Harnessing Anger to Persuade: The Moderating Roles of Retributive Efficacy and Prior Attitudes

ABSTRACT

Given unresolved questions about the conditions under which anger appeals persuade, this investigation examined whether an anger appeal is more persuasive if it includes efficacy content emphasizing retribution. It also examined whether effects hinge on initial attitudes toward the advocated issue – here, regulation of soda marketing to improve children’s health. Findings from this 3 (offense component: high, low, none) $\times$ 3 (efficacy component: retributive, non-retributive, none) between-subjects experiment (N=1,244) indicated a high offense component unintentionally generated anger toward the message source among counterattitudinal individuals but, intriguingly, also boosted their behavioral intentions. Exposure to any efficacy cues increased policy support. Although the offense and efficacy components did not interact, post hoc analyses revealed the expected pattern whereby message-intended anger more strongly predicted policy support at higher levels of retributive efficacy beliefs. These findings suggest promising new directions to revisit predictions made by extant theories of anger appeals.

Anger can facilitate thoughtful cognitive processing (Nabi, 2002) and spur political action to rectify social injustices (Turner, 2007). As such, whereas ambient anger unrelated to the task at hand can be deleterious for decision-making (Yip & Schweitzer, 2019), intentionally evoking anger through persuasive messages that highlight a blameworthy offense may be a promising strategy to encourage audiences to partake in activism around social issues. Yet meta-analytic findings paint a complex portrait of the conditions under which anger appeals can persuade, suggesting that only under certain circumstances can anger appeals lead to attitudinal or behavioral change (Walter et al., 2018).

In particular, two moderating factors may help elucidate the boundary conditions of anger appeals’ influence: efficacy cues in the message and audience’s prior attitudes. Regarding efficacy, to effectively translate emotion into attitudinal or behavior change, persuasion theories (e.g., the Extended Parallel Process Model, the Cognitive-Functional Model) argue an emotional appeal should elicit the target emotion and also provide efficacy cues about ways to address the situation (Dillard & Nabi, 2006; Nabi, 1999; Turner, 2007; Witte, 1992). These efficacy cues should match the motivational goal for the emotion evoked in order for the message to be maximally effective (Dillard & Nabi, 2006). The efficacy component of a threat appeal, for instance, should focus on how the recommended response will offer protection against the threat, given that protection is the motivational goal underlying fear (Bigsby & Albarracín, 2022; Frijda et al., 1989; Lazarus, 1991). Regrettably, researchers have not yet formally tested this strategy in studying the effects of messages that appeal to anger. Anger motivates the individual to seek retribution against the perceived offender (Frijda et al., 1989; Lazarus, 1991), so it follows that the anger-eliciting component of an anger appeal should be relatively more persuasive if it is accompanied by an efficacy component focused specifically on how taking action will effectively punish the offender. Because researchers have yet to formally test this proposition in the context of anger, extant studies are likely underestimating the potential for anger appeals to persuade.

Additionally, an appeal to anger may not be persuasive for all people. According to the Anger Activism Model (Turner, 2007), an anger appeal should be most likely to persuade audiences that are initially favorable toward the advocated issue, but for counterattitudinal audiences, an anger appeal should elicit not anger toward the wrongdoer depicted in the message but anger toward the messenger, consistent with the notion of psychological reactance (Brehm & Brehm, 1981), which should result in diminished persuasion outcomes. Despite the intuitive premise that anger appeals should only persuade pro-attitudinal groups, surprisingly little work has investigated this possibility (Walter et al., 2018). This study marks an important step forward in the literature by formally testing whether initial attitudes moderate the persuasiveness of anger appeals.

The current experiment sought to fill these gaps in the persuasion literature, testing whether offense components of an anger appeal are more likely to promote support for social policies and intentions to partake in activism (a) when combined with retributive efficacy cues or (b) for pro-attitudinal but not counterattitudinal groups. This investigation examines these matters in a health communication context, directing anger toward soda companies for excessively marketing their sugary products to kids.

Anger and persuasion

Anger is a high-arousal, negative emotion accompanied by appraisals of other-responsibility (Frijda et al., 1989), goal interference (Averill, 1982), and unfairness or illegitimacy (Frijda et al., 1989). For these reasons, anger is responsive to perceptions that harm has been done by an actor other than oneself (Lazarus, 1991), that the actor was acting intentionally (Russell & Giner-Sorolla, 2011), and that the actor was aware of the harm done (Laurent et al., 2016). Functional theories of emotion suggest anger motivates the individual to remove the obstacle, typically by lashing out against or attacking the appraised wrongdoer in order to “rectify injustice” (Frijda et al., 1989, p. 1078; Lazarus, 1991).

Because of their potential to guide decision making and motivate behavior (Frijda et al., 1989), emotions have played a central role in scholarship on persuasive emotional appeals. Although threat-based appeals to fear have dominated this literature, anger appeals have increasingly captured the attention of persuasion researchers. Anger appeals should be particularly appropriate when the communicative goal is to promote high-commitment behaviors (e.g., political activism) aimed at righting wrongdoings in the world (Turner, 2011).

Two theories in particular offer useful predictions about the conditions under which an anger appeal is persuasive. The Cognitive-Functional Model (CFM; Nabi, 1999) states that communicators can craft persuasive messages to arouse discrete negative emotions (including anger). Furthermore, it predicts how different emotional experiences are associated with varying levels of motivation to attend to and process the remainder of the persuasive message and argues for the importance of including an efficacy component in any emotional appeal. In a similar vein, the Anger Activism Model (AAM; Turner, 2007) suggests it is possible for anger to lead to activism if the individual believes they have the ability to enact a recommended response and believes that performing the response will be effective at addressing the issue. Individuals who feel high levels of anger and hold strong efficacy beliefs should be most likely to engage in activism, whereas individuals who are neither angry nor efficacious should be least likely to engage in activism. As such, a successful appeal to anger is one that evokes anger and also conveys efficacy.

