Speaking style and candidate evaluations

ABSTRACT I examine how a politician’s speaking style influences how voters evaluate the candidate. I argue that, above and beyond the content of the message, how a candidate conveys the message has important effects for voter evaluations of the candidate. I focus on two speaking styles: a powerful, straightforward and direct speaking style, and a powerless style, marked by hesitations, hedging and questions. Using original experimental data, I find that candidates who adopt a powerful speaking style in a debate are evaluated more favorably than those with a powerless speaking style. I also find that this effect is somewhat dependent upon the speaker’s gender – women are penalized more than men for adopting a powerless speaking style. Among female participants, the gender gap in evaluations is eliminated for women who adopt a powerful, but not a powerless, speaking style. Among male participants, however, the gender gap exists regardless of speaking style. I additionally find that powerless speaking style makes candidates more likely to be interrupted in the 2016 Presidential primary debates.

In the political world, it is increasingly important not just what candidates say, but how they say it. Here, I focus specifically on the speaking style of candidates and politicians. This line of inquiry is perhaps of ever increasing importance. During the presidency of Donald Trump, we have witnessed an important rise in unscripted moments, both in public appearances and on Trump's prolific Twitter feed. Indeed, recent coverage of a Trump event was described by The Independent as "descend [ing] into incoherent rambling." 1 This stands in stark contrast to traditional avenues of political speech, where Barack Obama was frequently criticized for his reliance on scripted remarks, often read from a Teleprompter. 2 As politicians are more frequently making unrehearsed remarks, often on social media, it is important to consider the political consequences for adopting atypical rhetorical styles.
I focus on two distinct speaking styles in this study. I compare a powerless speaking style, which is marked by pauses, hesitations, qualified and deferential statements to a powerful speaking style, which is clear and direct, and free of the markers of powerless speech. Generally, a powerful speaking style typifies how we might expect our leaders to speakclear, direct, and with confidence. In contrast, a powerless speaking style is often indirect, muddled, and conveys a lack of confidencespeech that is seemingly inappropriate for a leader. Given the rise of social media, not only are politicians increasingly prone to making off-the-cuff, unprepared remarks, but individuals are more likely to capture video of these remarks made in public events. Candidate speech may matter more today, as politicians have more avenues for their true speaking style to show.
Unscripted or spontaneous speech moments are especially ripe for the use of powerless speech. While scripted and prepared remarks are typically read in a clear, direct style, spontaneous speech is often littered with hesitations and pauses, words like "um" or "uh," that do not contribute to the content of the remarks (Corley and Stewart 2008). These hesitations are clear markers of a powerless speaking style (Erickson et al. 1978), and may have unique consequences for politicians engaging in unscripted speech.
While a powerless style may be harmful for political candidates, it may not have an equal effect on all types of candidates. Research on speech patterns shows that women and men are rated differently based on how they speak (Carli 1991;Reid, Keerie, and Palomares 2003). These findings suggest that women face an interesting dilemma as politicians: as politicians, they are expected to speak in a direct, powerful style, but as women, it would benefit them to speak in a powerless style. Considering this, I expect candidates will generally be punished by voters for adopting a powerless speaking style, and this punishment will be especially harsh for women, especially among female voters. I use original experimental data that allows me to isolate the effects of speaking style on candidate evaluations. Additionally, I use data from the 2016 Presidential primary debates to determine the real world consequences for candidates who adopt a powerless speaking style.

Speaking style and inferences about speakers
In this paper, I examine how speaking style influences candidate evaluations. To examine this, I look at speaking style as a concept that has been defined in social psychology research -I compare a "powerless" speaking style to a "powerful" speaking style. Powerless and powerful speech have long been examined in social psychology research, and the terms were coined first by Erickson et al. (1978) in a study of witness testimony in courtrooms.
Following Erickson et al. (1978), I define a powerless speaking style based on existing work, and focus heavily on four markers of speechhedging, intensifiers, tagging and hesitations. Hedging involves the use of qualifying words to weaken statements, such as "kind of" or "probably," intensifiers involve use of words like "really" or "very" to intensify speech, tagging involves the use of adding questions at the end of sentences, such as saying "That's it, right?" instead of simply "that's it," while hesitations are the use of audible pauses such as "umm" or repeated words in speech (Bradac and Mulac 1984). These forms of speech cause messages to be rated as less powerful, effective and authoritative (Bradac and Mulac 1984).
