FEAR-AROUSING COMMUNICATIONS AND PERSUASION

The impact of vulnerability on processing and accepting fear appeals

Natascha de Hoog

De Hoog, Natascha Fear-arousing communications and persuasion: The impact of vulnerability on processing and accepting fear appeals ISBN: Drukker: 90-8559-084-1 Optima Grafische Communicatie

FEAR-AROUSING COMMUNICATIONS AND PERSUASION

The impact of vulnerability on processing and accepting fear appeals

ANGSTCOMMUNICATIES EN OVERTUIGING
De invloed van kwetsbaarheid op verwerking en acceptatie van angstboodschappen (met een samenvatting in het Nederlands)

Proefschrift Ter verkrijging van de graad van doctor aan de Universiteit Utrecht op gezag van Rector Magnificus Prof. dr. W.H. Gispen ingevolge het besluit van het College van Promoties in het openbaar te verdedigen op donderdag 27 oktober 2005 des ochtends te 10.30 uur

door Natascha de Hoog geboren op 11 november 1976, te Rotterdam

Promotor: Prof. dr. W. Stroebe Co-promotor: Dr. J.B.F. de Wit

The research reported in this thesis was funded by the Research Institute for Psychology & Health, an institute accredited by the Royal Dutch Academy of Arts and Science.

Contents

Chapter 1

Introduction Theoretical approaches to the study of fear-arousing communications A stage model of processing of fear-arousing communications Overview

9 10 24 28

Chapter 2

The impact of fear appeals on processing and accepting action recommendations: the key role of vulnerability Experiment 2.1 Method Results Discussion

33 38 38 41 45

Chapter 3

The impact of fear appeals on processing and accepting action recommendations: the role of response costs Experiment 3.1 Method Results Discussion

51 55 55 58 63

Chapter 4

The impact of fear appeals on processing and accepting action recommendations: the role of dispositional optimism and neuroticism Experiment 4.1 Method Results Discussion

69 72 72 75 78

Chapter 5

Getting the whole picture: How biased processing leads to persuasion Experiment 5.1 Method Results Discussion Experiment 5.2 Method Results Discussion Experiment 5.3 Method Results Discussion General discussion

81 86 87 88 90 91 91 93 94 95 96 97 103 104

Chapter 6

The impact of fear appeals on information processing and persuasion: A meta-analysis Method Results Discussion

109 117 120 136

Chapter 7

General discussion

143

References

157

Summary

169

Samenvatting

175

Chapter 1: Introduction

Each time smokers take a cigarette out of their cigarette pack they are confronted with a warning displayed on the front of the pack that may read Smoking causes lung cancer, Smoking can make you infertile or Smoking causes a slow and painful dead. The warnings on packs of cigarettes are supposed to make smokers concerned about the hazardous consequences of smoking in such a way that they will stop smoking, yet most smokers continue to smoke (Defacto, 2002). Does that mean that the warnings are ineffective? If these warnings are indeed ineffective, what should be put on a cigarette pack to convince smokers to quit smoking? For example, would it help if the message made smokers feel like they themselves were personally at risk to die from smoking? Or would it help if just as in, for instance, Canada pictures of people suffering from cancer were displayed on the packs? Alternatively, would it help if smokers were informed how to stop smoking? These are some of the questions that will be addressed in the present thesis. Warnings that are printed on cigarette packs are just one example of feararousing communications. Fear-arousing communications have been used in health education campaigns since the 1950s, and have been applied to numerous different topics. The most well known health education campaigns that have used fear-arousing communications in the Netherlands are probably those targeting excessive alcohol consumption (DRANK maakt meer kapot dan je lief is and Do you know do you care?; NIGZ, 2000), drunk driving (3VO, 2003) and the unsafe use of fireworks (Je bent een rund als je met vuurwerk stunt; SIRE, 2003). Fear-arousing communications usually consist of two parts, namely (1) a fear appeal that stresses the severity of and personal vulnerability to a health risk, and (2) an action recommendation that emphasizes how to reduce or eliminate the health risk. The basic assumption underlying fear-arousing communications is that the more one succeeds in making individuals concerned about the negative consequences of a certain behavior, the higher will be the probability that they change their health impairing behavior. Over fifty years of empirical research on fear-arousing communications has resulted in a large body of evidence that shows that high fear messages are generally more effective than low fear messages in changing individuals attitudes, intentions and behavior (see Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000). However, these empirical studies have not succeeded in fully explaining the processes underlying the effects of fear-arousing communications on persuasion. Moreover, insufficient attention has been devoted to explaining why health education campaigns using fear-arousing communications are sometimes ineffective in changing individuals health impairing behaviors. In addition, fear-arousing communications in both

