistered.
6 See Kam, & Franzese (2007) for justification of this model.
26 Table 6
Multiple Regression Models to predict election outcome, percentage of votes and predicted chance of winning from attractiveness, dominance, and interactions with ideology
*p < .05. **p < .01. ***p < .001.
1All models with electoral outcome as response variable are binomial logistic regressions (pseudo-R2), all models with percentage of votes/predicted chance of winning as response variable are multiple regressions (R2).
Study 3: testing hypotheses in a more internally valid way Method Study 3
Study 2 shows conclude attractiveness and dominance likely explain part of the effect of low voice pitch on the predicted chance of winning elections. However, no influence of attractiveness and dominance on electoral outcome and percentage of votes received was found. In study 2, voices of real politicians and audio taken from YouTube videos were used.
Therefore, political context may play a role as a confounding factor, lowering the internal validity. This may explain why no influence of vocal attractiveness and vocal dominance on the electoral outcome and percentage of votes received was found. This also means it may be hard to establish a trustworthy cause-and-effect relationships. Therefore, the more internally valid study 3 was conducted to further check the results of study 2.
Outcome Electoral
outcome p- value
Percentage of votes received
p- value
Predicted chance of winning
p- value
Attractiveness 1.18
(2.14)
.929 17.24 (14.15)
.225 .165 (.119)
.167
Dominance .416
(.67)
.584 -10.25 (12.55)
.415 .114 (.105)
.281 Voice intensity -.005
(.044)***
.023 -.1.79 (1.21)***
.140 .016 (.010)
.117 Ideology x attractiveness 1.02
(.33)
.940 -3.19 (2.51)
.206 -.011 (.021)
.599 Ideology x dominance 1.21
(.31)
.681 1.66 (2.19)
.45 .008 (.018)
.677 Attractiveness x
dominance
.93 (.16)
.675 .03 (1.30)
.981 .003 (.011)
.792
Constant .790
(1.24)
.881 42.36 (12.42)***
.001 .51 (.104)***
<.001
1R2/pseudo-R2 0.039 0.03 0.718
27 Firstly, 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, by manipulating the F0 and F0SD of voices, 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. Also, the effect of attractiveness and
dominance on electoral votes was tested in a more controlled way. Furthermore, in this study the effect of ideology and how it influences these relationships was examined further by looking at personal ideology instead of national ideology. A questionnaire using a between- respondents design was conducted together with the questionnaire for study 2. The
participants for study 2 and 3 were the same. Furthermore, this study was preregistered (for full preregistration see https://doi.org/10.17605/OSF.IO/W8R2P).
Instruments
Six males voices saying: “When the sunlight strikes raindrops in the air, they act as a prism and form a rainbow“ (Fairbanks, 1960) were downloaded. Each of these voices was manipulated in pitch (either lowered or heightened by .5 equivalent rectangular bandwidths (ERB) in pitch, which constitutes to around 20 hertz) using Praat (Boersma & Weenink, 2013), leading to two different versions of each voice. Praat uses the Pitch Synchronous Overlap Add Method (PSOLA) algorithm to alter F0 (Boersma & Weenink, 2013), without altering other aspects of the recorded voices. Furthermore, each of these voices were
manipulated in pitch variability (either lowered (to around 10) or heightened (between 30 and 40) in F0SD whilst keeping a similar mean F0), leading to two more different versions of each voice (one monotone and one with more intonation). Thus, for every one of the 6 voices, 4 different versions were created (low F0, high F0, low F0SD, high F0SD). The different pairs of voices were divided up into two different groups, each consisting of three pairs of the same voice differing in F0 and three pairs of the same voice differing in F0SD. Participants
28 were randomly assigned to one of the two groups and thus presented with the corresponding six pairs of two corresponding voices (either low and high F0 or low and high F0SD of the same voice), and asked to decide which one they would vote for in a national election.
Furthermore, they were asked to indicate which of the two voices they find most attractive and most dominant.
Procedure
After having completed study 2, participants were once again given a short introduction (see appendix G2), after which they were presented with the 6 pairs of manipulated voices. Afterwards, respondents were asked to self-indicate their political ideology on a 7-point liberal-conservative scale as used in the American National Election Studies (see appendix G2). Lastly, they were directed to the concluding page where they are thanked for their participation.
Method of analysis
For this study, each participant listened to six randomized sets of voices, consisting of three voices manipulated in F0 and three voices manipulated in F0SD. For each of these voice pairs, participants chose their preferred voice to vote for, the most attractive and the most dominant voice. After the treatment, participants ideology scores on a scale ranging between 1 (extremely liberal) and 7 (extremely conservative) were measured for each of the
participants. First, the hypothesis that voices with lower F0 will be scored higher in
attractiveness and dominance, and will receive more votes than voices with higher F0 and that voices with lower F0SD will be scored higher in dominance lower in attractiveness and receive more votes than voices with higher F0SD was tested. To do this, the proportion of trials for which each voice was chosen as being more dominant, attractive, or voted for was calculated. Therefore, dominance, attractiveness and voting scores for all voices used in the analyses reflect the proportion of trials in which this voice was chosen for that particular
29 attribution. An independent samples t-test was performed to compare these mean
attractiveness, dominance and voting scores between voices low and high in F0 and between voices low and high in F0SD.
` Secondly, the hypothesis that ideology influences the scoring of vocal attractiveness, vocal dominance and percentage of votes received was tested. To do so, an index was created for F0 and F0SD separately, in a similar way as by Laustsen et al. (2015). This index was created for each participant over the three rounds of the experiment (three rounds in which a choice was made between low and high F0 and three rounds of the experiment in which a choice was made between low and high F0SD). This index reflects the degree of preference for the voice manipulated to be lower or higher (in either F0 or F0SD). This scale ranged from 1 to 2, with a score of 1 reflecting maximum preference for voices manipulated to be lower in F0/F0SD and 2 reflecting maximum preference for voices manipulated to be lower in
F0/F0SD. These scores were then regressed on ideology for voices manipulated in F0 and F0SD separately.
Results study 3
The relationship between F0 and attractiveness, dominance and received votes.
Three independent-samples t-test were conducted to compare vocal attractiveness, vocal dominance and the percentage of hypothetical votes received in the low F0 and high F0 conditions. There was a statistically significant difference between vocal attractiveness scores in the low F0 (M = .612, SD = .086) and high F0 (M = .388, SD = .086) conditions; t (10)7 = 4.54, p = .001. The effect size for this analysis (d = 6.41) was found to exceed Cohen’s (1988)
7,8,9
This study compares mean attractiveness, dominance and voting scores between voices low and high in F0. Therefore, the sample of these studies is the 12 different voices (low and high in F0).
30 convention for a large effect (d = .80). As expected, voices manipulated to have a lower F0 were found statistically significantly more attractive than those manipulated to have a higher F0. There was, however, no statistically significant difference between vocal dominance scores in the low F0 (M = .489, SD = .201) and high F0 (M = .511, SD = .201) conditions; t (10)8 = -.189, p = .854. Even more so, although not statistically significant, the effect was found in the opposite direction as predicted with voices with higher F0 being scored as more dominant. Lastly, there was no statistically significant difference between amount of
hypothetical votes received in a national election in the low F0 (M = .576, SD = .139) and high F0 (M = .423, SD = .140) conditions; t (10)9 = 1.90, p = .086. Although the difference was not statistically significant, there was a statistically near-significant difference in the expected direction, with voices with low F0 being voted for more often than voices with high F0. These outcomes did not differ between men and women (see appendix, table F1.1).
Therefore, gender was not accounted for in further analyses.
The relationship between F0SD and attractiveness, dominance and received votes.
Three independent-samples t-test were conducted to compare vocal attractiveness, vocal dominance and the percentage of hypothetical votes received in the low F0SD and high F0SD conditions. There was no statistically significant difference between vocal
attractiveness scores in the low F0SD (M = .554, SD = .170) and high F0SD (M = .446, SD = .170) conditions; t (10)108= 1.12, p = .286. Even more so, although not statistically significant, the effect was found in the opposite direction with voices with higher F0SD being scored as less attractive. Furthermore, there was no statistically significant difference between vocal dominance scores in the low F0SD (M = .425, SD = .183) and high F0SD (M = .575, SD = .183) conditions; t (10)11 = -1.42, p = .185. Even more so, although not statistically
10,11,12 This study compares mean attractiveness, dominance and voting scores between voices low and high in F0. Therefore, the sample of these studies is the 12 different voices (low and high in F0).
31 significant, the effect was found in the opposite direction with voices with higher F0SD being scored as more dominant. Lastly, there was no statistically significant difference between amount of hypothetical votes received in a national election in the low F0SD (M = .524, SD = .192) and high F0SD (M = .476, SD = .192) conditions; t (10)12 = .435, p = .673. Although the difference was not statistically significant, there was a difference in the expected direction, with voices with low F0SD being voted for more often than voices with high F0SD. These outcomes did not differ between men and women (see appendix, table F2.1). Therefore, gender was not accounted for in further analyses.
