From the standpoint of causal inference, procedures like the Life Event Inventory Task are attractive in that specific emotions clearly occur prior to message processing. One can then presumably draw uncluttered inferences concerning how emotions influence message processing. This is an interesting question, but it is different from that of how message-induced emotions might bring about persuasion. And, of course, many communication researchers are less interested in supposedly pure manipulations of emotion and more concerned with how the affect that is induced by one message might shape processing of a subsequent persuasion effect. For example, Chang (2011) examined how affect induced by a magazine editorial influenced reactions to an advertisement that was presented after the editorial. Because of its correspondence with real-world media contexts, the use of messages to evoke feelings in message-irrelevant affect designs enhances external validity. However, they also present challenges to internal validity as a result of the complexity of the affect-inducing stimulus.
The Effects of Affect
Evidence of Effects
For at least two millennia, students of suasory discourse have embraced the idea that emotions can persuade (Aristotle, 2007). Evidence of this point can be seen in the analysis of emotional appeals by early rhetoricians, including Aristotle, and can be traced through history to the present time. But, the more nuanced question—how should the emotion-persuasion link be analyzed and understood?—is still under debate. There are several options to consider. From the perspective of the dimensional valence model, the affects vary along a single dimension (positive vs. negative) and should be tested accordingly. That is, researchers should measure affective responses on a series of semantic differential scales, then correlate the sum of those scales with the persuasion outcome of interest. The categorical valence model directs researchers to similar options, except that positive and negative affects are measured independently of one another and treated as two separate predictors. The pleasure-arousal model, also a dimensional model, ordains the use of two sets of bipolar scales and, thus, two continuous predictors: Pleasant (vs. unpleasant) and arousal (vs. subdued). The discrete emotions position insists that each emotion is defined by a different function and different response patterns. Thus, emotions must be measured individually, and their effects on persuasion must be evaluated using methods that allow for unique effects to manifest themselves.
To date, there are many existing studies that test the relationship between affect and persuasion. Only a subset has produced evidence that can be used to differentiate the various models of affect. Studies that assess emotions on only bipolar scales must be excluded because the measurement procedures assume, rather than test, the structure of affect. If affects are measured on a positive versus negative scale, researchers cannot then unscramble the differences between sadness and anger or hope and happiness because those distinctions are lost at the moment of data collection. The same problem extends to pleasure-arousal models, except that there are two dimensions rather than one. In fact, the only data that can be used to empirically evaluate the worth of the different models come from investigations that utilize discrete emotion procedures. It is possible to aggregate specific emotions into larger categories, such as pleasant and unpleasant, but impossible work in the other direction, that is, to disassemble global affective judgments into discrete elements.
There are four types of evidence in the persuasion literature that distinguish the discrete emotions model from the alternatives. Each is illustrated next using investigations from different contexts:
1. There is variation in the effects of negative emotions. Although public opinion remains divided about the wisdom of a Western-led war with Iraq, there is no question that a sizeable portion of the U.S. population is critical of their government’s action. In a survey of undergraduate U.S. citizens, Iyer, Schmader, and Lickel (2007) examined the associations of anger, guilt, and shame with intentions for political action. They found that both anger and shame predicted intentions to advocate withdrawal from Iraq, while the association with guilt was nonsignificant. The results are instructive in that they illustrate how two negative emotions can exert independent effects on a single intention, and, in the same analysis, a third negative emotion may show no significant effect at all. In further illustration of this same general point, Nan (2009, p. 437) presents results showing that anger toward public service announcements is counter-persuasive, while guilt seems to generate agreement with the message. These findings too are at odds with any model of affect that anticipates uniform effects within valence-defined categories, that is, the bipolar valence model, the categorical valence model, and the pleasure dimension of the PA model.
2. There is variation in the effects of positive emotions. Participants in the Kim, Park, and Schwarz (2010) were asked to list three life events that were either peaceful or exciting. Then, under the guise of a separate study, data were gathered on their evaluations of a message that promoted a vacation to Japan. There were two versions of the advertisement. One of them emphasized the possibilities for adventure and excitement via exposure to anime/manga, electronics, and Japanese drumming, while the other promised a serene and tranquil experience focused on hot springs, tea ceremonies, and religious traditions. Responses to the question “I would like to take a vacation in this country” were most favorable when message content matched the pre-existing emotional state. In other words, the effects of these two positive emotions were reversed as a function of message content. Nan (2009, Table 2) also provides evidence of the independent, and generally opposing, effects of two other positive emotions. In her data, happiness is associated with greater persuasion, and contentment with less.