In contrast to a powerless speaking style, a powerful speaking style uses direct and terse language, free of hedging, tagging and hesitation (Bradac and Mulac 1984). Essentially, a powerful speaking style is clear and direct, without any of the noted arkers of powerless speech. Powerful language is likely how we expect our politicians to speakclear and to the point (perhaps the opposite of how we might expect scholars of political science to write). As such, politicians who engage in powerful speech should enjoy an advantage over those who take a powerless speaking style.
Indeed, language and speaking should be important in political life, where a vast amount of political information is received via the spoken word from media and politicians alike. Schlesinger (1974) dramatically notes, "We infer the spirit of the nation in great measure from the language" (557). Tetlock (1983) finds that conservative and liberal politicians use different types of language when they are speaking. Incumbents and challengers also differ in their speaking styles during presidential elections, with incumbents focusing more on optimism and commonality and challengers being more patriotic and ideological (Hart 2000).
Despite the focus on speech in general in politics, most attention has been focused on the content of speeches, rather than the speaking styles they use to convey these words. When policies are described with words that individuals have a strong negative reaction towards, such as torture or cancer, these policies enjoy lower levels of support (Utych 2018). Politicians also frequently use rhetoric in an advantageous way to win over public support for their policies (Riker 1996;Lakoff 2004;Nelson 2004). Of course, this line of work focuses on how words can vary, but what consequences are there for politicians based not on their words, but on how they speak?
Roderick Hart (2000) argues that certain political candidates often had less than ideal speech patterns, such as Ross Perot's tendency to be too circuitous and long-winded in his public statements, and Bob Dole's refusal to use a teleprompter and offering up too many off the cuff statements. Powerless speech harming political candidates is not a new phenomenon. In his speech the night before the election in 1976, Gerald Ford addressed the nation in a powerless way, using hedges and deference or tagging frequently (Hart 2000). These markers of powerless style seem to make Ford's comments less impactful: rather than stating his attacks on Carter directly, he used hedging statements during the attacks. This caused the public to perceive the attacks as Ford's opinion, while a powerful style may have caused the public to view the attacks as the truth. This style was common from Ford, whose 1976 Presidential debates featured numerous powerless markings, including significant hesitations. Even when describing his own policies, Ford seemed uncertain, saying "In my judgment the best way to get jobs is to uhexpand the private sector, where five out of six jobs today exist in our economy. We can do that by reducing Federal taxes as I proposed uhabout a year ago when I called for a tax reduction of $28 billionthree-quarters of it to go to private uh taxpayers and uh one-quarter to the business sector." 3 From this evidence, it is clear that presidential candidates are expected to speak in a way that is direct and powerful.
Candidate traits play a powerful role in how candidates are viewed. Perceptions of candidate competence and trustworthiness are often correlated with overall evaluations of the candidate (Markus 1982;Kinder 1983;Rahn et al. 1990), and politicians make great efforts to portray themselves to their constituents as competent and honest (Fenno 1978). Traits related to individual warmth are often reported as less important in evaluations of politicians than skill-related traits (Kinder et al. 1980). As such, I examine how speaking style influences impressions of candidate traits, as they have an important impact on overall candidate evaluations. I further examine whether speaking style has a different effect on skill-related evaluations and warmth or personal related evaluations.
The use of powerful speech has been studied primarily in two areas by scholars of language and psychology: its effect on evaluations of the speaker, and its effect on the persuasiveness of a message. Those who use powerful language are generally rated as more credible speakers than those who use powerless language (Conley, O'Barr, and Allen Lind 1979;Haleta 1996;Sparks, Areni, and Chris Cox 1998). Furthermore, speaking in a powerful style leads to more favorable ratings of speakers on a host of other positive traits, such as intelligence, competence, likability and trustworthiness (Conley, O'Barr, and Allen Lind 1979;Hosman 1989;Gibbons, Busch, and Bradac 1991;Hosman and Siltanen 2006). While these markers are typically studied simultaneously, work that focuses on individual markers suggests that intensifiers may make speech seem more authoritative (Hosman 1989), even though their inclusion still leads to lower ratings of candidates (Bradac and Mulac 1984). Since those who use powerful speech are rated more positively than those who do not, it is important to then ask if these individuals are also better at persuading others.
