“The jury is still out on Mitt Romney. He looked stiff, as he always does, and slightly nervous, which he sometimes does.” This point, made by a BBC journalist in 2012,Footnote 1 was echoed across the media. At the time, NPR broadcast a show called “Is Romney Nervous in London?” in which one of the guests argued that “He looked very nervous, very confused.”Footnote 2 Both reports hinted at the negative impact that his visible lack of assuredness might have on his image and campaign.
Previous research in psychology, communication and leadership studies suggests a strong connection between confidence, leadership (Shipman and Mumford Reference Shipman and Mumford2011; Northouse Reference Northouse2007) and persuasion (Birch, Akmal and Frampton Reference Birch, Akmal and Frampton2010). Journalists and political pundits have speculated that a display of nervousness hurts a politician’s electoral prospects. Yet there has been little systematic effort in the political communication literature to estimate the impact of nonverbal displays of confidence on candidate ratings. Moreover, the literature offers somewhat conflicting assessments of the importance of verbal and nonverbal communication for voters’ evaluations of candidates (for example, Hall et al. Reference Hall, Goren, Chaiken and Todorov2009; Nagel, Maurer and Reinemann Reference Nagel, Maurer and Reinemann2012).
Inferences about the impact of these dimensions are difficult to make for three reasons. First, much of the previous research is based on well-known political figures, making it difficult to discern whether the impact of appearing ill at ease is due to the behavior itself, to the fact that it is displayed by the particular politician or to the contrast with his/her opponent. Second, the findings are often based on experimental studies in which the nonverbal aspect of a speech is presented in isolation from other aspects of communication (for example, Hall et al. Reference Hall, Goren, Chaiken and Todorov2009; McHugo et al. Reference McHugo, Lanzetta, Sullivan, Masters and Englis1985; Spezio et al. Reference Spezio, Loesch, Gosselin, Mattes and Alvarez2012). Third, there is the possibility, rarely explored, that verbal and nonverbal cues might affect evaluations differently, depending on the type of evaluation being elicited (Noller Reference Noller1985). This article uses an innovative experimental approach to address these problems and provide a conservative estimate of the effects of nonverbal confidence on electoral persuasion. From a practical perspective, it helps answer critical questions about the role of appearing confident for political newcomers and party primaries.
The Impact of Confidence and Message Quality on Evaluations
There is strong empirical evidence linking nonverbal confidence to power and persuasiveness in various fields of advocacy research. In the politically proximate domain of leadership studies, (self-) confidence is listed as one of the main traits of effective leadership (Northouse Reference Northouse2007, chapter 2). Carli, LaFleur and Loeber (Reference Carli, LaFleur and Loeber1995) find that competent and dominant leadership styles are associated with nonverbal behavioral markers indicating confidence. Ames and Flynn (Reference Ames and Flynn2007) find evidence corroborating the relationship between “speaking up confidently” and perceptions of leader assertiveness. Awamleh and Gardner (Reference Awamleh and Gardner1999, 346) argue that a “confident presence” is a “key determinant of [leaders’] charisma.” In developmental psychology, Birch, Akmal and Frampton (Reference Birch, Akmal and Frampton2010) found that children as young as two recognize cues of confidence and are more willing to follow the instructions of a confident adult.
The impact of nonverbal confidence in politics has not been systematically explored, but nonverbal cues more generally have been found to play a key role in evaluations. Simple exposure to a candidate’s facial features, for example, can have an immediate and significant impact on ratings (for example, Riggio and Riggio Reference Riggio and Riggio2010; Rosenberg et al. Reference Rosenberg, Bohan, McCafferty and Harris1986; McHugo et al. Reference McHugo, Lanzetta, Sullivan, Masters and Englis1985), with an important part of the evaluation occurring automatically (Hall et al. Reference Hall, Goren, Chaiken and Todorov2009). Candidates’ non-facial appearance (for example, their hair and clothing) also matters (Spezio et al. Reference Spezio, Loesch, Gosselin, Mattes and Alvarez2012). Finally, candidates’ facial displays of emotion generate emotional responses that in turn mediate candidate evaluations (McHugo et al. Reference McHugo, Lanzetta, Sullivan, Masters and Englis1985; Sullivan and Masters Reference Sullivan and Masters1988; Stewart, Waller and Schubert Reference Stewart, Waller and Schubert2009). In light of this evidence, our baseline expectation is that:
Hypothesis 1: A candidate’s nonverbal confidence should lead to more positive voter ratings.