At this juncture, it is worth establishing the terminology used in the remainder of this investigation to signify the different parts of a persuasive anger appeal. To denote the part of the message that highlights the wrongdoing committed, which theoretically should incite anger, this paper uses the term offense component, echoing Lazarus’s (1991) language for the core relational theme of anger (“a demeaning offense against me and mine,” p. 222) (see also Nabi, 1999). To denote the part of the anger appeal that contains content about suitable ways to take action, this paper uses the term efficacy component, consistent with other theoretical frameworks of emotional appeals in persuasion (e.g., the Extended Parallel Process Model; Witte, 1992). This paper uses anger appeal to describe an offense-to-efficacy message in its entirety that is designed to persuade by directing anger toward a blameworthy actor.

As for empirical evidence, anger and persuasion was the focus of a recent meta-analysis (Walter et al., 2018), which revealed anger manipulations are more persuasive when they include efficacy content (compared to when efficacy content is absent). This finding supports the CFM’s and AAM’s assumptions that anger about an issue will be mostly likely to lead to persuasion when the individual perceives that the recommended response will be effective (see also Turner et al., 2020). A pivotal question that communication scholars have yet to empirically investigate is – effective at doing what exactly?

Tailoring response efficacy for anger

In the context of threat appeals, response efficacy refers to the extent to which one perceives a course of action will effectively protect oneself against a threat (Witte, 1992) and serves as a kind of outcome expectation – that is, perceptions about the consequences of taking some course of action (Bandura, 1997). This conceptualization of response efficacy for threat appeals is firmly rooted in functional theories of emotion because the “emotivational” goal (Roseman, 1984) of fear is to protect oneself against the appraised threat (Frijda et al., 1989; Lazarus, 1991), and threat appeals are indeed more effective when they include cues that the recommended action will be effective at reducing the threat (Bigsby & Albarracín, 2022).

However, protecting oneself is not the motivational goal associated with anger. Chiefly, anger motivates the individual to seek retribution for an appraised offense – a position supported not only by functional emotion research in psychology (Frijda et al., 1989; Lazarus, 1991) but also psychological work on retributive justice. This work has shown anger mediates the relationship between perceived seriousness of the offense and willingness to impose harsh punishment on the offender (Carlsmith et al., 2002; see also Carlsmith & Darley, 2008). Angry individuals should therefore be most likely to perform actions that they believe will punish the offender.

Connecting this logic to message design, Dillard and Nabi (2006) argued, “ … whereas fear appeals might require information suggesting an efficacious response to protect against threat, an anger appeal would be more effective if it suggested an efficacious response to retaliate against the offending agent” (p. S132). Theoretical justification can also be found in the CFM, which notes that people feeling anger intentionally evoked by the message will seek reassurance cues related to retribution (Nabi, 1999). Similarly, the emotions-as-frames perspective argues emotions motivate selective attention to and processing of information (e.g., persuasive media messages) in line with the action tendency of the emotion (Nabi, 2003). Whereas fear and sadness prime people to prefer information related to protection and assistance, respectively, the emotions-as-frames account predicts that anger primes preferences for information related to punishment (Kühne & Schemer, 2015; Nabi, 2003).

This paper uses the term retributive efficacy component to describe an efficacy component that conveys how a recommended action will effectively punish or hold accountable an offender and that should instill corresponding retributive efficacy beliefs. Retributive efficacy can be thought of as a subtype within the umbrella concept of response efficacy. If response efficacy is essentially an outcome expectation about the perceived consequences of an action, then response efficacy can be conceptualized with different end goals depending on the relevant emotion of interest: exerting punishment for a transgression (retributive efficacy, which should matter for angry decisionmakers) versus ensuring protection from a threat (what has traditionally been called response efficacy, which should matter for fearful decisionmakers). In an initial conceptualization of retributive efficacy, Skurka (2021) introduced a self-report measure of the construct and established the measure’s convergent and discriminant validity across three studies and across multiple health and environmental issues. Specifically, perceived retributive efficacy is correlated with (but distinct from) perceived issue efficacy (the perception that the action will address the larger social issue). Most importantly, retributive efficacy predicts political outcomes (support for public policies and intentions to partake in activism) above and beyond these other efficacy beliefs (Skurka, 2021), underscoring its value as a unique psychological construct worth intentionally targeting with persuasive messaging, though such an experimental endeavor has not been pursued until now.

The AAM identifies the importance of efficacy perceptions for those experiencing anger intentionally elicited by the message, but it does not specifically posit that response efficacy should be conceptualized with retribution as the functional goal. As an illustration, one recent experimental test of the AAM (Bessarabova et al., 2023) used a high efficacy message that stated, “Even small activities can help – simply sign a petition or engage in a sit-in!” (p. 5). In another experimental test (Turner et al., 2020), the high efficacy message read, “And there’s a lot we can do about it” (p. 19) by describing how the proposed actions would be “impactful.” By not identifying retribution-focused efficacy as the key efficacy belief that matters for angry decisionmakers, the AAM may be understating the potential for message-intended anger to translate into appropriate action.

This study uses a health communication context, examining the effects of anger appeals describing the exploitative practices employed by sugary drink corporations to market their products to children. Retributive efficacy cues in this context would explain how various solutions (e.g., restricting soda marketing to kids) could punish the wrongdoer (by cutting into beverage industry profits). Yet a strict reading of the AAM suggests that efficacy cues focusing on, for instance, whether the solution effectively addresses the larger social issue (reducing childhood obesity) would also enhance the effectiveness of an anger appeal because it provides a means to productively channel one’s anger. Thus, although the AAM would refer to this type of efficacy as response efficacy, this paper instead uses the label issue efficacy to avoid confusion with how response efficacy has been conceptualized in threat appeal theorizing (i.e., efficacy-as-protection) and tests the effects of both retribution efficacy cues as well as issue efficacy cues. To be clear, I define issue efficacy as a kind of outcome expectation specifically focused on perceptions (or message cues) about the extent to which a course of action will be effective at remedying the larger social issue (here, childhood obesity).