While, on average, powerless speakers are rated lower on positive trait characteristics and as less persuasive than powerful speakers, research also suggests that there may be a gender differential between men and women on such evaluations. Some research finds that women and men are rated equally on traits when using powerless language (Conley, O'Barr, and Allen Lind 1979;Bradac and Mulac 1984). However, Bradac and Mulac (1984) rely on written text with only a small gender name manipulation; generally, audio is more effective in conveying powerful speaking style (Sparks, Areni, and Chris Cox 1998), and reading text should make gender considerably less salient than hearing a gendered voice. While Conley, O'Barr, and Allen Lind (1979) do use audio recordings, their manipulation of powerless speech is small, and they do find that, observationally, women are more likely than men to use powerless language. Furthermore, it appears that the gender of the individual receiving a message matters along with the gender of the person speaking it. In a discussion, women are more persuasive with other women when they use powerful language, while they are less persuasive with men when they speak in a powerful way (Carli 1991). Similar patterns emerge with recorded speech, instead of face to face discussions, as women were rated as more likeable and more trustworthy by male subjects when they used a powerless speaking style, though the reverse was true for female subjects (Carli 1991;Reid, Keerie, and Palomares 2003). When women speak in an assertive way, they are rated as equally likeable and influential as men, but when women use tentative speech, a large gender gap emerges (Bongiorno, Bain, and David 2014). In political advertisements, women's voiceovers are seen as more credible on women's issues, though less credible on traditionally masculine issues, than men's voices (Strach et al. 2015).
Powerful speech should be a useful tool for politicians, though it may be decidedly less useful for women than for men. Indeed, citizen's expectations of leaders, and what they expect about women, often diverge (Eagley and Karau 2002). Women face gender stereotypes in terms of their perceived issue positions, and stereotypes that they may be less emotionally suited to be a politician (Sanbonmatsu 2002;Bauer 2015). Men are often perceived to be stronger leaders than women, though women can be seen as having better personal skills (Huddy and Terkildsen 1993;Sanbonmatsu and Dolan 2009). Issues such as education and family life are seen as stereotypical women's issues, where women candidates are evaluated more favorably, while issues like foreign policy lead to more favorable evaluations of men (Holman, Merolla, and Zechmeister 2016). Women candidates are more vulnerable to trait-based negative advertising, especially when they are challenged on issues that are seen as stereotypical strengths for women (Cassese and Holman 2018).
An instructive example comes from perhaps the most prominent woman in American politics: Hillary Clinton. Clinton's speech patterns have become markedly more masculine over time, save for a brief change during the end of her unsuccessful 2008 presidential campaign (Jones 2016). When women present themselves as more masculine, they may not face a substantial backlash (Schneider 2014), but this backlash does seem to occur among out-partisan voters (Krupnikov and Bauer 2014;Bauer 2017). This suggests women face a challenge as politicianscandidates may do better among co-partisans by highlighting masculine issues, but worse among members of the other party. Women are often evaluated differently from men on non-verbal or non-content related speech. Women who use assertive hand gestures are rated less favorably, while men who use these same gestures are rated more favorably (Everitt, Best, and Gaudet 2016). Even a woman's vocal pitch can influence evaluations of her as a leader (Anderson et al. 2014), where citizens prefer women leaders with lower pitched, or more masculine, voices (Klofstad, Anderson, and Peters 2012). Women members of Congress often speak with greater emotional intensity, and a different vocal pitch, when discussing women's issues, compared to both men and other issues (Dietrich et al. n.d.).
