Evaluating the Extended Parallel Processing Model
Witte and Allen’s (2000) meta-analysis seems to be of two minds. The meta-analytic results comparing the four EPPM conditions (i.e., high threat-high efficacy, high threat-low efficacy, low threat-high efficacy, and low threat-low efficacy) indicate main effects for both threat and efficacy, but no interaction effect. Specifically, consistent with the EPPM, the high threat-high efficacy condition produced the greatest attitude change. Inconsistent with model predictions, the low threat-high efficacy and high threat-low efficacy conditions both generated greater attitude change than did the low threat-low efficacy condition. (The EPPM predicts that all three of these conditions should generate equal amounts of attitude change, though for different reasons.) In short, although the EPPM predicts a multiplicative interaction between threat and efficacy, these results suggested an additive process where threat and efficacy separately influence responses.
In an effort to save the EPPM’s multiplicative prediction, Witte and Allen (2000) performed effects-coded analyses pitting the predicted multiplicative model with the additive model suggested by the data. Using unreported contrasts to represent the two effects, Witte and Allen report that both the additive and multiplicative patterns significantly predicted outcomes, suggesting that the predicted multiplicative model might fit the data after all. The flaw in that logic, however, is that the two models’ predictions are strongly interrelated. Specifically, the multiplicative interaction contains the two main effects from the additive model. Therefore, any set of effects codes representing the additive model will correlate strongly with those representing the multiplicative model.1 In such a case, it would be nearly impossible to generate effect-code models that could differentiate these two sets of predictions.
In summary, the accumulated data appear similar to tests of the original formulation of the protection motivation explanation. When considering the separate impact of predictor variables, the data match predictions reasonably well. Both threat and efficacy positively influence attitude and behavior change. When considering the specific combinations of threat and efficacy, however, the accumulated data do not fit predictions as clearly. Specifically, the “off” cells (i.e., high threat-low efficacy and low threat-high efficacy) produce greater attitude change than the model predicts. Why this might be and how it could be explained from the EPPM perspective remains unclear.
The Stage Model of Fear Appeals
The most recent explanation for fear appeals is the stage model (e.g., de Hoog, Stroebe, & de Wit, 2007; Das et al., 2003; Stroebe, 2000); not to be confused with the transtheoretical (or stages of change) model (see Prochaska & DiClemente, 2005). As is typical for this literature, the stage model combines concepts from earlier explanations (e.g., parallel response, protection motivation, and the EPPM) but also combines them with the dual-process model of message processing (Chaiken, 1980), stress-coping explanations (e.g., Lazarus & Folkman, 1984), and models of message evaluation, evidence, and inference (e.g., Kunda, 1987). The model is quite recent, more complex than its predecessors, and relatively untested. Therefore, this review will be brief.
Like the other explanations, the stage model assumes that fear appeal components include threat (i.e., severity and susceptibility) and coping (i.e., response and self-efficacy). These components are presented, and processed, in stages (thus the explanation’s label) with threat assessments potentially influencing subsequent efficacy assessments (Das et al., 2003; Stroebe, 2000). The stage model considers threat components (i.e., severity and susceptibility) as differentially influencing depth of message processing (heuristic or systematic), receiver’s processing goals (i.e., accuracy or defensive), and persuasive outcomes (i.e., attitude or behavior change). For example, while both severity and susceptibility are predicted to influence attitudes, only susceptibility is predicted to influence behaviors and intentions. What is more, both the processing of the threat and efficacy message components are predicted to be biased (the threat component negatively and the efficacy component positively). When threat is strong, audience members will attempt to counterargue and find logical flaws (loopholes) in message arguments (Das et al., 2003; Stroebe, 2000). Failing to find such loopholes, receivers will likely accept any coping response unless it is totally implausible and/or impossible to enact. Such a positive bias allows the receiver to feel good in the face of a severe threat.
Evaluating the Stage Model
The stage model is intriguing; however, evaluating it is difficult, first, because few published studies directly test it and, second, its predictions are slippery. For example, de Hoog et al. (2007) assert several times that main effects either could or should be moderated. Thus, it is not clear whether main effects or interaction effects should influence responses. The body of studies in the de Hoog et al. meta-analysis included only those studies that manipulated either severity or susceptibility or manipulated both, but separately (i.e., studies that confounded severity and susceptibility, e.g., Janis & Feshbach, 1953 and all of Witte’s work, were not analyzed). Meta-analytic results (de Hoog et al.) indicate that several predictions were inconsistent with meta-analytic results. For example, susceptibility predicted behaviors, but not attitudes (as expected), however, severity significantly predicted attitudes (where only efficacy judgments were supposed to have the influence). Finally, although response efficacy predicted attitudes, but not behaviors, self efficacy predicted both attitudes and behaviors (while only the former was predicted).
The Current State of Fear Appeal Theory and Research
* * *
On the whole, the current state of fear appeal theory and research is simultaneously clear and confused. The results of several meta-analytic investigations are relatively clear. Fear appeals work, for most audiences and messages. As Witte and Allen (2000) noted, both the strength of fear appeals (including both severity and susceptibility) and the efficacy of the coping response (both self-efficacy and response efficacy) influence experienced fear, attitudes, intentions, and behaviors.
That fear appeals work for most audiences and messages, however, represents an important caveat to this conclusion. Witte and Allen (2000) report that, in most all cases, effect sizes vary more across studies than is expected by chance alone. Such heterogeneity in study effect sizes suggests that fear appeals are more effective for some topics and for some audiences than they are for others, due presumably to some unknown moderator variable (or variables). Consideration of such variables will be taken up shortly.
Why Don’t Fear Appeal Explanations Work?
Given the state of the literature, a summary evaluation of fear appeal explanations has not changed substantially since 1984 when Boster and Mongeau claimed “none of the fear appeal explanations are consistent with the available evidence” (p. 366). Fear appeals work (most of the time at least), but it is safe to say that fear appeal explanations do not. Specifically, the drive model never came close to explaining the data; the parallel processing model was untestable; and predictions from the most recent explanations (i.e., protection motivation, extended parallel processing, and stage models) were inconsistent with how various fear appeal components (i.e., severity, susceptibility, self-efficacy, and response-efficacy) influence attitude, intention, and behavior change (de Hoog et al., 2007; Rogers, 1983; Witte & Allen, 2000).
There are several potential reasons why fear appeal explanations don’t match the accumulated data. Some of these reasons center on the nature of the explanations themselves, while other reasons are a function of the research methods used to test them. I will consider each of these factors in turn.
Issues Relevant to the Explanations Themselves
There are at least three factors associated with fear appeal explanations (both what they have and what they lack) that likely lead to their demise. First, ever since Hovland and colleagues’ (1953) seminal work, each new fear appeal explanation has taken something from previous work, changed it and/or combined it with other perspecti
ves and/or concepts, in order to develop something new. Parallel processing was built with bricks from the drive model. The EPPM combined parts of the drive, parallel response, and protection motivation explanations. Most recently, the stage model combined elements from all these explanations with several other persuasion concepts (e.g., Chaiken, 1980). In short, instead of becoming more consistent, terse, and compelling over time, fear appeal explanations are becoming more bloated, convoluted, and no closer to explaining the accumulated data. Such a state of affairs is due, in part, to the perpetration of several central constructs that are poorly conceptualized and operationalized (e.g., fear control, danger control, and defensive avoidance). What is more, one has to question whether the accumulated data necessitate this level of conceptual complexity.
A second shortcoming of fear appeal explanations is that they consistently highlight cognition and shortchange physiological aspects of emotion. This cognitive focus likely reflects social science paradigm shift over the past half-century. This shift has generated considerable useful inquiry and theory (e.g., G. A. Miller, 2003) in many scholarly areas, including persuasion (see, e.g., Eagly & Chaiken, 1993, and many chapters in this volume). On the other hand, the concept of fear (and emotion more generally), particularly the concept of physiological arousal, has faded into the background. This leaves explanations of how fear appeals work (or don’t work) simplistic and incomplete.
With the exception of the EPPM, recent fear appeal explanations are more accurately characterized as threat appeal explanations (Witte, 1992) as they focus nearly exclusively on the cognitive processing of messages. Largely gone is any consideration of arousal (e.g., physiological or emotional) or emotional labeling. Even within the EPPM, physiological processes given a tertiary role, guiding message rejection more than message acceptance (Witte, 1992).
Given the imbalance between cognitive and emotional (particularly physiological) factors in fear appeal explanations, we know more about danger control (i.e., message acceptance) than we do about fear control (and message rejection). Moreover, much of the thinking on the fear control centers on defensive avoidance, both of which are particularly poorly defined constructs. Rather than constructs used to make predictions, fear control and defensive avoidance are typically utilized to explain (after the fact) the failure of a fear appeal. What is more, discussions of defensive avoidance frequently include a variety of responses that likely stem from different (cognitive or emotional or both) processes that occur during message processing, long afterward, or both. Different defensive avoidance processes likely produce different outcomes, but this claim has not been adequately considered or tested. As it stands, defensive avoidance seems to be more of a post-hoc shorthand description of study outcomes rather than a compelling theoretical variable. Although the accumulated data appear consistent with danger control processes, greater theory and research needs to focus on fear control processes as well.
A third area where fear appeal explanations are lacking is the consideration of moderator variables (i.e., any variable that influences the direction and/or strength of relationship between two other variables; Stiff & Mongeau, 2003). When I teach persuasion, each major topic typically involves discussing at least one moderator variable (e.g., vested interest as a moderator of the attitude-behavior relationship; Sivacek & Crano, 1982). The recent study of fear appeals is an exception. Consideration of moderator variables is important given the heterogeneity of fear appeal effects (Witte & Allen, 2000). Although early fear appeal research considered several moderator variables (e.g., source credibility, trait anxiety, participant age, etc.; Boster & Mongeau, 1984), the study of such variables has largely died. (One exception is trait anxiety, see Hale, Lemieux, & Mongeau, 1995; Witte & Morrison, 2000.)