Based on these arguments, two initial preregistered hypotheses (“PH”) are offered, the first predicated on the AAM and the second predicated on the CFM and emotions-as-frames perspective:

PH1: A high offense component will lead to greater persuasive effects on policy support (PH1a) and activism intentions (PH1b) when it is followed by any efficacy content than when it is not followed by any efficacy content at all.

PH2: Though including any efficacy component should heighten the persuasive effects of a high offense component (per PH1), these effects on policy support (PH2a) and activism intentions (PH2b) should be greatest when the efficacy component conveys retributive efficacy.

The moderating effects of prior attitudes

The discussion above assumes all audiences respond to anger appeals similarly. Yet according to the AAM (Turner, 2007), this should not be the case. The AAM maintains anger appeals will only be persuasive for audiences already in favor of the advocated issue (pro-attitudinal audiences) because the appeal aligns with their views and elicits message-consistent anger, spurring them to act appropriately. Audiences initially against the issue (counterattitudinal audiences), by contrast, will not experience message-intended anger toward the culprit depicted in the message. They should instead experience anger toward the source of the message for trying to push their (counterattitudinal) views on the individual. Though not formally mentioned in Turner’s (2007) explication of the AAM, this defensive response experienced by counterattitudinal groups can be thought of as psychological reactance (Brehm & Brehm, 1981).

Reactance is a defensive motivational state that is a blend of negative cognitions and anger toward the message source, and it is initiated by the perception that one’s freedom is being threatened (Dillard & Shen, 2005). “In terms of the theory [of psychological reactance], perceived persuasive intent acts to increase magnitude of threat. When a person views another as intending to influence his or her behavior, the person is clearly aware of being pressured to comply” (Brehm & Brehm, 1981, p. 63). Accordingly, when an individual encounters a media message advocating a position that diverges from their own, they are likely to view the appeal as having strong persuasive intent, triggering a reactance response. Because reactance motivates the individual to restore the threatened freedom, the reactance state (of which anger toward the source is a key ingredient) should reduce the likelihood of persuasive success, potentially resulting in a boomerang effect whereby attitudes, intentions, or behaviors shift in the direction opposite what was intended by the communicator (Byrne & Hart, 2009).

In summary, because an attitudinally incongruent message is seen as a freedom-threatening attempt that challenges one’s preexisting views, the anger response that counterattitudinal individuals have should be a kind of message-unintended anger directed toward the messenger for being manipulative, à la reactance, that should result in a boomerang effect. If so, this means communicators should be extremely judicious in using anger appeals only to reach audiences who hold attitudes aligned with the message’s position. However, evidence for divergent effects of anger appeals as a function of prior attitudes is in short supply – so much so that Walter et al. (2018) were unable to test prior attitudes as a moderator in their meta-analysis. Nonetheless, based on the AAM, it is hypothesized (though not preregistered) that:

H1:

A high offense component will promote policy support (H1a) and activism intentions (H1b) among pro-attitudinal groups but decrease these outcomes among counterattitudinal groups.

Additionally, initial attitudes should moderate message effects on psychological reactance responses and message-intended anger. That is, the offense component of the message should be more successful at evoking message-intended anger (here, anger toward the beverage industry) among pro-attitudinal groups than counterattitudinal groups. Ness et al. (2017) found an anti-immigration website designed to induce anger evoked more anger among anti-immigration participants than participants in favor of immigration. Furthermore, because an anger appeal challenges their existing predispositions, counterattitudinal individuals should experience psychological reactance, including anger directed at the source (not message-intended anger at the offender in the message). To be specific, I expect that regardless of how much an offense message emphasizes a demeaning offense, counterattitudinal audiences should still be more reactant toward any message (relative to pro-attitudinal audiences) that does not align with their prior views. However, a high offense message should amplify this reactance gap between counter- and pro-attitudinal audiences because the high offense message is especially strong in making claims that oppose counterattitudinal audiences’ views. I anticipate an analogous but reversed pattern for message-intended anger: Pro-attitudinal audiences should experience greater message-intended anger than counterattitudinal audiences regardless of the offense level depicted in the message, but this gap should be exacerbated by a high offense message.

PH3: Counterattitudinal individuals will exhibit greater reactance outcomes (PH3a) and feel less message-intended anger (PH3b) than pro-attitudinal individuals.

PH4: Though these differences in reactance outcomes (PH4a) and message-intended anger (PH4b) should appear for both the high offense and low offense message, they should be wider in response to a high offense message.

The context of this study

Though a host of factors have contributed to the rise of childhood obesity, advocates have identified food and beverage companies as key players – especially because of their excessive marketing to kids (Nestle, 2015). Exposure to advertising for calorie-dense, nutrient-poor products correlates with adolescents’ consumption of obesogenic products (Olafsdottir et al., 2013). Not to mention, the soda industry is aware of the implications of their actions (Nestle, 2015). These facts are precisely the kind of content likely to stimulate appraisals of blame that underlie anger. This topic is also ideologically divisive in that research has documented polarized responses to messages describing “upstream” social determinants that impact health (Byrne & Hart, 2009).

Methods

Recruitment and sample

In January 2023, a sample of US adults was recruited through CloudResearch. Participants were eligible to participate if they had a 95% approval rating or higher on prior tasks, had completed 100 or more tasks, and if they were CloudResearch-Approved, meaning they had passed a previous screening process. The raw dataset included N = 1,303 responses. After data exclusions based on preregistered criteria, N = 1,244 complete cases remained for analysis. The mean age was 42 years (SD = 12.4). With respect to gender identity, 51% were men, 48% were women, and 1% were non-binary or another identity. Most participants were non-Hispanic (94%) and white (84%). On a political ideology scale of extremely liberal (1) to moderate, middle of the road (4) to extremely conservative (7), the typical participant leaned slightly liberal (M = 3.57, SD = 1.81). These demographics did not differ significantly across conditions (see Table S1 in Appendix). Preregistered power calculations, hypotheses, and analysis plans are available online. Data and syntax are available through the author’s Open Science Framework page (Skurka, 2023), and this study was exempted by the Penn State University Institutional Review Board.