However, it appears female politicians face another stereotype that may be more daunting: while women often have similar expectations for how women and leaders should speak, men may expect women to speak in a powerless way, and leaders in a powerful way. It is important to consider how speaking style matters for the success of a woman as a politician, and trying to determine how they can manage to balance the expectations of their speaking patterns as both a woman and a political candidate. Perhaps one opportunity would be to downplay gender; gender salience seems to influence speaking patterns in women (Reid, Keerie, and Palomares 2003), and perhaps making gender salient may influence expectations of speaking patterns from men. Powerless speech may be a negative for candidates, regardless of gender, but it is important to consider if this has a stronger effect for women.

How speaking style influence candidate evaluations
Based on this theory, I expect that political candidates who use a powerful speaking style will be evaluated more favorably and receive greater levels of support than powerless speakers (hypothesis 1). Powerful speaking style will lead to more positive inferences on a host of traitspowerful speakers will be seen as more competent, experienced, honest, and as better leaders than powerless speakers. Given that non-verbal competence cues like appearance can positively influence trait evaluations (Olivola and Todorov 2010), it is likely that another non-verbal cue, speaking style, will have a similar effect.
I also expect the effect of speaking style on candidate evaluations to be conditional on gender. Based on research in other fields that shows women and men speakers are evaluated quite differently (Bongiorno, Bain, and David 2014), I expect that women will be viewed less favorably than men when they engage in a powerless speaking style, an effect driven by the evaluations of female voters (hypothesis 2). Expectations for male voters are less clear, as they may prefer a powerful speaking style from a leader, but prefer a powerless speaking style from a woman. For women, expectations converge: they would prefer both women and leaders to engage in a powerful speaking style.
Additionally, I expect that powerless speech will have an effect on candidates' debate performance. As candidates use more powerless speech, they should be more likely to face negative consequences in a debatewhich I operationalize as being interrupted by another debater.

Speaking style and candidate evaluations: An experimental test
To test these hypotheses, I turn to an experimental design. An experiment is especially well suited for testing my hypotheses, as I want to isolate the effect of speaking style on evaluations of each candidate while holding all other factors constant. To the greatest extent possible, I want to keep the general text of the candidate's statements the same, while only adding in markers of a powerless speaking style. This allows me to control for confounding factors in an observational analysis of speaking style, such as the content of the speeches and the identity (along with other unique speech patterns) of the person speaking.
I collected data using a sample of undergraduate students 4 at a private Southern university in November 2014. Students in political science courses were recruited to participate in an omnibus study on politics in exchange for extra credit in their courses. A total n of 301 students participated in the present study, which took under five minutes to complete. In this study, subjects were asked to listen to an approximately one-minute long audio recording they were told was from a gubernatorial debate in another state. The topic of this text was a discussion of cap and trade policy. The primary difference between the two treatment texts were the inclusion of markers of powerless style in the powerless treatment, and the absence of such markers in the powerful treatment. These full texts are available in Appendix 1. This policy was selected to minimize the impact of perceived partisanship of the candidates on evaluationsthe text of the debate featured both environmental and economic arguments for cap and trade policy, appealing to both liberals and conservatives. 5 Since partisan cues can often override impression driven evaluations of candidates (Iyengar and Barisione 2015), I take care to minimize any partisan cues in this study. 6 Afterwards, subjects were asked to answer questions about the candidate whose debate response they had just heard.
The experimental design for this study takes a classic 2 × 2 design. Subjects were assigned to listen to a speech either in a powerful or powerless style. Additionally, the gender 7 of the speaker was variedhalf of the subjects were assigned a woman speaker, while the other half were assigned a man speaker. This design allows me to make comparisons across both speaking style and across sex. This design is presented visually in Figure 1.

Results
First, I examine the direct effect of speaking style on candidate evaluations. Subjects were asked to evaluate candidates on experience, leadership, competence, honestly and warmth. 8 In these analyses, I pool results for speaker gender. This test allows me to examine the effects of powerful and powerless speaking style in general, before turning to an analysis of these results conditional on speaker gender. 9 These results are presented in Table 1. As shown in Table 1, there is a large effect of speaking style on candidate evaluations. Candidates who take a powerful speaking approach are viewed as more experienced, a better leader, and more competent than those who use a powerless speaking style. These results are large in magnitude, all over 1 point on a seven-point scale, with some effects approaching two points. These effects dwarf the effects of standard demographic predictors of candidate evaluations. These traits relate to candidate skillhowever, we see weaker effects related to personal traits like honesty and warmth. Powerful and powerless speakers are seen as roughly equally honest, with powerful speakers rated only about .17 points higher in honesty (p∼.18, two-tailed) and rated only about 1/3 of a point higher in warmth. Here, powerful speakers are still given an advantage, but not nearly as much as they are given on traits related to skill.