3. Discrete emotion models exhibit better predictive power than the alternatives. In one of the few direct comparisons of affect models, Dillard and Peck (2001) examined emotional reactions to public service announcements on a variety of prosocial topics, such as exercise, charitable giving, and avoiding contraction of HIV. They conducted two series of regression analyses, each of which used perceived effectiveness of the messages as the criterion variable. In the discrete emotions analyses, anger, fear, sadness, guilt, happiness, and contentment were used as separate predictors. For the valence analyses, anger, fear, sadness, and guilt were combined to form a single, negative-affect predictor; happiness and contentment were combined to form a single, positive affect variable. A ratio of the R2 change for each set of equations permitted a judgment of the relative predictor power of each affect model. The discrete emotions analysis was superior to the valence analysis by a factor of two.
4. Different emotions produce different persuasion outcomes. Appraisal theory presupposes that emotions are means of solving functionally distinct types of problems. From that premise, it seems reasonable to infer that different emotions would produce different persuasive outcomes. An illustration of these variations is offered by Brader (2005), in an experiment conducted in the context of a Democratic primary race for governor. Two messages were designed, both of which manipulated various nonverbal cues to create either enthusiasm or fear. More concretely, one message paired uplifting music and images of children with message content that described conditions that were “good and getting better” (p. 392). The other appeal, which argued that things were “bad and getting worse,” (p. 392) was accompanied by tense music and pictures of violence. In a crossed design, both messages were attributed to both candidates. The data revealed that the enthusiasm message solidified pre-existing voter preferences and dramatically increased intention to vote (a 29% increase). In contrast, the fear message was more effective at changing voter preferences, but had no significant influence on intention to vote.
The research cited in preceding paragraphs is illustrative: Much more evidence exists. The research that appears in the literature prior to 2001 is summarized by Dillard and Meijnders (2002). And, accrual of data since that time has only supported the conclusion that a discrete emotions perspective is not merely preferable, b
ut rather, necessary for understanding the effects of emotion on persuasion (e.g., Huddy, Feldman, Taber, & Lahav, 2005; Lerner, Gonzalez, Small, & Fischhoff, 2003; Lerner & Keltner, 2000).
Mechanisms of Effect
The appraisal model of persuasion (Figure 10.1.) depicts movement from emotion to persuasion outcome with a simple, single arrow. However, because emotions are complex states that implicate a host of perceptual, cognitive, physiological, and motivational systems, they might influence the persuasion process through multiple means (Dillard & Nabi, 2006).
Perhaps the simplest claim regarding the mechanism is that emotions can have a direct effect on persuasion outcomes (Dillard & Nabi, 2006). This idea traces back to the assumption that stimulation of an action tendency is a defining feature of any emotional response. Each emotion has some implications for action that can be understood as approach/engagement or avoidance/withdrawal. These motivational tendencies are guides to behavior with direct implications for attitudes and intentions. However, because emotions are evolutionarily designed to solve specific problems, the action tendency associated with each emotion is a more specific variation on one, the other, or both of those two broad themes. Table 10.2. provides a summary of emotions and their predominant action tendencies. The qualifier predominant is important in this application because action tendencies vary as a function of context. For example, the predominant action tendency for fear is acquiescence. But, under some circumstances, fear motivates aggression. The table also attempts to elaborate on the point that action tendencies can be analyzed at both the general and specific levels. At the general level, anger and happiness are both approach emotions in that they prompt engagement with the emotion-inducing stimulus. However, at the specific level, they are quite different from one another. The logic of the direct effects mechanism predicts that action tendency effects will be moderated by various aspects of the advocacy or the context in which the message is presented.
Table 10.2 Emotions and Their Action Tendencies
a. These should be understood as approach/engage and avoidance/withdrawal.
Another class of mechanisms reflects heightened sensitivity to appraisal-relevant message features. The most commonly observed type of effect is matching, in which emotion-consistent beliefs are amplified by the evocation of the corresponding emotion. Just as perceived risk induces fear, fear increases estimates of risk. Conversely, prior instigation of anger attenuates perceived risk (Lerner et al., 2003). Kim et al. (2010) describe matching effects for pre-message emotions and subsequent issue judgments that are mediated by expectancies, a specific form of belief. Turner (2007) reports findings in which matching effects for anger carry over to politically oriented action tendencies, including signing a petition, talking to friends, and organizing an event.
There is also evidence of emotional inductions that can activate the behavioral approach and avoidance systems, which, in turn, render audience members sensitive to noncontent message features. Using the Life Event Inventory procedure, Yan, Dillard, and Shen (2010) demonstrate a variation on the sensitization phenomenon such that when sadness or fear are used to stimulate the behavioral inhibition system, loss framed messages become more persuasive. In parallel, inducing anger or happiness—both of which are approach emotions—activates the behavioral approach system and causes gain-framed messages to be evaluated more favorably (Yan, Dillard, Shen, in press).