Procedure and stimuli

After consenting, participants reported their attitudes about the target issue in this study (regulation of soda corporations). To reduce demand characteristics, participants also reported their attitudes toward two other policy solutions that have been proposed to address various social issues, randomly selected from a larger pool of topics (e.g., immigration, gun violence). Participants were then randomly assigned to read one of nine different messages, following a 3 (offense component: high, low, none) $\times$ 3 (efficacy component: retributive, non-retributive [issue efficacy], none) between-subjects design. After reading their assigned message, participants self-reported their emotional responses, reactance, support for the policy solutions described in the efficacy messages, efficacy perceptions for each of the solutions, and activism intentions. Participants received US$1.25 for their time.

Each message was formatted to resemble an online opinion article, including a header and a stock image of soda, to enhance ecological validity (see Supplemental Appendix for all messages). The offense messages began by linking childhood obesity to human actions (consumption of sugary drinks), then pivoted to highlight the role of industry actions – namely, soda companies targeting kids with exploitative marketing for their unhealthy products. The high offense versions went on to emphasize the harmful effects that the industry’s actions have had, describing how the industry is aware of the harm inflicted by their actions and that the industry deliberately engaged in these activities. To hold the information constant in substance and length, the low offense version downplayed the industry’s intentionality and awareness, casting doubt on whether the industry was aware or acting intentionally. These differences were also reflected in the message headers (“The Sugary Drink Industry Is to Blame for Childhood Obesity” vs. “The Role of Sugary Drinks in Childhood Obesity”). Given that perceptions of harm committed by another actor, actor intentionality, and actor awareness are all linked to anger intensity (Laurent et al., 2016; Russell & Giner-Sorolla, 2011), these claims in the high offense version were strategically emphasized to provoke anger toward the industry. This follows from recommendations by the CFM to provide cues in the message targeting the underlying appraisals for the emotion of interest (Nabi, 1999).

The efficacy messages recommended three solutions to address the issue (two public policies and personal actions). The efficacy messages summarized each proposed solution and maintained that each solution would be effective at helping address childhood obesity (i.e., issue efficacy, e.g., “Reducing sugary drink advertising to kids is one promising solution that will help reduce kids’ demands for these beverages … ”). Both efficacy messages provided relevant evidence for the effectiveness of the proposed solutions at reducing childhood obesity, but crucially, the retributive version also emphasized that each solution would be effective at holding companies accountable for their actions (e.g., “ … reduce kids’ demands for these beverages and make companies pay for their actions”). Thus, the issue efficacy version just focused on the efficacy of the solutions at addressing childhood obesity, not that the solutions would punish corporations. These different foci were captured in the headers for the two efficacy messages (“Solutions to Stop Childhood Obesity and Hold Corporations Accountable” vs. “Solutions to Stop Childhood Obesity and Reduce Weight Gain”). Both the efficacy and offense messages underwent pilot testing to confirm they were able to effectively influence retributive efficacy perceptions and anger intensity, respectively (see Appendix for pilot study details). Participants in the no offense/no efficacy condition read a message about a migratory bird called the shorebird to ensure they would not have a substantially different survey-taking experience from participants in the other conditions.

Measures

Initial attitudes

Using semantic differential scales of negative-positive, bad-good, undesirable-desirable, unnecessary-necessary (scored from 1 to 7), participants reported their attitudes toward increased regulation of soda companies to address childhood obesity prior to message exposure (M = 4.26, SD = 2.00, Cronbach’s α = .98).

Message-intended anger

To capture message-intended anger, participants were asked how much they felt several emotions toward the soda industry (angry, outraged, infuriated, disgusted) on a scale of not at all (1) to a great deal (7) (M = 3.14, SD = 1.84, α = .96). Several other emotion items were included to minimize demand effects.

Efficacy perceptions

Participants were presented with the three solutions mentioned in the efficacy messages, and for each solution, they reported their retributive and issue efficacy perceptions, shown in random order. On a scale of strongly disagree (1) to strongly agree (7), participants indicated their agreement with three statements measuring retributive efficacy borrowed from Skurka’s (2021) measure: Implementing this policy would hold soda companies accountable, …would punish soda companies, and …would negatively impact the profits of soda companies. The nine retributive efficacy items (3 solutions $\times$ 3 items) were averaged into a scale (M = 4.72, SD = 1.07, α = .85).

Issue efficacy was measured as agreement (on the same 7-point scale) with three statements about the extent to which the solution would be effective at preventing the issue of childhood obesity (Shen & Dillard, 2014) (e.g., Implementing this solution would be a sure-fire way to address childhood obesity). These nine items (3 solutions $\times$ 3 items) were averaged into an issue efficacy scale (M = 4.02, SD = 1.36, α = .93).

The retributive and issue efficacy scales were strongly but not perfectly correlated (r=.58), suggesting they are conceptually related but tapping into different constructs. Furthermore, confirmatory factor analyses indicated a two-factor model, in which the efficacy scales load onto their respective latent constructs, was a better fit for the data than a single-factor model, as indicated by a chi-square difference test, $\mathrm{\Delta }$X² (df = 1) = 1256.9, p < .001, and lower AIC and BIC values for the two-factor model (AIC = 74914, BIC = 75104) compared to the single-factor model (AIC = 76169, BIC = 76354).

Dependent variables

On a scale of strongly oppose (1) to strongly support (7), participants indicated their support for the two policies mentioned in the efficacy messages (Barry et al., 2009): prohibit all sugary-drink advertising on media that children are exposed to and prevent soda companies from deducting marketing and advertising expenses from their federal income taxes (M = 4.87, SD = 1.69, r = .67). On a scale of extremely unlikely (1) to extremely likely (7), the survey asked participants how likely they are to engage in several activism behaviors in the next month to take action on childhood obesity (adapted from Turner et al., 2020) (e.g., sign a petition, contact your local officials) (M = 2.74, SD = 1.52, α = .91). Policy support and intentions were moderately but positively correlated (r = .38, p < .001).