To further examine how speaking style influences candidate evaluations, I turn to analysis of additional, non-trait dependent variables. Subjects were asked how many previous offices they believe the candidate has heldthis variable is recoded into Previously Elected which takes a value of 1 if subjects believed the candidate has ever held previous elected office and 0 if they believe they have not. Subjects were also asked how likely the candidate is to win the election, on a 5 point scale from very likely to very unlikely. Subjects also rated the quality of the candidates argument on a seven-point scalesubjects rated the credibility, quality, trustworthiness and persuasiveness of the argument, and these were indexed (Cronbach's alpha = .89) into a scale, Argument Quality. 10 The results of these analyses are presented in Table 2.  Once again, the difference between powerful and powerless speech is staggering. Powerful speakers are seen by roughly 90% of subjects as having held political office before, while only about 47% believe powerless speakers have held previous elected office. Powerful speakers are seen as over 1 point more likely to win the election on a five-point scale. Subjects are also likely to rate a powerful speaker's argument as higher quality, by over 4 points or roughly 18% of the total scale.
Overall, these results provide strong support for hypothesis 1. Powerful speakers are rated higher on a host of traits, especially those relating to skill. Powerful speakers are considered to be more experienced, competent, and better leaders than powerless speakers. They are also perceived to be strong candidates, as evidenced by an increase in perceptions that they are likely to win the election. Additionally, the arguments of powerful speakers are rated as higher qualitydespite the fact that the content of the argument was exactly the same in the powerful and powerless conditions; subjects rated powerful speaker's arguments as considerably higher quality than those of powerless speakers. Powerful speakers, in the political world, gain a huge advantage. To further test these effects, I analyze the effects of powerful speaking style broken down by candidate and subject sex.

Powerless speaking style and gender
To test hypotheses based on gender, I turn to a set of analyses broken down by the gender of the speaker. I expect that women will be punished more harshly than men for engaging in a powerless speaking style, and that this effect will be driven by female respondents' views of the speaker. To test these hypotheses, I have created two scales from the candidate evaluation variables: Skill evaluation and Personal evaluation. Skill evaluation is measured by indexing the variables for competence, experience and leadership (Cronbach's alpha = .89), and ranges from 0 to 18 (mean 8.53, s.d. 3.83). Personal evaluation indexes the variables for honesty and warmth (Cronbach's alpha = .64), and ranges from 0 to 12 (mean 6.15, s.d. 1.94). These results are presented in Table 3.
These results show that significant differences between evaluations of men and women emerge, but only when a powerless speaking style is employed, and only on evaluations of candidate skill. When a powerless speaking style is adopted, men are rated significantly higher on skill evaluations than women, a difference of 1.74 points on the 18 point scale, or roughly 10% of the scale. When a powerful speaking style is employed, this gap disappears. Men are still rated slightly higher, by just less than half a point, but this difference does not approach statistical significance. There is a lack of significant gender differences on personal evaluations, as men and women are rated roughly equally on honesty and warmth regardless of their speaking style. It appears that speaking style influences evaluations of candidate skill, rather than candidate personality, and that women are punished significantly more than men for engaging in powerless speech.