Nonverbal effects on political evaluations are likely to be strongest when people do not have access to any other information (such as in picture-based studies; for example, Noller Reference Noller1985), but we know from debate-based research that verbal messages can have powerful persuasive effects. An early study of the debate between Dole and Mondale (Krauss et al. Reference Krauss, Apple, Morency, Wenzel and Winton1981) found that viewers’ inferences could be traced primarily to the content of the candidates’ speeches. Similarly, a study of the 2005 German leaders’ debate found that the verbal information explained much more of the variation in viewers’ evaluation of the politicians than nonverbal cues (Nagel, Maurer and Reinemann Reference Nagel, Maurer and Reinemann2012). Awamleh and Gardner (Reference Awamleh and Gardner1999) also find that leaders gain from delivering visionary speeches. Message quality also has a positive impact on persuasion (for example, Mackie, Worth and Asuncion Reference Mackie, Worth and Asuncion1990; Chong and Druckman Reference Chong and Druckman2007; see Park et al. Reference Park, Levine, Westerman, Orfgen and Foregger2007 for a brief meta-analysis). High message quality is usually operationalized using “strong and valid [arguments]” (Mackie, Worth and Asuncion Reference Mackie, Worth and Asuncion1990, 814), “sound logic, valid reasons and presentation of data” (Park et al. Reference Park, Levine, Westerman, Orfgen and Foregger2007, 87) or “effective” arguments (Chong and Druckman Reference Chong and Druckman2007, 641). This leads us to our second hypothesis.
Hypothesis 2: Higher-quality messages should increase voters’ appreciation of the candidate.
However, as previous research often manipulates the verbal and nonverbal communication dimensions separately, it is unclear what the power of the message quality is relative to that of nonverbal cues. Evidence from the communication and issue-advocacy literature (Noller Reference Noller1985; Marsh, Hart-O’Rourke and Julka Reference Marsh, Hart-O'Rourke and Julka1997) raises the possibility that the relative weight of these dimensions varies with the types of judgments viewers are required to make. Noller (Reference Noller1985, 44) notes that “the importance of the verbal channel … is greater for more cognitive (rather than affective) tasks.” The affective tasks included evaluating a speaker’s likability, friendliness, sincerity and rating their attitude on a positive-negative dimension; the cognitive tasks involved making predictions about a speaker’s future behavior.
Hypothesis 3: The nonverbal expression of confidence has a stronger impact on affective judgments of the candidate (for example, their honesty, likability and friendliness) than on evaluations of their job qualifications (for example, competence, knowledge and intelligence).
Hypothesis 4: Message quality should have a stronger effect on assessments of candidates’ expertise than on affective evaluations.
There is broad scholarly agreement that members of the public potentially rely on multiple cues to evaluate and endorse political candidates (for example, McGraw Reference McGraw2011). Candidates’ personality traits are just one such dimension. Another is perceived electability, and these perceptions play an important role, especially in intraparty elections (Rickerhauser and Aldrich Reference Rickershauser and Aldrich2007). Electability may be particularly subject to communication effects, as voters must make inferences not just about the candidate’s persona but also about how other voters will react. Indeed, previous studies have found that the amount of media coverage (Bartels Reference Bartels1988), celebrity endorsements (Pease and Brewer Reference Pease and Brewer2008) and information about previous poll performance (Collingwood, Barreto and Donovan Reference Collingwood, Barreto and Donovan2012) can significantly influence these assessments. It is unclear, however, to what extent new candidates have non-mediated power to influence these perceptions themselves. In this study, we provide a first test of the impact of the verbal and nonverbal features of a candidate’s speech on their electability ratings. We expect both verbal and nonverbal aspects to matter, but given the lack of prior studies to provide guidance, we refrain from making any initial predictions about their relative importance.
The Experiment
To examine the effects of nonverbal confidence and message quality on political evaluations, 104 participants of voting age (85 percent students, 49 percent female, average age 22) were recruited from a major English-speaking Canadian university for a 2 (Nonverbal confidence: high vs. weak)×2 (Message quality: low vs. high) experiment administered on the computer. To eliminate the confounding effects of partisanship, participants were pre-selected to be supporters of the Liberal Party of Canada.