Reactance

This study assessed three components of reactance: perceived freedom threat, negative cognitions, and anger toward the source. For freedom threat, participants indicated their agreement (strongly disagree = 1, strongly agree = 7) with four statements taken from Dillard and Shen (2005) (e.g., The message tried to manipulate me) (M = 3.04, SD = 1.64, α = .92). Three statements (strongly disagree = 1, strongly agree = 7) were used to capture negative cognitions (Reynolds-Tylus et al., 2021): The thoughts I had about this message were bad/unfavorable/negative (M = 2.99, SD = 1.67, α = .95). For anger at the source, participants were asked how much they felt three emotions toward the people who wrote the message: angry, annoyed, and irritated (Dillard & Shen, 2005), measured on a scale of not at all (1) to a great deal (7) (M = 1.96, SD = 1.46, α = .93).

Results

Manipulation checks

A one-way ANOVA for anger toward the industry, F(2,1244) = 10.40, p < .001, $\mathit{\eta }$² = 0.016, with Holm comparisons indicated anger was significantly greater in the high offense condition (M = 3.43, SD = 1.88) than the low offense (M = 3.14, SD = 1.78, p=.042, Cohen’s d = 0.16) and no offense conditions (M = 2.84, SD = 1.82, p < .001, d = 0.32). The low offense and no offense conditions also differed from one another on anger (p = .042, d = 0.16).¹ I do note that these effect sizes – particularly between the high and low offense conditions – are small in magnitude, so I primarily focus on comparing the high offense and no offense condition for my hypothesis tests.

Regarding the effects of the efficacy manipulation on perceived retributive efficacy, F(2,1244) = 6.44, p=.002, $\mathit{\eta }$² = 0.01, the retributive efficacy message produced marginally greater retributive efficacy perceptions (M = 4.87, SD = 1.06) than the issue efficacy message (M = 4.71, SD = 1.02, p = .054, d = 0.16) and significantly greater retributive efficacy perceptions than no efficacy message (M = 4.61, SD = 1.11, p = .001, d = 0.25), which did not differ from one another (p=.16, d = 0.10). There was also a main effect of efficacy condition on issue efficacy beliefs F(2,1244) = 6.11, p = .002, $\mathit{\eta }$² = 0.01, with both the retributive efficacy message (M = 4.17, SD=1.35) and issue efficacy message (M = 4.06, SD=1.31) increasing issue efficacy beliefs compared to no efficacy message (M = 3.85, SD = 1.41, p = .002, d = 0.24, and p = .041, d = 0.24, respectively). Critically, issue efficacy beliefs did not differ between the retributive and issue efficacy messages (p = .24, d = 0.08). It is also noteworthy that the retributive efficacy message produced stronger retributive efficacy beliefs than it produced issue efficacy beliefs (M = 4.87 vs. M = 4.17). Because the goal of the retributive vs. issue efficacy manipulation was to impact beliefs about whether the solutions would punish the wrongdoer and not beliefs about whether the solutions would address the larger issue, these findings indicate a successful retributive efficacy induction. However, considering the marginal difference on perceived retributive efficacy between the retributive efficacy and issue efficacy conditions, I focus on comparing the retributive efficacy condition and no efficacy condition as the key comparison of interest for testing my hypotheses.

There was also an unexpected effect of the offense manipulation on perceived retributive efficacy, F(2, 1241) = 3.25, p = .039, $\mathit{\eta }$² = .01. The high offense message yielded greater retributive efficacy beliefs (M = 4.83, SD = 1.02) than not being exposed to an offense message (M = 4.64, SD = 1.09), p = .036, d = 0.18. The low offense message produced retributive efficacy beliefs between these two conditions (M = 4.71, SD = 1.10) but did not differ from the high offense condition, p = .21, d = 0.11, or the no offense condition, p = .36, d = 0.06. I return to this unexpected but potentially theoretically insightful finding in the discussion section.

Efficacy cues moderating persuasion effects (PH1, PH2)

The first two hypotheses predicted interaction effects between the offense component and efficacy component on policy support (PH1a, PH2a) and intentions (PH1b, PH2b). These interactions were tested with two-way ANOVAs.

Regarding policy support, there was no main effect of offense condition, F(2,1235) = 1.35, p = .26, $\mathit{\eta }$² = .002, but there was a main effect of efficacy condition, F(2,1235) = 5.86, p = .003, $\mathit{\eta }$² = .01. Comparisons with Holm corrections indicated that the retributive efficacy message (M = 5.02, SD = 1.69) increased policy support compared to no efficacy message (M = 4.66, SD = 1.72), p = .004, d = 0.23, as did the issue efficacy message (M = 4.94, SD = 1.70), p = .020, d = 0.18. Policy support did not differ between the two efficacy conditions, p = .48, d = 0.05. The two-way interaction on policy support was not significant, F(4,1235) = 1.51, p = .20, $\mathit{\eta }$² = .005.

Regarding intentions, there were no main effects of the offense manipulation, F(2,1235) = .27, p = .76, $\mathit{\eta }$² = .0001, or efficacy manipulation, F(2,1235) = .16, p = .85, $\mathit{\eta }$² = .0001, and the interaction between the manipulations was not significant, F(4,1235) = 1.25, p = .29, $\mathit{\eta }$² = .004. These results do not support PH1 or PH2.

Initial attitudes moderating effects on persuasion (H1)

H1 proposed initial attitudes would moderate the effects of the offense component on policy support (H1a) and intentions (H1b). This question was addressed with two linear regression models, which included the following predictors: initial attitudes, the condition variables (dummy coded with the no offense condition and no efficacy condition as the reference levels), and condition × attitude interaction terms.

Initial attitudes did not moderate the effects of the offense or efficacy manipulations on policy support (ps = .44–97). Turning to intentions, attitudes did not moderate the effects of efficacy condition (ps = .64, .75) but did moderate the effects of the offense manipulation (b_{low offense} = -.11, p = .029, b_{high offense} = -.10, p = .042). These interactions were probed using PROCESS (model 1) and are visualized on the left side of Figure 1. Among counterattitudinal participants (M_attitude–1SD = 2.26), the high offense message led to greater intentions than no offense message (b = .28, p = .041), but the low offense message did not affect intentions relative to no offense message (b = .21, p = .14). Among those at the sample mean (M_attitude = 4.26) and among pro-attitudinal participants (M_attitude+1SD = 6.26), no effects of the offense manipulations emerged. The Johnson-Neyman analysis indicated the region of significance for the effect of the high offense message (vs. no offense) was attitude $\le$2.58, and 27% of the sample fell in this region. In sum, the high offense message increased intentions but only for counterattitudinal participants.

Figure 1. Interactions between the offense manipulation and initial attitudes.

Bands represent 95% confidence intervals.

Initial attitudes moderating effects on message processing (PH3, PH4)

With PH3, it was hypothesized that reactance outcomes would be higher (PH3a) and message-intended anger would be lower (PH3b) among counterattitudinal compared to pro-attitudinal participants. PH4 predicted these differences in reactance (PH4a) and message-intended anger (PH4b) would be larger for high offense messaging (vs. low offense). A set of hierarchical regressions was run. The high offense and retributive efficacy conditions were set as the reference levels to permit comparisons between the high and low offense conditions and comparisons between the retributive and issue efficacy conditions.

Block 1 of these models, in which initial attitudes was the only predictor, allowed me to test PH3. In support of PH3a and PH3b, less favorable initial attitudes were associated with greater freedom threat perceptions (b = −.26, p < .001, R² = .10), more negative cognitions (b = −.27, p < .001, R² = .11), more anger toward the source (b = −.14, p < .001, R² = .05), and less anger toward the industry (b = .44, p < .001, R² = .25).

Block 2 of these models, which included condition × attitude interaction terms, provided tests of PH4. Initial attitudes did not moderate the effect of the offense manipulation on freedom threat (p = .57), negative cognitions (p = .39), or message-intended anger (i.e., anger toward the industry, p = .83). Initial attitudes did moderate the effects of the offense manipulations on anger toward the source (b_{low v. high offense} = .10, p = .053, b_{no vs. high offense} = .14, p = .005, $\mathrm{\Delta }$R² = .011), which is visualized on the right side of Figure 1. Among counterattitudinal participants (M_attitude–1SD), the low offense message led to less anger toward the source than the high offense message (b = −.33, p = .020), as did no offense message compared to the high offense message (b = −.56, p < .001). Among participants holding average initial attitudes, no offense message led to less anger toward the source than the high offense message (b = −.28, p = .005), but there was no difference between the low and high offense messages (b = −.13, p = .18). No differences in anger toward the source were observed across the offense conditions for pro-attitudinal participants (M_attitude+1SD). The region of significance for the effect of the low offense message (vs. high offense) was attitude $\le$3.49, and 35% of the sample fell within this region. The region of significance for the effect of no offense message (vs. high offense) was attitude $\le$4.79, and 54% of the sample fell within this region. To summarize, the high offense message prompted anger toward the messenger but only for counterattitudinal individuals. These results partially support PH4a but do not support PH4b.

Post hoc: Interactions between self-reported anger toward the industry and perceived efficacy

Up to this point, the analyses have focused on the effects of the experimental manipulations. Although the anger and retributive efficacy inductions were significant, it could reasonably be argued that these manipulations were not robust enough to permit a strong test of the argument that message-intended anger should lead to persuasive success at higher levels of (retributive or issue) efficacy. To this end, a series of post hoc regression analyses were performed to test for interactions between self-reported anger toward the industry and self-reported efficacy beliefs on policy support and intentions.

Because the efficacy scales combined beliefs about policies and personal actions, four new efficacy scales were created for these post hoc analyses: a 6-item scale of retributive efficacy specific to the policies (M = 4.97, SD = 1.14, α = .83), a 3-item retributive efficacy scale specific to personal actions (M = 4.23, SD = 1.48, α = .88), a 6-item issue efficacy scale specific to the policies (M = 3.93, SD = 1.51, α = .94), and a 3-item issue efficacy scale specific to personal actions (M = 4.21, SD = 1.62, α = .94). The policy support model included the following predictors: anger toward the industry, the policy-specific retributive efficacy scale, the policy-specific issue efficacy scale, the experimental condition dummy variables, initial attitudes (given its notable correlations with policy support, r = .61, and intentions, r = .39), and interaction terms between self-reported anger and the two policy-specific efficacy scales. A similar regression was run with intentions as the outcome variable but using the efficacy scales that were specific to personal actions.

In the policy support model ($\mathrm{\Delta }$R² = .02), anger interacted with retributive efficacy (b = .05, p = .008) and issue efficacy (b = −.10, p < .001). These interactions are plotted in the top two panels of Figure 2. Anger positively predicted policy support at all levels of retributive efficacy but was especially predictive at higher levels of retributive efficacy (b = .65, p < .001 at M_{retributive efficacy}+1SD = 6.11 vs. b = .55, p < .001 at M_{retributive efficacy}–1SD = 3.83), which is consistent with the logic underlying PH2. A different pattern emerged when looking at issue efficacy as a moderator. Anger positively predicted policy support when issue efficacy was low (p $\le$ .05 when issue efficacy $\le$ 2.10, 16% of the sample) but negatively predicted policy support when issue efficacy was high (p $\le$ .05 when issue efficacy $\ge$ 5.53, 88% of the sample). In sum, retributive efficacy amplified anger’s relationship with policy support, whereas issue efficacy compensated for low levels of anger.