10

Chapter 1

research and practice have mainly focused on the effects of stressing the severity of a health risk, but much less attention has been paid to the importance of making individuals feel vulnerable to a health risk. The current thesis will examine the processes underlying persuasion through fear-arousing communications, in order to provide more insight into the conditions under which fear-arousing communications are (in)effective, and, more importantly, to specify how and why they are (in)effective. A stage model of the processing of fear-arousing communications is used as the theoretical framework (Das, de Wit & Stroebe, 2003; Stroebe, 2000), which will be described in detail in this introduction. I will start of by giving an overview of all the main theoretical approaches to the study of fear-arousing communications that have been proposed since the 1950s, ending with a description of the stage model that has been underlying the experiments reported in this thesis. This description will be followed by an overview of the experiments reported in each of the chapters in this thesis. Theoretical approaches to the study of fear-arousing communications Drive reduction model Early research on fear appeals in the 1950s was guided by the assumptions of the drive reduction model (Hovland, Janis & Kelley, 1953; Janis, 1967). The main assumption of the drive reduction model is that fear or emotional tension has the functional properties of a drive. A drive is a motivator for action, and the stronger the drive, the more motivational function it possesses. According to the drive reduction model, when individuals are presented with threatening information they will be motivated to search for responses that reduce the threat. Whatever response reduces the intensity of the emotional tension produced by a fear appeal will be reinforced, and will become part of ones permanent response repertoire. However, when a recommendation fails to reduce the fear level (i.e. if there is residual fear), because the recommendation is perceived as either irrelevant to the threat or impossible to carry out, spontaneous responses will be tried out to find a response that reduces emotional tension. In sum, higher fear should result in more persuasion, but only if a recommended action is perceived as effective in preventing the threat. When the recommendation fails to reduce the fear level, the drive reduction model (Hovland et al., 1953; Janis & Feshbach, 1953) describes several defensive reactions to a fear appeal that may serve to reduce fear. The first is defined as inattention to message content, which results either from motivated attempts to avoid thinking about the threat or from cognitive factors such as reduced concentration due to too much fear. Secondly, individuals could react to a fear appeal with aggression toward the communicator, which shows itself

Introduction

11

in rejection of the message content or a derogation of the message source. Thirdly, when no defensive reaction to a health threat is perceived as successful in reducing emotional tension, delayed defensive avoidance may occur, which means that individuals will try to actively avoid any subsequent thinking about the health threat or minimize the importance of the threat. Especially high levels of fear are proposed to evoke defensive reactions that could undermine persuasion. The introduction of these defensive reactions alters the relationship between the level of aroused fear and persuasion. Although an increasing level of emotional tension should lead to increased persuasion, at some point the emotional tension will reach a level at which the reassuring recommendation will not sufficiently reduce the tension. Residual emotional tension might give rise to defensive avoidance, causing a decrease in persuasion. This results in a curvilinear relationship between the level of emotional arousal and persuasion. Extensions of the drive reduction model Extensions of the drive-reduction model were later suggested by Janis (1967) and McGuire (1968). Janis (1967), with his family of curves model, elaborated the prediction of a curvilinear relationship between fear arousal and persuasion, and proposed that the effects of fear appeals on persuasion were mediated by reception (i.e., attention and comprehension) of and yielding (i.e., acceptance) to message content. Janis proposed that fear arousal would result in facilitating and interfering effects on reception of and yielding to message content. Facilitating effects of fear arousal are assumed to be caused by a heightened awareness and a heightened need for reassurance. A heightened awareness will result in increased attention to a threat and thinking about ways to deal with this threat. Whereas a heightened need for reassurance increases an individuals readiness to accept an action recommendation. These facilitating effects are proposed to be more prominent in low to moderate fear than in higher levels of emotional arousal. Interfering effects of fear arousal on reception of and yielding to message content, according to Janis (1967), are mainly defensive avoidance tendencies that interfere with acceptance. Interfering effects are more prominent in high levels of fear, than low or moderate levels, because of a higher need for reassurance. The combined effects of facilitating and interfering processes are proposed to result in a curvilinear relationship between fear arousal and persuasion: low to moderate fear arousal will lead to higher levels of persuasion, whereas higher levels of fear will result in decreased persuasion. McGuires reception-yielding model (1968) describes a similar theory of the relationship between fear arousal and acceptance as the family of curves model (Janis, 1967). McGuires reception-yielding model (1968) assumes a curvilinear relationship between fear arousal and persuasion, and proposes that