Manipulation
After completing a preliminary questionnaire, participants were asked to evaluate the content of the webpage of one of the Liberal Party candidates in the recent Canadian federal election, presented in the form of a video tutorial.Footnote 3 The tutorial started with a 20-second static shot of a message posted by the candidate showcasing his policy experience on environmental matters (see the Online Appendix).Footnote 4 This message ensured that all of the participants started with the same information about the candidate. After the static message, respondents saw one of four versions of a candidate-recorded webcast, which contained a campaign speech on environmental policies. The production details of the speech mirrored real candidate web productions on YouTube. Depending on their treatment group, participants watched the candidate deliver a high- or low-quality message with or without nonverbal confidence. The four webcast versions were of similar length and were produced in the studio in a single day, using the same professional actor to play the candidate and a professional crew to film and edit the material.Footnote 5
The Candidate
The actor hired to play the candidate was pre-selected from a list of five actors, based on ratings from an independent sample of 35 university students blind to the purposes of the study. The candidate was selected to be moderately attractive (to avoid the potential confounding influence of physical attractiveness) and to look about 50 years old.
The Message Quality Dimension
The high- and low-quality speeches differed in their organization of ideas and the quality of their arguments (see the Online Appendix for the texts). Following research by Atkinson (Reference Atkinson2004) and Awamleh and Gardner (Reference Awamleh and Gardner1999), the high-quality speech (hereafter labeled “V+”) used strong rhetorical devices: sequences of three (for example, “…three priorities: to save jobs, to save on energy consumption and to save on consumption of fossil fuels!”), position taking (“as a Liberal I say we can and we will!”) and repetition (for example, “one, we need to save jobs” […] “two, we need to save on energy consumption” […] “three, we need to save on consumption of fossil fuels”). The high-quality speech was structured according to Monroe’s motivated sequence (Cook Reference Cook1989). In line with previous studies (Mackie, Worth and Asuncion Reference Mackie, Worth and Asuncion1990; Park et al. Reference Park, Levine, Westerman, Orfgen and Foregger2007), the arguments were bolstered using concrete examples of cuts to environmental programs implemented by the opposition party. The low-quality speech (hereafter labeled “V−”) addressed the same ideas, but was poorly structured, lacked precise examples and used more passive voice (21 percent, compared to only 12 percent for the high-quality speech). The speeches were of about equal length (315 and 316 words).
The Nonverbal Confidence Dimension
Delivery was manipulated by instructing the actor to vary his body movements, facial expressions and voice pitch throughout the speech to convey a more or less confident attitude. The behavioral instructions were adapted from Carli, LaFleur and Loeber (Reference Carli, LaFleur and Loeber1995), Awamleh and Gardner (Reference Awamleh and Gardner1999) and Marsh, Hart-O’Rourke and Julka (Reference Marsh, Hart-O'Rourke and Julka1997).
For the confident delivery (N+), the actor was instructed to sport an upright, but slightly backward-leaning and open body position; to keep his hands relatively apart from his body; to make firm hand gestures coordinated with the speech; to display vivid emotional facial expressions; to speak loudly and with continuous debit; and to articulate clearly. These nonverbal elements are signs of confidence associated with the competent, dominant and social leadership style types proposed by Carli, LaFleur and Loeber (Reference Carli, LaFleur and Loeber1995). We decided not to have the actor display confidence markers consistent with just one leadership style, because each of these styles also contains aspects that might diminish persuasion. For example, confidence in the dominant style is also associated with threat; in the competent style, confidence displays can suggest a lack of emotional involvement; and in the social style openness can be associated with weakness. The portrayal of confidence in our study closely mirrors the “strong delivery” used by Awamleh and Gardner (Reference Awamleh and Gardner1999) in their study of charismatic leadership.
In the non-confident delivery (N–), the actor was instructed to adopt a rather slumped, forward-leaning and closed body position; to exhibit little facial emotion; to keep his hands close to his body; and to speak softly and monotonously with verbal stumbles and hesitations.
Manipulation Checks
We performed several manipulation checks to establish both the validity and the credibility of our manipulation. First, we analyzed the candidate’s nonverbal behavior in each of the four conditions using a visual analysis computer program (Dumitrescu Reference Dumitrescu2010). We selected a large number of frames (N=275) taken randomly every two seconds from each video and used the software to obtain pixel-level estimates of the positioning of the candidate’s head and hands in the frames. As expected, the candidate held his head up higher (F (1, 271)=649.36, p=0.00), opened up his body more (F (1,271)=85.46, p=0.00) and gesticulated more in the N+ conditions than in the N– ones.
Second, after completing all the dependent variable questions in the post-questionnaire, we asked respondents to rate the delivery of the speech on a scale ranging from non-confident (–5) to confident (5). Ratings were higher in the N+ than in the N– conditions (F(1,100)=49.34, p=0.00). There were no significant verbal or interaction effects. Participants’ comments about the candidate made at the end of the study contained no indications that they suspected fakery.