Figure 2. Interactions between self-reported anger toward the industry and efficacy beliefs.

Bands represent 95% confidence intervals. In the middle figure, anger becomes negatively associated with policy support at especially high levels of issue efficacy ($\ge$5.53), but this negative relationship is not apparent here because, for the sake of consistency, the figures all display efficacy levels based on standard deviations from the mean.

In the intentions model ($\mathrm{\Delta }$R² = .01), retributive efficacy did not interact with anger (b = −.02, p = .26), but issue efficacy did (b = .06, p < .001). Consistent with the logic underlying PH1, as shown in the bottom panel of Figure 2, anger positively predicted intentions at all levels of issue efficacy, but this relationship was especially strong when issue efficacy was high (b = .44, p < .001 at M_{issue efficacy}+1SD = 5.83 vs. b = .24, p < .001 at M_{issue efficacy}–1SD = 2.59).

Discussion

Persuasion studies have largely neglected the questions of whether anger appeals might benefit from tailoring efficacy to match anger’s motivational goal of retribution (Dillard & Nabi, 2006) or whether prior attitudes moderate anger appeals’ impact (Walter et al., 2018). This investigation marks an important empirical step forward on both fronts, clarifying the boundaries of when anger appeals are most likely to change minds and intentions. Most notably in the findings, offense messaging about a wrongdoer’s actions did not have uniform effects. As the Anger Activism Model (AAM) would predict (Turner, 2007), counterattitudinal individuals felt angrier toward the messenger when presented with a high offense message, but pro-attitudinal individuals did not exhibit such a defensive response. It should be noted that perceptions of freedom threat were unaffected by the messages, so the high offense message did not trigger reactance, per se, but simply annoyed counterattitudinal groups.

Intriguingly, in contrast to the AAM’s predictions, an offense appeal was only persuasive among counterattitudinal groups, as they expressed greater intentions to engage in activism-related behaviors upon exposure to a high offense message (relative to not receiving this messaging). No persuasive effects of the offense appeals emerged for groups that were neutral or favorable toward the advocated issue. From the standpoint of the AAM, reactance theory (Brehm & Brehm, 1981; Dillard & Shen, 2005), and even motivated reasoning principles more broadly (Taber & Lodge, 2006), these findings are puzzling. One possible explanation involves the nature of attitudes. The AAM focuses on attitude valence (whether one’s evaluation of the attitude object is positive or negative), but attitude valence may matter less to the processing of offense appeals than attitude certainty (one’s conviction of one’s evaluation of the attitude object; Tormala & Rucker, 2007). If so, people in this study reporting negative initial attitudes may not have been especially planted in their views, rendering them more open to the ideas presented.

A second explanation – though somewhat theoretically provocative – is that it is possible for an anger appeal to persuade the opposition. In a meta-analysis of political messaging experiments, Coppock (2023) found no consistent evidence that persuasive political appeals produce boomerang effects among counterattitudinal individuals – or even that political messages consistently produce different effects for pro- and counterattitudinal individuals. Although such effects can and have occurred (see Byrne & Hart, 2009), Coppock’s findings suggest it is more common for counterattitudinal groups to update their views in line with the message’s position. Of course, only once additional findings have accrued can firm theoretical recommendations be made, but these ideas would have tremendous implications for emotional appeal theories like the AAM (and persuasion theories more broadly, such as psychological reactance theory). Namely, these ideas would suggest (a) we should not assume defensive reactions (e.g., anger at the source) always translate to reduced persuasion and (b) boomerang effects are the exception rather than the norm – at least, when the message presents compelling arguments and when the messaging avoids explicitly partisan cues (for similar conclusions, see Guess & Coppock, 2020).

Offense messaging aside, efficacy cues mattered across the board for participants in this study. Individuals exposed to any type of efficacy information (focused on the effectiveness of the recommended responses at addressing the larger issue or focused on retribution toward the wrongdoer) were more supportive of the advocated policies than individuals not exposed to any efficacy information. This effect is consistent with meta-analytic findings that communication about a policy’s effectiveness promotes support for the policy (Reynolds et al., 2020). These findings are relevant for persuasion theorists in that they underscore the value of designing persuasive messages with clear efficacy cues about the desirable consequences of enacting the proposed solutions. They are also relevant for practitioners, as they suggest the need to provide sound evidence of social policies’ efficacy to marshal public support for social issues in civic life.

With respect to the role of psychological processes (i.e., message-intended anger and efficacy beliefs) in predicting persuasion, post hoc analyses revealed a nuanced set of interaction patterns. First, message-intended anger was associated with greater policy support at all levels of perceived retributive efficacy, but perceived retributive efficacy strengthened this relationship. In other words, individuals were most likely to express policy support when they felt angry at the wrongdoer and also believed those policies would hold the wrongdoer accountable. Applying Holbert and Park’s (2019) typology, this moderation reflects a divergent positive contributory pattern. Of course, causal claims cannot confidently be made given the cross-sectional nature of these analyses, but this interaction between subjective anger and perceived retributive efficacy is consistent with the argument drawn from functional theories of emotion that anger motivates preferences for actions and information that serve a retaliatory purpose (Frijda et al., 1989; Lazarus, 1991; Nabi, 2003). It also coheres nicely with research on the psychology of punishment, which has shown anger plays a critical role in explaining why people seek to punish transgressors for committing blameworthy offenses (Carlsmith & Darley, 2008; Carlsmith et al., 2002). Perhaps most importantly, this finding – though cross-sectional – speaks to the need for emotional appeal theorists to conceptualize and operationalize efficacy constructs in a way that corresponds to the motivational goal of the target emotion (Dillard & Nabi, 2006). Failing to do so means persuasion theorists may very well underestimate the potential for anger intentionally aroused by the message (or perhaps other message-intended emotions) to facilitate persuasion. Future research could expand on these findings in the context of appeals to other emotions. For example, guilt’s effects might be enhanced by perceptions that a recommended response will make amends for a personal transgression (Dillard & Nabi, 2006).