12

Chapter 1

persuasion depends on the outcome of two processes, reception of the persuasive message and yielding to the action recommendation in the message. McGuire proposed that the fear aroused by exposure to a feararousing communication can act as a drive or a cue. When fear acts as a drive, it can motivate individuals to accept a protective action recommendation. However, when fear acts as a cue, it can elicit responses that interfere with the acceptance or reception of a recommended action. The reception-yielding model proposes that these cues and drives combined result in an overall curvilinear relationship between the fear aroused and attitude change, in which a moderate amount of fear arousal should produce the most attitude, intention and behavior change. The main contribution of the drive reduction model and its extensions (Hovland et al., 1953; Janis 1967; McGuire, 1968) has been the focus on defensive reactions that individuals may display in reaction to a fear-arousing communication. In addition, a number of issues that are explicitly addressed in later theories are also already presented in the drive reduction model, albeit not prominently. For example, the drive reduction model addresses the role of perceived efficacy of a recommendation and ones ability to carry it out, two concepts that were incorporated and featured prominently in later theories. Support for the assumptions of the drive reduction model comes mainly from Janis and Feshbachs (1953) classic dental hygiene experiment. Janis and Feshbach found greater attitude and behavior change when high school students were presented with mild rather than strong fear appeals regarding dental hygiene. However, this finding has only rarely been replicated, and most fear appeal studies have failed to find support for the curvilinear relationship implied by the drive reduction model and its extension. The majority of experiments on fear-arousing communications have found that higher levels of fear lead to more persuasion than lower levels (see Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000). The drive reduction models assumption that the persuasiveness of a feararousing communication depends on both the level of fear aroused and the effectiveness of the recommendation to reduce fear (Hovland et al., 1953), leads to the prediction of an interaction between these two factors. However, most fear appeal studies have been unable to find significant fear-efficacy interactions on persuasion (Chu, 1966; Dabbs & Leventhal, 1966; Griffeth & Rogers, 1976; Maddux & Rogers, 1983; Rogers & Mewborn, 1976; Rogers & Thistlethwaite, 1970). Thus, research has resulted in only limited support for the predictions of the drive reduction model. Parallel response model In reaction to the shortcomings of the drive reduction model (Hovland et al., 1953), the parallel response model was developed (Leventhal, 1970). In this

Introduction

13

model a mediator between environmental threat and action was introduced, namely threat appraisal, and the notion that emotional arousal is a necessary antecedent of the adaptation to danger was abandoned. According to the parallel response model, an environmental threat is cognitively evaluated by the individual and this appraisal can give rise to two parallel or independent processes, namely danger control and fear control. Fear control involves the initiation of responses in an attempt to reduce the unpleasant feeling of fear and is proposed to be guided largely by internal cues. Fear control entails an individuals attempt to control the unpleasant affect evoked by fear arousal by denying or avoiding the threat. Danger control is a problem solving process, which involves the selection and guidance of responses aimed at averting the danger. The information used in the danger control process consists largely of information about the environment. Danger control involves cognitive appraisal of the coping responses. According to the parallel response model, fear control and danger control are two separate processes, which might interfere with one another (Leventhal, 1970). These two processes are parallel instead of sequential, which implies that the process of danger control, which may lead to adaptive action, is presumed to occur regardless of whether someone feels frightened. Both processes affect the selection and guidance of instrumental actions. In fear control, these actions are directed at controlling subjectively experienced fear and they may have little or no effect on averting or coping with the actual threat. In danger control, these actions are directed at coping with the threat. Therefore, the acceptance of a recommendation is proposed to depend mainly on the process of danger control. The importance of the parallel response model to the understanding of feararousing communication lies in its movement away from the notion of fear as the central explanatory concept in persuasion resulting from fear-arousing communications. Instead, the parallel response model recognizes that individuals responses to fear-arousing communications involve adaptive behavior motivated by a desire to avert the anticipated danger and expectations as to how this can be done. However, the parallel response model lacks precision and it does not specify the conditions that lead to each of the two processes of fear and danger control. It also does not specify when and how these two processes interact or how people switch from one process to the other. The model only broadly categorizes emotional and cognitive responding as two distinct reactions to a health risk. Because of the parallel response models lack of precision and specification of relevant processes, it has been difficult to test this model empirically.