We also validated our manipulation using independent ratings. After the study, we recruited an independent sample of 103 students from the same university who were blind to our manipulation. They watched silent excerpts from the webcasts and rated the politician’s confidence on a 1 to 10 scale. The candidate’s behavior was rated as more confident in the N+ conditions than in the N– ones, irrespective of the verbal valence of the original webcast (t>2.51, p<0.03). The students were also randomly assigned to read one of the two speeches. The V+ speech received better evaluations for the strength of its arguments (|t|=3.27, p=0.00), structure (|t|=4.22, p=0.00), clarity (|t|=4.02, p=0.00) and persuasiveness (|t|=1.83, p=0.07).
Dependent Variables
After watching the speech, we asked respondents “how well does each of the following terms describe the candidate whose speech you have just watched?” The terms were: likable, honest, warm, competent, knowledgeable, intelligent, strong leader. Each term was placed in a separate row in a table (in random order) and was ranked from 1 to 5. Subsequently, we also asked “Compared to other Liberal candidates, do you think the candidate had a much better chance, a somewhat better chance, same chance, somewhat worse chance, much worse chance of “persuading undecided voters to vote for the party” and “winning their constituency.” This question was placed on a separate page, and the order of the two items was randomized.
Based on the results of a principal component analysis (see results in the Online Appendix), we created three separate scales: an affective rating index (containing evaluations of likability, honesty and warmth, alpha=0.70), a qualifications rating index (containing evaluations of competence, intelligence and knowledge, alpha=0.74) and an electability rating index (containing evaluations of leadership, persuasive power and district viability, alpha=0.82).
Results
Twelve participants who expressed either high discomfort during the study (see fn. 3) or did not follow the initial reading instructions were removed from the analyses. The final analyses are performed with a sample of N=92 (49 percent female), equally distributed by condition. A randomization check revealed no significant between-conditions differences in terms of participants’ age, environmental opinions, strength of Liberal identification and personality. We therefore perform two-way ANOVAs on our dependent variables with just the message quality and the nonverbal confidence as independent factors. Figure 1 graphs the mean evaluations with 95 percent confidence intervals. The results are available in the Online Appendix.
Mean predicted candidate affective, qualifications and electability ratings (with 95 percent confidence intervals) by experimental condition Note: N+ indicates a confident attitude, N– indicates a non-confident attitude, V+ indicates a high-quality speech and V– indicates a low-quality speech.
The results generally support Hypotheses 1 and 2. Both the confident delivery and the message quality affect the qualifications index and the electability index. A confident delivery improves evaluations of a candidate’s qualifications by 0.30 on a 1 to 5 scale, and evaluations of a candidate’s electability by 0.52, all else equal. A high-quality speech improves ratings of qualifications by 0.31 and electability by 0.28. For the qualification ratings, the nonverbal confidence dimension and the message quality dimension are each responsible for about 50 percent of the explained variance.Footnote 6 For the electability ratings, nonverbal confidence is responsible for about 70 percent of the explained variance, while the quality of the message is responsible for about 20 percent. The interaction between these two dimensions, which approaches significance, accounts for about 10 percent. In R-squared terms, the amount of variation explained by the communication variables is 0.11 and 0.17, respectively, for the qualifications and electability ratings. Admittedly, these are modest effects. But given that we only sampled voters who felt positively about the candidate to begin with, these results suggest that paying attention to these communication variables is important even when a candidate presents himself to a sympathetic crowd.
Contrary to our expectations, we do not find any communication effects on affective ratings. In fact, ratings on this index are very close to the midpoint, and vary little by condition. With hindsight, this could be explaining the verbal and nonverbal manipulation, as we did not include any elements that might indicate that either the candidate or somebody else (his political opponents, individually or as a party) was dishonest. Nor did the candidate’s performance include any explicit displays of friendliness or effusiveness. It is possible that expressing confidence affects ratings of honesty, likability and warmth, but that the effects are too weak for our experiment to be able to detect them.
Hypothesis 3 predicted that nonverbal confidence would play a stronger role in affective ratings than in qualifications ratings, but it turns out, as we have already discussed, that nonverbal confidence only influences impressions of qualifications. Hypothesis 4 is supported, as we find that the quality of the message explains significantly more of the variation in qualifications ratings than in affective ratings.