Second, issue efficacy beliefs interacted with message-intended anger in different ways depending on the outcome in question. Mirroring the divergent positive contributory pattern that emerged between message-intended anger and perceived retributive efficacy, perceived issue efficacy strengthened the relationship between message-intended anger and intentions – a pattern predicted by the AAM, which maintains anger is most likely to translate into appropriate action when it co-occurs with the belief that taking action will effectively address the problem. As such, this is one of few investigations to provide evidence for the AAM’s proposed interaction between anger and efficacy, which other studies have failed to find evidence for (Bessarabova et al., 2023; Skurka, 2018; Turner et al., 2020). However, in direct contradiction to the AAM, individuals experiencing high levels of message-intended anger were generally supportive of policies almost regardless of whether they believed implementing the policy would address the issue, whereas less angry individuals were more supportive of policies if they believed the policies would address the issue. Put differently, issue efficacy compensated for low levels of message-consistent anger. In fact, at the highest levels of issue efficacy, message-intended anger negatively predicted policy support – a pattern Holbert and Park (2019) refer to as convergent cleaved moderation.

What, then, are the theoretical implications of these findings? The AAM posits efficacy cues should follow anger-inducing content but does not explicitly state that the efficacy claims should be about retribution, per se. The current findings suggest there may be a benefit to explicitly incorporating the notion of retributive efficacy into the AAM, as implied by the CFM (Nabi, 1999) and emotions-as-frames perspective (Nabi, 2003). When promoting policy support is the rhetorical goal of an anger appeal, instilling beliefs about the effectiveness of retribution may confer a benefit (i.e., amplifying the link between message-intended anger and policy support), but instilling issue efficacy beliefs may not be particularly important, considering that individuals angered by the message were highly supportive of policy initiatives regardless of their issue efficacy beliefs. When the goal is encouraging personal behaviors, the AAM’s predictions may hold more water, in that anger appeals would benefit from efficacy messaging that strengthens beliefs about how taking action will address the larger social issue in question.

Finally, not only did the high offense message heighten anger toward the industry; curiously, in manipulation check analyses, it also heightened beliefs that taking action will hold the offender accountable, even though the offense messaging made no reference to potential solutions or their effectiveness. Perhaps this finding is capturing a kind of motivated cognition phenomenon in which anger induced by the offense messaging makes one want to believe that taking action will effectively punish the transgressor (Hughes & Zaki, 2015). Similar evidence has been offered for the relationship between threat and efficacy in which threat appeals increase perceptions of efficacy (Hornsey et al., 2015) – a phenomenon Hornsey and colleagues referred to as motivated control. In this way, just as anger drives motivated attention to ideas related to retribution because of anger’s approach orientation and appraisal of blame, as predicted by the emotions-as-frames account (Nabi, 2003), so too might anger drive motivated acceptance of the belief that a proposed solution will teach the wrongdoer a lesson. If such an effect is replicated in future work, this could have interesting implications for the design of anger appeals: It may not be necessary to provide retributive efficacy cues in the message because the anger evoked by the first part of the message might organically lead message recipients to believe the recommended responses will effectively punish the wrongdoer.

Limitations

There are several limitations of this study to keep in mind. Chiefly, the anger and efficacy inductions were small-to-moderate in size, which likely undermined the ability to detect theoretically derived moderation effects. Worth noting, however, is that with respect to the anger manipulation, my anger measure consisted of high arousal emotion terms (e.g., infuriated, outraged), which may not be easily felt with exposure to a single message. Also worth noting is that other recent experiments examining anger appeal effects have yielded anger and efficacy induction effects very similar in magnitude as what I see here (Bessarabova et al., 2023; Turner et al., 2020). These consistent patterns highlight the logistical challenges persuasion scholars face when attempting to induce absolutely high and low levels of target emotion states (and not just relatively “higher” and “lower” levels of those states). Second, because the efficacy messages developed for this study primarily advocated for public policies rather than individual-level behaviors, this study did not examine the role of self-efficacy. Third, this study involved brief exposure to persuasive messages about a single context with immediate measurement of outcomes. Finally, the convenience sampling approach used here may limit the generalizability of the findings.

Conclusion

The findings from this experiment point to several promising directions to build upon dominant theories of anger and persuasion – the AAM in particular. First, there may be conditions under which an anger appeal not merely avoids a boomerang effect but in fact persuades the opposition. This notion ought to be investigated in future work by considering the certainty with which audiences hold their preexisting attitudes as well as exploring the provocative idea that counterattitudinal groups can update their position in line with the message even if they engage in defensive message processing (being annoyed at the messenger). Second, the AAM should more explicitly consider retribution-specific efficacy as a crucial moderator of anger’s influence – particularly when the aim is to foster support for systemic, policy-level solutions.

Open scholarship

This article has earned the Center for Open Science badges for Open Data, Open Materials and Preregistered. The data and materials are openly accessible at https://osf.io/z3hme/?view_only=15334bb528514f099716550e3f092bf0.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Supplementary material

Supplemental data for this article can be accessed online at https://doi.org/10.1080/15213269.2024.2352747

data availability

The data will be made available upon request by contacting the corresponding author

Additional information

Funding

This material is based upon work supported by the National Science Foundation under Grant No. 1824212.

Notes

1. The offense manipulation did not influence how anxious, afraid, guilty, hopeful, or compassionate participants felt. Exposure to either of the offense messages increased sadness compared to no offense ($\mathit{\eta }$² = .013), though the sadness means in the high and low offense conditions (M = 2.97 and M = 2.79, respectively) were noticeably lower than the anger means. There was also an effect of the efficacy manipulation on compassion, $\mathit{\eta }$² = .01, such that the retributive efficacy message (M = 1.92) and the issue efficacy message (M = 1.96) both decreased compassion relative to the no efficacy condition (M = 2.20).