A gap also emerges on perceptions of candidates' likelihood of winning election, as shown in Table 4. Men are always seen as more likely to win election, but this difference is larger in the powerless speaking condition compared to the powerful speaking condition (difference = 0.26, p < .01). No significant differences based on speaker gender, however, appear in reference to argument quality, either in the powerful speech or powerless speech condition. Table 5 demonstrates the relationship between gender and speaking style, broken down by experimental subject sex. 11 Male subjects generally rate the man higher, regardless of speaking style. They give the man an advantage of over a point in the powerful speaking style treatment, and a similar magnitude in the powerless speaking style treatment. The difference is substantively small at .13, and does not reach statistical significance. However, female subjects behave quite differently. They actually give a slight advantage in skill evaluation to the woman in the powerful speech condition, though this effect is small at .24 and not statistically significant. However, in the powerless condition, female respondents give a large advantage of over 2 points on skill ratings to the man. The difference for female subjects between the conditions is nearly 2.5 scale points.  These results demonstrate that female participants tend to punish women who engage in a powerless speaking style rather harshly, while male participants tend not to provide this punishment. This suggests that when women are punished for taking a powerless speaking approach, this punishment is primarily doled out by female citizens.
These results provide some support for hypothesis 2, that women are punished more harshly than men for engaging in powerless speech, but in quite specific ways. Women and men, when adopting a powerless speaking style, are rated roughly equally on personal evaluations and argument quality. However, women are punished significantly more than men for taking a powerless speaking style in terms of skill evaluations and their projected viability as a candidate. And, these results are primarily driven by female subjects more negative evaluations of women for adopting a powerless speaking style. These results also demonstrate that, when a powerful speaking style is adopted, ratings of men and women become indistinguishable, but only among female respondentsmale respondents continue to rate the man more positively.

Powerless speaking and the 2016 presidential debates
Given that speaking style influences how candidates are evaluated, it is important to consider the potential consequences that candidates may face by adopting a powerful, or powerless, speaking style. Given that many candidate speeches are prepared remarks read from a teleprompter or script, analyzing a candidate's true speaking style is often difficult. To this end, I examine how often candidates employ a powerless speaking style in the 2016 presidential primary debates, and the extent to which powerless speaking style causes a specific negative outcomebeing interrupted in the debate. Interruption behavior is common among those trying to get attention in group discussion settings, such as debates (see Grob, Myers, and Schuh 1997 for discussion). Are powerless speakers interrupted more often than powerful speakers? I predict that, given that powerless speech is often seen as a weak form of speech, individuals who adopt a more powerless speaking style will be interrupted more often in debates than those who adopt a powerful speaking style.
To test this prediction, 12 I turn to a content analysis of all 2016 presidential primary debates. These debate transcripts were collected from the American Presidency Project, 13 and hand coded for markers of powerless style 14 and interruptions by a research assistant who was blind to the hypotheses of the project. Each individual candidate in the debate was coded for their use of powerless speech markers, and the number of times they were interrupted 15 , in each debate. A total of 12 Republican debates, and 10 Democratic debates, had available transcripts, and were coded in this analysis. This provided for a total of 145 unique candidate-debate records.
Perhaps unsurprisingly, Donald Trump stands as an outlier in speech and interruption behavior. Trump was far and away the Republican who used the most powerless speech markers, averaging 140 powerless language markers per debate, more than doubling Jeb Bush and Rick Santorum at just over 60 per debate. Trump showed an especially disproportionate use of intensifiers, in line with his rhetorical style on Twitter, and an increased amount of hesitations, perhaps suggesting his unfamiliarity with the debate format. Trump was also the Republican most likely to be interrupted, and most likely to interrupt others. Among Democrats, Bernie Sanders and Hillary Clinton were remarkably similar in their use of powerless language and in interruption behavior. I analyze this data by examining how the use of powerless speech by candidates influences the likelihood of the candidate being interrupted by another candidate in a debate. 16 These results are presented in Table 6. Controls are included for the logged number of times a candidate spoke in the debate, as those who speak more often have more opportunities to be interrupted, and more opportunities to use a powerless speaking style. An additional control for whether or not the participant was in a Democratic debate was included, in order to account for potential differences between the candidate fields. Models 1 and 3 include candidate fixed effects, to account for variation among the candidates, while Models 2 and 3 present results without these fixed effects, with robust standard errors clustered by candidate.