We did not make any predictions about the relative effect of the two aspects of communication on electability, but we find that both have a significant effect. Interestingly, nonverbal confidence plays a significantly stronger role than message quality in explaining the variation in electability ratings. Nonverbal confidence also has a stronger effect on electability ratings than on qualifications ratings.
Finally, it is worth noting that an interesting pattern emerges from post-hoc comparisons between the different conditions. For both qualifications and electability, the condition that clearly pushes the candidate ahead is the one in which he delivers a high-quality message with confidence. The relevance and implications of this finding are addressed in the conclusion.
Discussion and Conclusions
We conclude by focusing on three aspects: the relevance of studying nonverbal confidence in politics, the implications of the method utilized and the implications of the findings.
First, this study makes an empirical contribution to the study of nonverbal communication and political persuasion by exploring a previously untested relationship in an experimental setting: the relationship between nonverbal confidence and political ratings. On the one hand, as the initial newspaper story about Mitt Romney suggests, candidates are constantly scrutinized for the overall display of confidence in their public interactions. Thus given its association with power and leadership, understanding the impact of confidence is highly politically relevant. On the other hand, the focus of this study on a broader attitudinal display (rather than on specific emotional displays) gives us a better grasp of the importance of nonverbal communication for persuasion.
This research also makes a methodological contribution. The main concerns driving the experimental protocol were control and realism. We did not want to expose participants to versions of a candidate’s speech that obliterated either the visual or the verbal part, because in real life, people will typically be exposed to both. However, we attempted to isolate the effects of the communication aspects as carefully as possible. We deliberately recruited participants to match the politician’s party. We also exposed them to just one version of the candidate’s performance, which differed only in the verbal and/or nonverbal aspect from other versions. Finally, we specifically selected an unknown candidate, about whom participants had only a handful of details.
These methodological choices have two implications. First, our results are limited to a specific context, in which a new, white male candidate presents himself to a sympathetic public. It is possible that confidence, when expressed by better-known politicians (or in reaction to other politicians) might have stronger or weaker effects. In particular, the effect of confidence in real life might be moderated by a candidate’s history. For a candidate who always acts very confidently, lacking in confidence in some instances might be more damaging. It might also depend on the context in which the candidate expresses or fails to express confidence. Thus it might have been more important for Romney to act confidently, to prove he is “presidential,” than for a legislative or local candidate. The candidate’s gender, ethnicity and age might also matter. It is likely that a voter’s party affiliation also matters, with unaffiliated voters more likely to take cues from confidence than voters who already have their minds set. Finally, it is also possible that displaying confidence might be more important for new candidates, as people need more information about them. These research ideas should be pursued in future studies.
Our methodological approach also has a second implication, which we think points to an important contribution. The act of communication is intrinsically complex. Logically speaking, we cannot disentangle the effects of confidence (or any other nonverbal attitude) from those that can be attributed to fame, individual personality, demographic characteristics or political context if we do not have a benchmark for nonverbal effects when these contextual and individual-level variables are held more or less constant. This study makes a step toward establishing such a benchmark. In doing so, it provides an extremely conservative test of the impact of confidence and message quality.
Our results show that nonverbal confidence and message quality have an effect even when variations in persuasion seem very unlikely, that is, when voters agree with the politician and share his party affiliation. The results also suggest that confidence pushes a candidate’s evaluations up even on dimensions that should be most affected by the verbal content of the communication, such as evaluations of his job qualifications. Moreover, we find that showing confidence has the strongest effects on impressions of how the candidate will be perceived by others and how he will fare in the future (his electability). These findings suggest that future studies should differentiate between various aspects of voters’ evaluations.
Finally, the improvement in ratings induced by a confident attitude or the message quality is, by itself, not always substantial. In fact, displaying confidence only helps attain a significantly higher rating when it is associated with a high-quality message. This does not necessarily mean that confidence and message quality have no effect when they are exhibited separately, since our relatively limited sample might fail to detect small effects. But it strongly suggests that the effect of communication is much stronger when confidence and message quality go together. This finding confirms, on the one hand, conventional wisdom. But on the other hand, it has some interesting normative and practical questions. It is good news that people do not consider a candidate to be more qualified and more electable when he exudes confidence but delivers a low-quality message. This suggests that, while the nonverbal information might be processed first, the processing does not stop there. At the same time, it is bad news that a high-quality message fails to elicit more support when delivered with a lack of confidence. This reaffirms the need to know more about when and why unspoken confidence matters in politics. So, while this study casts initial light on the question of the role of nonverbal confidence in electoral persuasion, it also opens up an exciting research agenda.