How ideology influences perceptions of vocal attractiveness and vocal dominance To determine if participants’ ideology influences perceptions of the vocal attractiveness and dominance, a linear regression was performed to predict the scores reflecting the preference for voices manipulated to be higher in F0/F0SD for both vocal attractiveness and vocal dominance based on ideology. There was no statistically significant relationship between ideology and vocal attractiveness for F0 (b = .006, p =.569, table 7, column 1). Although the interaction was not significant, the effect was found in the expected direction. The more liberal a participant was (low ideology score), the more often the
participant perceived voices manipulated to be lower in F0 as more attractive. There was no statistically significant relationship between ideology and vocal dominance for F0 (b = -.002, p =.851, table 7, column 2). Although the interaction was not significant, the effect was found in the expected direction. The more conservative a participant was (high ideology score), the more often the participant perceived voices manipulated to be lower in F0 as more dominant.
There was no statistically significant relationship between ideology and vocal
attractiveness for F0SD (b = -.002, p =.842, table 7, column 3). Although the interaction was not statistically significant, the effect was found in the expected direction. The more liberal a participant was (low ideology score), the more often the participant perceived voices
32 manipulated to be lower in F0SD as more attractive. There was no statistically significant relationship between ideology and vocal dominance (b = -.016, p =.150, table 7, column 4).
Although the interaction was not statistically significant, the effect was found in the expected direction. The more conservative a participant was (high ideology score), the more often the participant perceived voices manipulated to be lower in F0SD as more dominant.
Table 7.
Effects of Ideology on Vocal Dominance and Vocal attractiveness score Vocal
characteristic
F0 F0SD
dependent variable
Attractiveness Dominance Attractiveness Dominance Ideology .006 (.011) -.002 (.012) -.002 (.012) -.016 (.011) Constant 1.37*** (.045) 1.52*** (.052) 1.46*** (1.05) 1.64 (.049)***
R2 0.019 .000 .000 .012
N 169 169 169 169
*p < .05. **p < .01. ***p < .001.
How ideology, attractiveness and dominance influence voting preferences
There was no statistically significant relationship between ideology and voting
preference for F0 (b = .021, p =.069, table 8, column 1). This corresponds with the hypothesis that liberals and conservatives prefer voting for low pitched voices equally. However, the interaction was nearly statistically significant, indicating that more liberal participants (low ideology score) voted for low F0 voices more often. There was no statistically significant relationship between ideology and voting preference for F0SD (b = .015, p =.189, table 8, column 2). Although the interaction was not statistically significant, the effect was found in opposite direction. The more conservative a participant was (high ideology score), the more often the participant voted for higher pitched voices, whereas more liberal participants (low ideology score) voted for low F0SD voices more often.
Furthermore, the relationship between the average amount of votes the different voices had received on the one hand, and the average dominance and attractiveness scores for these voices on the other hand was checked. There was a statistically significant relationship
33 between attractiveness and voting preference for F0 (b = .734, p =.016, table 8, column 3).
There was also a statistically significant relationship between dominance and voting
preference for F0 (b = .606, p =.005, table 8, column 5). There was a statistically significant relationship between attractiveness and voting preference for F0SD (b = 1.04, p <.001, table 8, column 4). There was also a statistically significant relationship between dominance and voting preference for F0SD (b = .720, p = .005, table 8, column 6). Thus, for voices
manipulated in both F0 and F0SD, voices scored higher in attractiveness/dominance also received more votes. This reaffirms the finding from study 2 that both attractiveness and dominance explain part of the effect of low F0 on election success. In sum, the results from study 3 show a similar picture to the results found in study 2.
Table 8
Effects of Ideology, Vocal Dominance and Vocal attractiveness on amount of votes received Vocal
characteristic
F0 F0SD F0 F0SD F0 F0SD
Ideology .021 (.011)
.015 (.012)
- - - -
Attractiveness - - .734*
(.254)
1.04***
(.104)
- -
Dominance - - - - .606**
(.170)
.720**
(.202) Constant 1.34***
(.047)
1.42***
(.048)
.133 (.132) -.021 (.055)
.169 (.090)
.140 (.108)
R2 0.018 .001 .454 .908 .560 .560
N 169 169 12 12 12 12
*p < .05. **p < .01. ***p < .001.
Conclusion and Discussion
The goal of this thesis was to find out more on how F0 and F0SD affect electoral success. Study 1 shows the majority of the findings by Banai et al. (2018) can be verified (Nuijten et al., 2018) and extended (Zwaan et al., 2018), showing the relationship between F0 and electoral outcomes of real-life elections is relatively robust. Furthermore, study 1 shows