These results 17 demonstrate that powerless speakers tend to be interrupted more than powerful speakers in the 2016 presidential debates. The number of powerless statements varies in the 2016 debates, from a minimum of 5 to a maximum of 327, with a mean of 64.3 and a standard deviation of 51.4. Substantively, the effects of powerless speech are somewhat strong. The OLS coefficient of .06 predicts that candidates should be interrupted one additional time for roughly each 17 additional powerless speech markers. As such, a one standard deviation increase in powerless speech predicts that a candidate will be interrupted 18 about 3 more times over the course of a debate. Model 2 makes similar predictions, with 25 powerless markers predicting an additional interruption, and a one standard deviation increase in powerless speech predicting roughly 2 additional interruptions per debate. With a low mean value of interruptions per candidate per debate of just over 3, these effects are quite substantial.
Of course, it is possible that not all markers of powerless speech operate in a similar fashion. To test the effects of each type of powerless speech, I turn to analyses in Models 3 and 4. Candidates who use hedging (statements such as "kind of" or "sort of"), contrary to expectations, are actually less likely to be interrupted during the debate than those who do not, while those who use intensifiers (statements such as "very" or "extremely") and hesitation (pauses or repeated words in their speech) are more likely to be interrupted than those who do not. There does not appear to be any effect of using tagging statements on the likelihood of being interrupted. Taken together, these results suggest that powerless speech can harm candidates in debates, but perhaps only certain types of powerless speech are negative for candidates. Powerless speakers are interrupted significantly more often than powerful speakers, but this effect appears to be driven primarily by hesitations and intensifiers, rather than hedging or tagging statements.

Conclusions
I find that candidates who engage in a clear, direct, powerful speaking style are viewed as more skilled than those who engage in a powerless speaking style, even when the content of their speech is the same. These effects are consistent and large in magnitudeindividuals view powerful speakers as more skilled, more experienced, more likely to win an election, and providing better quality arguments than powerless speakers. These effects persist regardless of the gender of the candidate. However, women are punished more harshly than men for engaging in a powerless speaking style, and this punishment is especially doled out by female respondents. However, when women engage in a powerful speaking style, these differences in evaluations become more subtlefor male participants, men remain preferred. For female respondents, however, the gender difference in evaluations of candidates disappears. While both men and women benefit from adopting a powerful speaking style, males rated men consistently higher than women, suggesting that, among male voters, men are always advantaged candidates. Furthermore, using data from the 2016 Presidential primary debates, I find some evidence that suggests that powerless speakers are more likely to be interrupted than their more powerful speaking counterparts.
This research has important implications for scholars of language in politics. Typically when we study language, we focus on what people have to saythe words they use to talk about political events. However, it appears that something beyond words also matters. It matters not just what a candidate says, but how they say it. Candidates who speak powerfully have a distinct advantage over those who speak in a powerless style. While many pundits and critics have bemoaned Barack Obama's use of the Teleprompter, it is likely that Obama's clear and direct style of speech leads to improved evaluations of his ability. Meanwhile, candidates who fumble over their words in debates or stump speeches may be viewed as less competent, less skilled, and less experienced.
This work also should have implications for other non-traditional political candidates. Black women typically face biases due to their hair and skin tone (Brown 2014). This highlights that minority women may face different concerns than white women. Indeed, minority women are more likely than minority men, and white women, to discuss issues related to gender in their official biographies (Brown and Gershon 2016), which may further serve to highlight their gender. Asian individuals are rated as less attractive when their English is heavily accented (Cargile and Giles 1998), and are rated as considerably lower-status than native English speakers (Cargile et al. 2010). This suggests that both race and gender should play a role in how speaking style is evaluated, something that should be further teased out in future research.
These effects, while quite large, were found in an artificial experimental environment and using a limited number of political debates. It is important to extend this work, in the future, further into the real political world. Future work could determine a more holistic view of candidate speaking stylebeyond simply looking at debate performance -, and determine whether real world candidates benefit electorally from adopting a powerful speaking style. Unfortunately, I am limited by available debate transcripts, and a small amount of women running for office where transcripts are available, to determine the effects of gender and speaking style using debate data. Additionally, while I have isolated a causal relationship, I am not able to compare the effects of speaking style to numerous other factors that influence candidate evaluations and vote choice. It is important to do so in order to contextualize how large of an effect speaking style has in the political world.