14

Chapter 1

Expectancy-value models Later theoretical models of fear appeals all built on Leventhals danger control process and tried to specify the cognitive process involved in danger control. These efforts were guided by the expectancy-value principle that holds that behavior is a function of its expected consequences and the perceived value of these consequences. Expectancy-value models assume that decisions between different courses of action are based on two types of cognition: the subjective probability that a given action will lead to an expected outcome, and the evaluation of these action outcomes. Individuals will choose among alternative courses of action that action that will be most likely to lead to positive consequences or avoid negative consequences. Expectancy-value models like Rosenstocks health belief model (1974), Rogers protection motivation theory (1975, 1983) and Suttons subjective expected utility model (1982), all tried to assess in a cognitive and rational manner what made individuals engage in health enhancing behavior. Health belief model Even though the health belief model (Rosenstock, 1974) was not developed especially to examine the effects of fear-arousing communications, but as a more general model of explaining when individuals engage in health enhancing behaviors, the model can be, and has been applied on occasion, to feararousing communications (Becker et al., 1977; Kirscht, Becker & Haefner, 1978). The health belief model (Janz & Becker, 1984; Rosenstock, 1974) assumes that the likelihood that individuals engage in a certain health enhancing behavior depends on the extent to which individuals believe that they are personally susceptible to a particular disease and their perceptions of the severity of the consequences of a disease. Susceptibility and severity combined are assumed to determine the perceived threat of the disease. When individuals perceive a threat of contracting a disease, the likelihood of engaging in a particular health enhancing behavior will depend on the extent to which individuals believe that the health behavior results in certain benefits that outweigh the barriers associated wit the health behavior (see Figure 1.1). In addition, the model proposes that certain cues to action can trigger health behavior when appropriate beliefs are held. These cues can include a wide range of triggers, namely individual perceptions of symptoms, social influence and even health education campaigns. The relationship between the variables specified by the health belief model has never been explicitly stated, although in most studies an additive combination is assumed, while a multiplicative function seems intuitively more plausible, at least for the combination of vulnerability and severity (see also protection motivation theory).

Introduction

15

Figure 1.1. The health belief model Perceived susceptibility Belief in a personal health threat

Perceived severity

Health behavior Perceived benefits

Perceived barriers

Belief in the effectiveness of a health behavior

Research on the impact of the components specified by the health belief model on protective health behavior has generally been supportive. Studies have found small yet significant positive associations between protective behavior and perceptions of susceptibility, perceptions of severity, and perceived benefits, as well as significant negative associations between perceived barriers and protective behavior (see Harrison, Mullen & Green, 1992; Janz & Becker, 1984). Protection motivation theory Protection motivation theory (Rogers, 1975, 1983) also elaborated on the parallel response models process of danger control by specifying the processes of cognitive appraisal. Rogers identified four message components of fear appeals that are assumed to motivate individuals to implement a protective action: severity of a threat, probability of the threats occurrence, the efficacy of a protective response, and ones ability to perform the protective response. Protection motivation theory differentiates between threat appraisal and coping appraisal processes. These two forms of appraisal are assumed to interact with one another: when both appraisal of the threat and appraisal of the coping responses are perceived as high, the most protection motivation and subsequent message acceptance is proposed to occur. There are two versions of protection motivation theory, namely the original version and a revised version both will be described in more detail in the next sections.

16

Chapter 1

The original protection motivation theory The aim of the original formulation of protection motivation theory (Rogers, 1975) was to specify the message components of fear appeals that were assumed to motivate individuals to implement a protective action. Rogers originally distinguished three components of fear appeals that determined an individuals response to a fear-arousing communication: a) the magnitude of noxiousness of a depicted event, b) the probability of the events occurrence, and c) the efficacy of a protective response. According to Rogers, protection motivation arises from the cognitive appraisal of a depicted event as severe and likely to occur, along with the belief that an action recommendation is effective in preventing the occurrence of the aversive event. The primary assumption was that these three variables have a fundamental influence on the motivation to implement a protective action, with Rogers assuming a multiplicative link between the processes. The amount of protection motivation was thought to be a monotonically increasing function of the product of these three cognitive appraisals. Revised protection motivation theory In his reformulation of protection motivation theory, Rogers (1983) extended the model into one that differentiates between threat appraisal and coping appraisal processes (see Figure 1.2). In addition, Rogers added self-efficacy as a fourth component of a fear appeal, which is defined as individuals confidence in their ability to carry out a protective response (Bandura, 1977). Rogers (1983) also added the concepts of response costs (i.e., perceived costs of engaging in adaptive behavior) and perceived rewards (i.e., benefits of maladaptive responses) to the model. According to Rogers (1983), threat appraisal is an appraisal of vulnerability to and severity of a threat and of the rewards associated with the maladaptive response. In appraising the threat, individuals are assumed to continue to engage in maladaptive behaviors if the rewards of performing the maladaptive behavior are greater than the perceived severity of the threat and their perceived vulnerability to the threat. Thus, increases in rewards heighten the probability of a maladaptive response while increases in perception of threat decrease the probability of a maladaptive response. Coping appraisal involves the appraisal of response efficacy and self-efficacy of the recommended protective action and the costs of adaptive responses. In appraising coping responses, protection motivation theory assumes that increases in perceived response efficacy and self-efficacy increase the likelihood of adaptive behavior while increases in response costs decrease the likelihood of adaptive behavior. These processes of threat appraisal and coping appraisal are assumed to interact with one another and together result in protection motivation. In sum, according to protection motivation theory (Rogers, 1983; Rogers &

Introduction

17

Mewborn, 1976), individuals will accept an action recommendation if they feel vulnerable to a severe health risk and if the recommendation is seen as an effective way to reduce the threat. The main contribution of protection motivation theory has been the specification of some of the important components of fear-arousing communications. Additionally, protection motivation theory has attempted to specify the cognitive appraisals of each of these components and the relationship between these cognitive appraisals and subsequent protective behavior.
Figure 1.2. Protection motivation theory Severity Vulnerability Threat appraisal Intrinsic rewards Extrinsic Response efficacy Self-efficacy Response costs

Protection motivation

Health behavior

Coping appraisal

Most research on fear appeals has incorporated one or more factors specified by protection motivation theory, and all main factors specified in protection motivation theory have been found to affect measures of persuasion: Increases in severity, vulnerability, response efficacy and selfefficacy all result in more attitude, intention and behavior change (see Floyd, Prentice-Dunn, & Rogers, 2000; Milne, Sheeran, & Orbell, 2000; Witte & Allen, 2000). In addition, decreases in maladaptive response rewards and adaptive response costs have also been found to promote change in attitudes, intention and behavior. However, even though support was found for main effects of severity, vulnerability, response efficacy and self-efficacy on measures of persuasion, little support was found for the 3-way interaction between severity, vulnerability and response-efficacy (Kleinot & Rogers, 1982; Maddux & Rogers, 1983; Rogers & Mewborn, 1976; Rogers & Thistlewaite, 1970; Sutton & Eiser,

18

Chapter 1

1984) as proposed by the original formulation of protection motivation theory. More support has been found for the threat appraisal by coping appraisal interaction predicted by the revised protection motivation theory (Rogers, 1983). Many studies have found two-way interactions between one of the threat appraisal variables (i.e., severity or vulnerability) and one of the coping appraisal variables (i.e., response efficacy or self-efficacy) on outcome measures such as attitudes, intention, and behavior (Maddux & Rogers, 1983; Rogers & Thistlewaite, 1970; Mulilis & Lippa, 1990; Wurtele & Maddux, 1987). However, specific interactions between the four variables have been inconsistent across studies. Whereas some studies have found that vulnerability interacts with response efficacy (e.g., Rogers & Thistlewaite, 1970; Mulilis & Lippa, 1990), others have found that severity interacts with self-efficacy to change attitudes, intentions or behavior (e.g., Maddux & Rogers, 1983; Wurtele & Maddux, 1987). Moreover, even though a large number of studies have found some interaction-effects between threat and coping variables, an equally large number of studies have been unable to find any of these interaction-effects (see Rogers & Prentice-Dunn, 1997). Subjective expected utility model The subjective expected utility model proposed by Sutton (1982) holds that individuals choose from competing alternatives a course of action that has the greatest subjective expected utility. A subjective expected utility is defined as a function of the subjective values or utilities attached to the possible outcomes of the alternative and the subjective probabilities that the alternative will lead to those outcomes. According to the subjective expected utility model, the decision facing individuals who have been exposed to a fear-arousing communication can be represented as a choice between continuing the health impairing behavior and trying to change the behavior (Sutton, 1982; Sutton & Eiser, 1984). The model proposes that the decision to change health impairing behavior will depend on three factors: (1) the utility individuals attach to the health impairment; (2) the reduction in the subjective probability of contracting this health impairment when individuals successfully change their behavior (probability difference); and (3) the subjective probability of succeeding in changing ones behavior, given that an attempt is made (confidence). According to the subjective expected utility model, these three factors should combine multiplicatively to determine the decision to try to change behavior. Studies assessing the effectiveness of Suttons subjective expected utility model in explaining behavior change have produced limited support. Although significant effects were found for both the probability difference and the utility of the health damage on intentions (Sutton & Eiser, 1984; Sutton & Hallet,