Despite the dearth of moderator variables in modern fear appeal research, future research might fruitfully investigate several important candidates. One potentially important moderator variable is the novelty of the threat. Fear appeal explanations suggest that describing the severity and susceptibility of the threat are important components of fear appeals. Recent research, however, suggests that this might not be the case (Nabi, Roskos-Ewoldsen, & Carpentier, 2008). When the threat is new to the audience, the threat component is indeed likely critically important to a fear appeal’s success. What we don’t know is whether it is necessary to remind audience members of a threat they are already aware of. During the Cuban Missile Crisis, did U.S. residents really need to be reminded of the dangers of a nuclear attack to convince them to build a basement bomb shelter? In the present, do tobacco smokers really need to be reminded of the health threats of their habit in order to motivate them to change? Nabi et al. suggest that when the threat is well-known to participants, that presenting only the efficacy information can be more persuasive, and produce less reactance, than presenting the full fear appeal. Although their results are somewhat equivocal (they work better for men when the issue is testicular self-examination than for women for breast self-examination), they do suggest that this is an important issue for future fear appeal research.
A second set of moderator variables relevant to the effectiveness of fear appeals represent the nature of the recommended behavioral change. The fear appeal literature (and persuasion in general) seems to assume that message factors work in a one-size-fits-all fashion. This is a questionable assumption because, first, effect sizes differ across studies and, second, behavioral changes differ dramatically across message topics. Some changes are drastic, occur over time, and require consistent maintenance (e.g., dietary change to reduce blood cholesterol levels; Prochaska & DiClemente, 2005). In such cases, not only must an individual decide to make drastic changes in one’s diet, but must consistently maintain those changes when making daily dietary decisions. Other behavioral changes only need to be made once and following that decision, no further elaboration is likely necessary (e.g., signing an organ donation card). The nature of the successful persuasive message (or campaign) is likely quite different if the behavior in question is ongoing rather than discrete. Also, eliminating a behavior (e.g., quitting smoking) might involve fundamentally different persuasion processes than initiating behaviors (e.g., starting an exercise regime; Floyd et al., 2000).
Another aspect of behavioral change that fear appeal explanations fail to consider is the receiver’s readiness to make a change. Smoking cessation, for example, is likely a long and complicated process requiring several stages (e.g., Prochaska & DiClemente’s, 2005, pre-contemplation, contemplation, and preparation stages). Convincing a smoker that they should quit likely has to precede convincing people how to quit. Put another way, where an individual might be in the process of change or their readiness for making a major multibehavioral, longitudinal change are likely important moderators of the effectiveness of fear appeals. These are issues discussed in detail by the transtheoretical (stages of change) explanation (Prochaska & DiClemente, not to be confused with the stages of change model discussed earlier, Stroebe, 2000).
Reasons Centering on Research Methods
Fear appeal explanations have changed dramatically since Janis and Feshbach’s (1953) drive model. The research methods used to test these explanations, however, have changed less drastically. Thus, a second set of reasons why fear appeal explanations don’t work lies not with the explanations themselves, but with the predominant methods used to test them. Specifically, the operationalization of fear arousal (i.e., perceived fear) seems particularly problematic.
Put simply, the drive model and the EPPM suggest, essentially, that the fear appeal’s threat component creates the emotion of fear while the coping component is supposed to reduce it. The operational definition of perceived fear (in most studies) is a self-report measure of how much fear (or concern or worry) participants experienced during message reception (Witte & Allen, 2000; for exceptions, however, see Dillard & Anderson, 2004; Mewborn & Rogers, 1979). Many times self-report measure of perceived fear represents little more than a manipulation check r
ather than an important theoretical variable in and of itself.
There are two problems with the self-report measure of fear in such an experimental design. First, the perceived fear judgment participants make is inherently ambiguous. Given a typical fear appeal, recipients are likely calm at the beginning of the message, emotionally aroused in the middle (after processing the threat component), and relatively calm at the end (after processing an effective coping component). If this is indeed the case, how does he or she respond to an item asking, after the fact, how much fear (or concern or worry) they experienced during the message? If the actual level of fear varied (perhaps dramatically) during the message, what does such an overall measure really tell us?
Moreover, given the typical post-test only design used in most fear appeal studies, not only are self-report data ambiguous, they are unable to capture a great deal of important information. As a fear appeal explanation, the EPPM centers on the creation and reduction of emotional arousal as a central determinant of attitude and behavior change. The typical (post-test only) fear appeal study provides no data that speak to this critically important point. Put another way, researchers need to utilize physiological, rather than exclusively self-report indicators of fear arousal and reduction.
If physiological measures of fear arousal are so important to testing fear appeal explanations, why are self-report measures the norm? Certainly they are less expensive, don’t require sophisticated equipment, and are easier to interpret. Moreover, Witte claims that self-report measures are the preferred operational definition of fear because they are likely more sensitive than physiological measures “because self-rated fear is more global in nature and more adequately reflects an overall emotional state, while physiological arousal fluctuates substantially during the presentation of a fear appeal” (Witte, 1992, p. 331).
The SAGE Handbook of Persuasion Page 39