Speaking style seems to have an important impact on how candidates are evaluated. Even when a candidate says the same thing, it matters how he or she says it. When taking a clear, direct and powerful speaking style, candidates gain an advantage. While this advantage may matter less than things like partisanship and issue positions, in close elections, candidates must seek out any advantage they can find. These findings suggest candidates should focus on speaking in a clear and direct manner in debates in addition to the content of their answers. Notes 1. http://www.independent.co.uk/news/world/americas/donald-trump-phoenix-arizona-rallycrowd-leaves-incoherent-rambling-dishonest-media-border-wall-a7908436.html. 2. https://www.washingtonpost.com/politics/republicans-mock-obamas-teleprompter-use/ 2011/10/18/gIQA6hEivL_story.html?utm_term=.133db9ea178a. 3. Text retrieved from the Commission on Presidential Debates, https://www.debates.org/votereducation/debate-transcripts/september-23-1976-debate-transcript/. 4. Roughly 52% of the sample was female, 51% Democrat, 41% Republican, and 75% white with a mean age of about 20. For group assignment, 66 individuals were assigned to the powerless man condition, 82 to the powerless woman condition, 79 to the powerful man condition, and 74 to the powerful woman condition. This is a convenience sample, and care should be taken when generalizing these findings to an older, less educated population. However, given the substantive magnitude of many of these effects, I feel that external validity concerns in this study are minimal. 5. The actual text was adapted from a 2014 debate between Jerry Brown and Neel Kashkari for the California Gubernatorial election. 6. As such, this study does not allow me to determine the effects of speaking style with partisan cues. Given the power of partisan cues on candidate preference, I do not expect speaking style would overwhelm partisanship. 7. In the discussion of results, I use guidance from Bittner and Goodyear-Grant (2017) to discuss the role of gender and sex. For politicians, gender was evoked by only a masculine or feminine voice in the statements. For participants, they were asked to identify whether they are male or female. As such, I will refer to candidate gender, but participant sex in discussion of these results. 8. Full question texts are available in Appendix B. 9. All results are also robust to OLS models with controls for speaker gender, age, subject sex, subject race, ideology, and partisanship. These models are presented in Appendix C. 10. This variable ranges from 0 (lowest quality) to 24 (highest quality) with an overall mean of 12.41 and standard deviation of 4.96. 11. Participants were asked a question about their gender, but only given options to identify as male and female, matching more closely to the concept of sex. 12. Unfortunately, debate data does not allow a thorough analysis based on gender, as there were only 2 women candidates (Hillary Clinton, a Democrat, and Carly Fiorina, a Republican) among the 2016 presidential candidates. 13. http://www.presidency.ucsb.edu/debates.php 14. Here, markers of powerless style included hedges, intensifiers, hesitations, and tagging. A complete set of coding rules are available in Appendix D. The research assistant was provided these details on how to code the speeches and was blind to all hypotheses in the study during the coding. Coding instructions required the research assistant to only code for data that met criteria in the coding rules set forth in Appendix D. 15. All analyses are restricted to using number of interruptions by another debater as the dependent variable. Results are substantively and statistically similar when including number of interruptions by anyone (debater or moderator), but there are no statistical or substantive effects of powerless language on being interrupted by just the moderator. 16. Interruption behavior may indeed be gendered, with women more likely to be interrupted than men. However, in this dataset, there is little evidence to suggest this occurred for the two female candidates in the sample. Hillary Clinton was interrupted with similar frequency to Bernie Sanders, and Carly Fiorina was interrupted below the mean rate for Republican debaters. 17. Results are somewhat robust to excluding Donald Trump from these analyses. With an N of only 134, powerless speech predicts a 0.024 increase in interruptions in Model 1 (p=.11), and a 0.019 increase in model 2 (p=.08). Looking at each marker individually, only hedging and hesitations retain their statistical significance. This is perhaps unsurprising, since Trump used intensifiers more commonly than other debaters, but was not any more likely to use hesitation, hedging, or tagging. 18. Interruptions range from 0 to 36, with a mean of 3.17 and a standard deviation of 5.69.