Introduction

19

1988), only mixed support was found for confidence (Hallet & Sutton, 1986; Sutton & Eiser, 1984). Moreover, no evidence has been found for the multiplicative combination of utilities and subjective probabilities (Sutton & Eiser, 1984; Sutton & Hallet, 1988, 1989). Extended parallel process model The extended parallel process model (Witte, 1992, 1994) combines and elaborates ideas from the parallel response model (Leventhal, 1970), protection motivation theory (Rogers, 1983), as well as the drive-reduction model (Hovland et al., 1953), and reemphasizes the role of fear in processing and acceptance of fear-arousing communication. Leventhals parallel response model forms the basis of the extended parallel process model, with ideas from protection motivation theory explaining danger control, and portions of the drive-reduction models explaining fear control. By combining ideas from these theories the extended parallel process model attempts to explain how, through danger control processes, fear-arousing communications can be effective in changing attitudes, intention and behaviors, and how, through fear control processes, they can be ineffective. According to the extended parallel process model (Witte, 1992) exposure to a fear-arousing communication initiates two appraisal processes, namely threat appraisal and coping appraisal. Individuals will appraise the threat that is portrayed in a fear-arousing communication. The more individuals believe they are vulnerable to a serious health risk (i.e., high perceptions of threat), the more motivated they are to engage in coping appraisal. If the health risk is perceived as irrelevant or insignificant (i.e., low perceptions of threat), the extended parallel process model proposes there is no motivation to process a fear-arousing communication any further, and individuals will simply ignore the remainder of a fear-arousing communication. In contrast, when a health risk is believed to be severe and individuals feel vulnerable, the extended parallel process model assumes that individuals will become scared. The fear aroused should motivate individuals to take any action that will reduce their fear. Perceived efficacy of the recommended action (i.e., response efficacy and self-efficacy) will determine whether individuals, who believe that they are susceptible to a serious risk, will engage in danger or fear control (see Figure 1.3). Witte (1992, 1994) assumes that individuals will mainly engage in danger control when they perceive the recommended action as effective in reducing the health risk, and they will mainly engage in fear control when they perceive the recommended action as ineffective in reducing the threat, or when they feel unable to perform the recommended action. In that case (i.e., high perceived threat and low perceived efficacy) defense motivation is elicited, which the extended parallel process model defines as individuals focusing on eliminating their fear through denial, defensive avoidance or reactance. In

20

Chapter 1

sum, the extended parallel process model assumes that perceived threat determines the magnitude of the response to a fear-arousing communication, whereas perceived efficacy determines the direction of the response (i.e., danger or fear control).
Figure 1.3. Extended parallel process model Protection motivation Message acceptance

Perceived efficacy (Response efficacy & Selfefficacy) Perceived threat (Susceptibility & Severity)

Fear Defense motivation Message rejection

No threat perceived (no response)

Note Adapted from Witte (1994)

The extended parallel process model has succeeded in integrating ideas of both protection motivation theory (Rogers, 1983) and the parallel response model (Leventhal, 1970), and has extended these ideas by identifying how threat appraisal and coping appraisal relate to each other, and by specifying the role of perceived fear in threat and coping appraisal. Research testing the predictions of the extended parallel process model has mainly been supportive (McMahan, Witte, & Meyer, 1998; Morman, 2000; Witte, 1994; Witte & Morrison, 2000). Studies have found effects of both threat and efficacy on intention and behavior, with intention and behavior change being the highest when both perceptions of threat and efficacy were high. However, the predicted interaction-effect between threat and efficacy has not always been found (see the previous section on protection motivation theory). Fear arousal models Recently, attempts have been made to integrate or elaborate the fear appeal theories described in the previous sections, by emphasizing the importance of fear arousal (LaTour, & Rotfeld, 1997; Ruiter, Abraham & Kok, 2001; Tanner, Hunt & Eppright, 1991). LaTour and Rotfeld (1997) propose an arousal model in which they assume a positive monotonic relationship between fear arousal and persuasion. They assume that the amount of fear aroused by a threatened

Introduction

21

event should correspond to the perceived probability of a threatened event. