The Tone Dilemma: Comparing the Effects of Flattery and Verbal Aggression in a Political Speech

In the realm of political communication, the effects of personal verbal attacks on political opponents have long been studied. However, less well understood are the effects of flattery on such opponents. I present an experiment showing that praising a political opponent elicits an audience’s positive emotions, which in turn positively influences source trustworthiness, and ultimately increases the likelihood of voting for that source. In contrast, attacking an opponent elicits aversion, which in turn negatively influences source trustworthiness, thus reducing the likelihood of voting for the source.

advantageous for a political candidate (Lau, Sigelman, & Rovner, 2007). In this regard, a critical distinction has been made between messages attacking an opponents' personality or behavior (i.e., person-based), and those pointing to political programs (i.e., issue-based). The former strategy risks backfiring on the source impression, whereas issue-based criticisms seem less risky, or even useful (Lau & Pomper, 2001).
To capture the effects of such negative strategies, experimental studies often include a "positive" control condition, in which the source develops the message around his or her own characteristics or political proposals. Considerably less attention has been devoted to studying the effects of symmetrical "positive" messages, that is, those in which the source flatters the rival. This scant attention is probably due to the fact that it is not common to find a political debate where a politician flatters the other. However, given the backfire effect induced by the negative personal attack and the persuasive effects of arguing against one's own interest (J. Burgoon, 1993;Petty, Fleming, Priester, & Feinstein, 2001), it may well be that such strategies are beneficial.
In an attempt to provide this insight, I present an experimental study directly comparing the effects induced by a fictitious politician who, during a speech, simply mentions his or her rival (control condition), or addresses a verbal attack (negative message condition) or a compliment (positive message condition) to the rival.

Personal Attack Effects
Research about negative campaigns has shown that personal attacks rarely achieve the goal of convincing the audience of the negativity of the target; instead, they may provoke a negative judgment about the source (e.g., Budesheim, Houston, & DePaola, 1996;. Such a boomerang effect has been explained both as the outcome of an inference mechanism (e.g., Carraro, Gawronski, & Castelli, 2010), and as the effect of an emotional reaction (Russo, 2011(Russo, , 2016. Indeed, Gawronski and Walther (2008) showed that individuals observing one person evaluating another (either in positive or negative terms) tend to attribute the observed evaluation to the source. For example, people tend to infer that someone conveying negative information about other people is unlikeable. This has been called the "transfer of attitudes recursively" (TAR) effect (Gawronski & Walther, 2008). In this line of reasoning, Carraro, Gawronski, et al. (2010) found that respondents expressed a less positive explicit evaluation of a fictitious politician when he criticized his opponent, in comparison with when he positively described himself.
In addition, scholars also have shown that negative propaganda strategies stimulate in the audience negative emotions targeted in particular toward the source (e.g., Haddock & Zanna, 1997). According to affective intelligence theory (Marcus, Neuman, & MacKuen, 2000), people react to political information through the activation of different emotions which can be clustered into two emotional systems: dispositional and surveillance. In particular, enthusiasm and aversion are part of the dispositional system that activates habitual approach and avoidance in respect to the source of those emotions. In contrast, a reaction of anxiety activates the surveillance system, warning that something potentially threatening needs more conscious attention. In this case, the attention will be focalized on the source of the threat. Thus, within the framework of affective intelligence theory, recent studies (Russo, 2011(Russo, , 2016 found that personbased attacks, especially when coming from a candidate of the preferred party, elicit aversion toward the source that mediates the (negative) effect on the likelihood of voting for that source.
However, studies on the effect of negative propaganda are often based on experimental designs which compare the effects of negative political messages with those of "positive messages," that is, messages focused instead on the political program promoted (e.g., Arceneaux & Nickerson, 2010), or on the competence and morality of the source (e.g., Carraro, Gawronski, et al., 2010). Therefore, the experimental conditions differ both in the message tone (positive vs. negative) and in the target of the communication (source vs. opponent; source's vs. opponents' political program). To date, we do not know whether a genuinely positive statement, that is, a compliment flattering the political rival may induce positive effects on the source, since empirical evidence issued from a direct comparison between negative and positive person-based claims are still lacking.

Flattery Effects
Addressing a compliment to someone is a means of ingratiating the interlocutor and affecting source evaluation and target compliance (Jones & Wortman, 1973). The persuasive effects of flattery have been mainly studied in the realm of interpersonal communication or commercial interaction (Grant, Fabrigar, & Lim, 2010;Gordon, 1996;Seiter, 2007;Seiter & Dutson, 2007). Beyond the influence of flattery on relevant targets, it has been shown that a flattering remark can have an effect that spread to observers, when it is expressed in the presence of other people who then form their own impression of the source (Campbell & Kirmani, 2000;Vonk, 2002). This is interesting for the case of a political candidate who flatters the rival, not in order to gain his or her approval, rather to gain positive evaluation and acceptance from the observing audience. Unfortunately, sound evidence for this effect is, at best, scant.
A recent experimental study carried out in the political domain included the manipulation of appreciation directed to a rival during a public speech (Cavazza, 2016). Results showed that being exposed to a political candidate who flattered the rival (vs. control condition without the appreciation) improved the perception of source trustworthiness (but did not affect perception of source competence), and indirectly increased the likelihood of voting for that candidate. These findings were interpreted in line with the TAR effect (Gawronski & Walther, 2008). In this case, observers tend to infer that sources who express positive evaluation of others are likeable. However, a different interpretation could be advanced. Flattery directed at political opponents would come as a surprise to people attending a candidate's speech, as politicians tend to attack their counterparts. According to the language expectancy theory (M. Burgoon & Miller, 1985) when a source conforms more than expected to social norms (i.e., to be kind to one's own interlocutors), she or he gain positive evaluation by the audience. Moreover, praising a rival is seemingly contrary to candidates' self-interests. Persuasion research suggests that the violation of expectations regarding what the source would argue, and in particular arguing against self-interest, provokes evaluations of source trustworthiness and message validity (e.g., J. Burgoon, 1993;Cialdini, 1984;Eagly, Wood, & Chaiken, 1978;Petty et al., 2001). Therefore, Cavazza's study (2016) still left some important questions unanswered. First, could the surprise caused by flattery have a role in making the strategy effective and positively influence the audience?
Second, the previous study did not directly compare the effects of flattery with those of personally attacking the rival. Also, Cavazza's (2016) study did not take into consideration the role of emotional reactions elicited in the audience by a positive claim about the rival. Could flattery of a rival elicit positive emotions (enthusiasm) with the potential to backfire on the source, just as verbal attacks do?
Finally, the previous study did not consider relevant audiences' evaluations of the flattered target. Indeed, one reason why politicians are so reluctant to flatter opponents may be the potential risk of contributing to their opponents' positive image, which might outweigh their own benefits. The present study seeks to answer these questions.

The Present Study
In the present study, the message tone was experimentally manipulated, whereas the target of the positive/negative sentence holds the same (i.e., the political rival). Therefore, I compared the effects exerted by a political speech in which the source (a politicians) neutrally evoked the opponent (control condition), with those induced by the same speech where a deprecation of the rival (negative message), or an appreciation of the same rival (positive message) was embedded. Since people also have different normative expectations about appropriateness of communication style for males and females, and verbal aggression in persuasive messages are more expected for male than for female source (M. Burgoon, Dillard, & Ooran, 1983), I also controlled for potential effects of source gender, through the attribution of the message to a male versus a female candidate.
Following the literature reviewed above, the general hypothesis is that flattery (vs. person-based attack) of a political opponent facilitates a positive evaluation of the source candidate, and in turn enhances the likelihood of voting for him or her. The theoretical framework and previous studies illustrated above suggest that this may be due to different mechanisms. Drawing on studies about the effects of negative messages, we could expect the following: Hypothesis 1a: As a result of the TAR effect, flattery should induce general source liking, and source trustworthiness, just as attacks undermine them (Carraro, Gawronski, et al., 2010), whereas, based on previous studies concerning both flattery (Gordon, 1996) and political communication effects (e.g., Castelli, Carraro, Ghitti & Pastore, 2009;Cavazza, 2016;Cislak & Wojciszke, 2008), source competence evaluation should remain unaffected. Hypothesis 1b: Source liking and source trustworthiness should in turn increase the likelihood of voting for the source (i.e., indirect or mediation effect of flattery/verbal attack on intention to vote through source liking and source trustworthiness).

Hypothesis 2a:
Flattery might elicit a positive emotion (i.e., enthusiasm), just as verbal attacks elicit aversion (Russo, 2011(Russo, , 2016, whereas no influence should be observed for anxiety. Hypothesis 2b: Following the affective intelligence theory, enthusiasm and aversion emotions of the dispositional system should drive participants' attention to the source instead of the target and backfire on source liking and trustworthiness. These evaluations in turn may influence the likelihood of voting for the source (two-step indirect effect of flattery/verbal attack on intention to vote through enthusiasm/ aversion, and source trustworthiness).

Moreover,
Hypothesis 3a: Flattery toward a political rival might be perceived as a positive expectancy violation, whereas attacks represent a more expected strategy. Hypothesis 3b: Based on the persuasive effects of expectancy violation (J. Burgoon, 1993;Cialdini, 1984;Eagly et al., 1978;Petty et al., 2001), we should observe a two-step indirect effect of the flattery (vs. attacks and control message) on the likelihood of voting for the source through the positive expectancy violation that should increase source trustworthiness.
These hypotheses are not mutually exclusive. On the contrary, the effects of these different mechanisms could be cumulative.
For all three mechanisms, receivers' attention should be focused on the source, and previous studies have, in fact, found that attacks do not achieve the goal of discrediting the target. This is the reason why we also expected a negligible effect of flattery on the audience evaluation of the target.

Participants
One hundred and thirty-eight Italian adults (65.4% women) aged 19 to 65 years (M = 31.51, SD = 11.87) were recruited through personal mailing lists, Facebook contacts, and snowball sampling. They were highly educated (39.2% of them have a high school degree and the remaining have a university or master's degree). They were also predominantly slightly progressives (51.2%), and moderates (22.9%), and mostly workers (45.6%) and students (36.8%). Participants were asked to complete an online questionnaire about political communication implemented on the LimeSurvey platform.

Design and Procedure
Participants were randomly assigned to the three conditions (type of message: flattering vs. attacking vs. control) of a single factor between participants design.
Having answered three questions about their political interest, information, and their participation in politics (political engagement index, Cronbach's α = .88), participants read the following instruction: "Now, imagine that in a few months a local election will be held in your town, there are two candidates running for mayor. While exploring the web to collect information, you find their pictures, some personal information and a speech one of them has given in a recent public meeting during the campaign." Some personal information was provided about the two candidates (age, education, work, marital status, and children) along with their pictures.
Next, participants read the passage (about 200 words long) from a speech about the relationship between young people and politics that had allegedly been given by one of the candidates during a meeting (the same passage used in Cavazza's study, 2016, see the appendix). In the control condition, the second sentence of this passage read: "I believe that my competitor will agree with me about the need to change this situation." In the two experimental conditions, this same sentence included also a source's opinion about the competitor (i.e., "I believe that my competitor, who is an upright and smart person, will agree . . . " vs. "I believe that, even though my rival has often proved to be unreliable and incompetent, will agree . . . "). I also controlled for a possible effect of a candidate's gender, manipulating the name of the source candidate and the competitor so as to refer either to two men or to two women (Paolo Albertelli vs. Mario Gambettini and Paola Albertelli vs. Maria Gambettini, respectively).
After reading the speech, participants answered a series of questions aimed at assessing whether the message was perceived as benevolent/contemptuous toward the rival, how much it violated expectations, source trustworthiness, competitor trustworthiness, emotions elicited by the message, and likelihood of voting for the source candidate. Finally, they reported their position on the left-right political spectrum and completed a standard sociodemographic form.

Dependent Measures
Perception of Message Expectancy Violation. As a check on whether the candidate's message induced expectancy violation as a function of the experimental conditions, participants responded on a 7-point scale ranging from 1 = not at all to 7 = very much, to indicate the extent to which they perceived the message as surprising, typical for a politicians and expected. An index of message expectancy violation was computed as mean of the three items (α = .68).
Emotions Elicited by the Message. Following Marcus, MacKuen, Wolak, and Keele (2006) and Russo (2011Russo ( , 2016, participants' emotional reaction in terms of enthusiasm, anxiety, and aversion was measured. For each of the three dimensions of emotional response, two items were included for which participants had to rate how much the message made them feel (from 1 = not at all to 7 = extremely). The scores (after reversing two positive items) were submitted to a factor analysis, forcing a three-factor solution. Oblique rotation was used because factors were expected to be correlated. The first factor (explained variance 41.74%) included the two items supposed to tap aversion (disgust and anger) with factor loadings >.93. The second factor (explained variance 28.02%) included the two items supposed to tap enthusiasm (hope and joy) with factor loadings >.91. The third factor (explained variance 15.30%) included the remaining two items supposed to tap anxiety (worry and fear) with factor loadings > .93. On this basis, the three emotional indexes were computed as item means.
Impression About the Source. Participants evaluated how well they thought six adjectives (sincere, reliable, dishonest, skilled, unqualified, and uninformed) described the speaker on a 7-point scale (from 1 = not at all to 7 = very well). Following previous studies (e.g., Abelson, Kinder, Peters, & Fiske, 1982;Cavazza, 2016;Wojciszke & Klusek, 1996), after reversing the score for the negative item, I computed a source trustworthiness index (α = .71) and a source competence index (α = .63). Furthermore, one item measured the general liking of the source of the message on a 7-point scale (from 1 = not at all to 7 = very much).
Impression About the Competitor. The same six adjectives were used to capture participants' impressions about the competitor. Because of low reliability values of the competitor trustworthiness index (α = .31), and competitor competence index (α = .39), the analysis was performed on a global index of impression about the competitor (α = .58).
Likelihood of Voting for the Source Candidate. Two questions were put to participants regarding the likelihood of voting for the candidate if they were members of the town constituency (r = .89).

Preliminary Analyses and Manipulation Checks
Preliminary checks on the effect of participant's gender, political engagement, and self-reported orientation on the left-right political spectrum showed significant differences of participants' political engagement in the three experimental conditions, F(2, 132) = 9.96, p < .001, η 2 p = .13. For this reason, political engagement score was included as a covariate in all the subsequent analysis.
Analysis of variance (ANOVA) on the two items evaluating the tone of the message showed that the speech containing the flattery was perceived as more benevolent (M = 4.53, SD = 1.32) than that containing the deprecation (M = 3.13, SD = 1.67), but not different from the neutral one (M = 4.55, SD = 1.35), F(1, 132) = 14.12, p < .001, η 2 p = .18. In the same way, the flattering message was perceived as less contemptuous toward the rival (M = 2.69, SD = 1.69) than the attacking one (M = 4.35, SD = 1.66), but not different from the neutral one (M = 2.41, SD = 1.37), F(1, 132) = 19.90, p < .001, η 2 p = .23.  Tone manipulation marginally influenced both general source liking, F(2, 134) = 2.75, p = .07, η 2 p = .04, and the perception of the source trustworthiness, F(2, 134) = 2.52, p = .08, η 2 p = .04, in the hypothesized direction, whereas no differences emerged in terms of source competence, F(2, 134) = 0.86, p = .43 (Hypothesis 1a). 1 In line with Hypothesis 2a, the analysis on enthusiasm revealed the main effect of tone manipulation, F(2, 134) = 5.99, p = .003, η 2 p = .08. The flattering message elicited more positive emotions as opposed to the attacking one, even though none of them differed from the neutral message at the post hoc Bonferroni test. The same analysis on aversion showed the significant influence of message manipulation, F(2, 133) = 3.37, p = .04, η 2 p = .05. The attacking message elicited more aversion than the flattering one, with neither of them differing from the control message at the post hoc Bonferroni test. In contrast, no significant effect was observed on anxiety, F(2, 134) = 0.70, p = .50.

Direct Effects of Message Manipulation
Contrary to Hypothesis 3, the same analysis on the perception of expectancy violation revealed no significant effects, F(2, 134) = 0.25, p = .78.
Finally, the independent variable did not directly affect either the impression about the competitor, F(2, 126) = 0.50, p = .61, or the likelihood of voting for the message source, F(2, 134) = 1.16, p = .32.

Indirect Effects of Message Manipulation
The weak or null influence of message manipulation on general source liking, source trustworthiness, competence evaluations, and expectancy violation seen above entails the rejection of Hypotheses 1b and 3b on the possible indirect effects of flattery or attack on the likelihood to vote for the source through these respondents' reactions.
To test Hypothesis 2b, two dummy variables were computed from the message manipulation: one for the flattery condition (coded 1) versus the others (coded 0), and one for the attacking conditions (coded 1) versus the others (coded 0). I then verified two two-step models whereby the flattering/attacking message induced enthusiasm/aversion, which in turn favors/undermine the perception of source trustworthiness, ultimately affecting the intention to vote for the source. To this end, I ran Model 6 of PROCESS, the SPSS Macro by Hayes (2013), setting 5,000 bootstrapped samples. This analysis allows testing of three indirect paths. The first includes only enthusiasm/aversion as the intermediate factor, the second includes only perceived source trustworthiness as the intermediate factor, and the third includes the complete sequence depicted in Figure 1.
Tables 3 and 4 show that the candidate flattering or attacking the rival indirectly influenced the likelihood of receiving votes. When the message was flattering, the effect passed through positive emotion (enthusiasm), which also entailed source trustworthiness ( Figure 1a). When attacking the opponent, the source elicited aversion, which in turn weakened perceived source trustworthiness, thus reflecting the influence on intention to vote for the candidate (Figure 1b).

Discussion
The findings of the present study supported the general hypothesis concerning the persuasive effects of praising political opponents, independent of candidates' gender. This is consistent with a previous study in the same field (Cavazza, 2016), which found that a candidate addressing a flattery to the rival in a political speech enhanced the audience perception of source trustworthiness and indirectly, willingness to vote for that source. In addition, the present study expands this evidence in three areas. First, it contributes to a broader understanding of the process through which this effect occurs. Previous studies (Carraro, Gawronski, et al., 2010;Cavazza, 2016) suggested that an attributional process (i.e., TAR effect; Gawronski & Walther, 2008) could be responsible for the source experiencing a  backfire effect for attacking and flattering the rival. In the present experiment, the influence of the message manipulation on source evaluation (i.e., general liking, trustworthiness, and competence) was rather weak. Instead, the stronger influence of the positive versus negative tone manipulation was observed on the receivers' emotions of enthusiasm and aversion. Therefore, the present findings supported Hypothesis 2 by showing that negative and positive political messages stimulate the receivers' dispositional emotional system as suggested by the affective intelligence theory (Marcus et al., 2000). These kinds of emotion do not signal that some threat is at stake; they keep people's attention on the message source, thus, affecting evaluation of the candidate and intention to vote. Thus, participants' positive and negative emotional reactions to the message seemed to activate the inference process to which the TAR effect refers, even though enthusiasm was sufficient to strengthen likelihood of voting for the source. Moreover, the compliment of the rival did not appear as a surprise (i.e., positive expectancy violation) to our participants, who were not particularly persuaded by the candidate because she or he was arguing against his or her own self-interest.
In synthesis, the present results confirmed that a genuinely positive appeal (i.e., flattery) is able to induce similar effects to those provoked by verbal aggression directed at a political opponent. Just as verbal attacks on the personal characteristics of a political rival often backfire on the source, because they elicit aversion (Russo, 2011(Russo, , 2016 and undermine source trustworthiness (Carraro, Gawronski, et al., 2010), brief appreciation of a rival in a speech revealed similar but positive effects.
Second, the present experimental paradigm, keeping the target of a compliment or attack constant, helps disentangle the effects of the positive versus negative message tone from those due to different message targets (source candidate vs. rival). While previous studies (e.g., Carraro, Gawronski, et al., 2010;King & McConnell, 2003;Lau et al., 2007) have already shown that attacking a political rival is a risky strategy in comparison with promoting one's own qualities or proposals, our results also highlighted that praising a rival may be less risky than insulting him or her.
Third, the present findings seem to suggest that a positive tone may well be a worthwhile communication strategy, as highlighting some good qualities of the competitor did not influence participants' impressions about him or her. 2 Although empirical evidence about the persuasive effect of flattery in the political domain is increasingly gaining traction, application of the present evidence to real situations requires caution, because results are issued from a very simplified political scenario. Participants did not know the political affiliation of candidates, and this may be crucial. However, revealing information about candidates' political orientation risks obscuring the influence dynamics. Future studies could overcome this limitation through, for instance, an experimental scenario of primary elections.
Furthermore, the results concerning indirect effects may suffer from the fact that observed constructs share a common method of measurement; thus, the intervention of a possible common variance bias could not be excluded (Podsakoff, MacKenzie, Lee, & Podsakoff, 2003).
Despite these limitations, the present findings contribute to the literature on the effects of political communication strategies, which until now focused on effects of verbal personal aggression and criticisms, in comparison with completely different messages (e.g., messages supporting a political proposal). The present findings also broaden knowledge on the potential of flattery as a persuasive tactic, providing evidence that observed flattery, previously studied in interpersonal context, may also work in one-to-many political communication context. In this regard, a promising avenue for future studies is to test the consequence of another "positive" tactic political leaders are used to employ in their public speeches: to directly flatter the audience.