Emotions can also influence the way in which messages are processed. Specifically, emotions of different sorts might propel individuals toward or away from a message topic, a phenomenon that Nabi’s (1999) cognitive-functional model (CFM) characterizes as motivated attention. In the same vein, message consumers may try to regulate the emotion via motivated processing (Nabi, 1999). Based on the type of emotion experienced, motivated attention sets a baseline attention level that will either impede (for avoidance emotions, like fear) or facilitate (for approach emotions, like anger) subsequent information processing. Expectation of reassurance from the message might then further shape style of processing. For example, given anger’s nature, the CFM predicts that it is likely to promote closer information processing. Fear is likely to promote more message scrutiny only when reassurance cues are unavailable. Huddy and Gunnthorsdottir (2000) suggest that message-induced emotion effects may be limited to message recipients who are knowledgeable about and involved with the issue.
Although the data on matching effects are appealing, Agrawal and Duhacheck (2010) suggest that they may be overridden, at least in some cases, by motivated processing. In their research, shame or guilt was evoked by describing others either observing or suffering from the negative consequences of binge drinking. The others-as-observers perspective produced shame because message recipients were presented as behaving in way that yielded unfavorable judgments of their friends and family. In contrast, when friends and family were described as bearing the costs of the research participants irresponsible behavior, the participants reported feeling guilt. Antidrinking messages that were matched on induced emotion, yielded mismatched persuasive effects presumably because participants were motivated to counterargue against appeals that further threatened their sense of self.
Other Themes in the Research Literature
Affect and Attitudes
The concept of attitude is inextricably bound up with that of evaluation. And, of course, so is affect. The fact that the two concepts both draw on valence as an underlying concept has meant that, at certain points in history of social science, researchers made little or no distinction between them. For example, the original version of the theory of reasoned action viewed affect and attitude as essentially synonymous (Fishbein & Ajzen, 1975). However, in subsequent years, theorists developed meaningful grounds on which to differentiate the two. One of the most compelling expositions is offered by Zanna and Rempel (1988) who show that the utility of treating (1) attitude as a summary cognitive judgment regarding the goodness-badness of a mental object and (2) affect as one or more feeling states that may influence attitude. Their position aligns perfectly with the appraisal model of persuasion given in Figure 10.1., and it has clear implications for the measurement of both attitude and affect. To wit, attitude is appropriately assessed via judgments of good or bad, favorable or unfavorable, and like or dislike, all of which capture the cognitive aspect of evaluations. In contrast, measures that index responses such as anxious, pleasant, tense, happy, and sad may be appropriate for the assessment of emotion, but they are invalid indicators of attitude. Some investigations continue to intermingle measures of affect and attitude, thereby diminishing the interpretability of their results.
The Affect Heuristic
Heuristics are short-cut decision-making rules that come into play when a more detailed analysis of persuasive circumstance is unwarranted or impossible. For instance, when faced with a nonfunctional computer, many people defer to the recommendation of their IT specialist rather than spend time themselves learning the technology of personal computers, disassembling their machine, then locating and fixing the problem themselves. Pertinent to this chapter is the claim that individuals often rely on an affect heuristic: A feeling that captures the positive or negative aspect of a stimulus (Slovic & Peters, 2006). This definition seems to carry on the tradition of conflating affect and attitude. However, the terminology and theorizing are more contemporary. Experiments conducted on the affect heuristic manipulate variables, such as time pressure, that are held to govern the activation of other cognitive heuristics (Slovic, Finucane, Peters, & MacGregor, 2002). Indeed, the results show parallel effects. However, the affect heuristic is defined in terms of feelings rather than cognition. This bears some similarity to the notion of signal value, that is, the idea that subjective experience is the means by which conscious awareness is informed of the change in the physiological, expressive, cognitive, and, motivational domains (described in more detail below). But, work on the affect heuristic seems to reject the conception of specific emotions. Despite its applicati
on to risk assessment, which is normally associated with fear, research on the affect heuristic makes only a distinction between positive and negative affect. In the final analysis, the notion of an affect heuristic has been successful at stimulating research. However, it seems to bring together concepts from appraisal theory, cognitive theory, and the valence approach in ways that are fundamentally incompatible with one another. Thus, it is difficult to know how to locate the research in a more general framework.
Issues for Future Research
* * *
The Regrettable Persistence of Valence Models
Taken together, the reasons to prefer a discrete emotions perspective over the alternative conceptions are quite strong (Nabi, 2010). It seems clear that continued dependence on valence models is a mistake. Nonetheless, the valence model still appears in the literature with remarkable regularity—a circumstance that seems to call for explanation.
One way of accounting for this curiosity is to recognize that valence judgments are salient throughout life. They are learned early, then practiced regularly. Rudimentary processes such as the startle response are quickly followed by an assessment of the intrinsic pleasantness of stimuli (Scherer, 1984). And, as theories such reasoned action tell us, even many initially complex decisions ultimately reduce to good versus bad judgments (chapter 8 in this volume). Jointly, these facts of life may bring valence judgments to the forefront of human consciousness, where they are as likely to bias the views of social scientists as they are lay persons.
The SAGE Handbook of Persuasion Page 32
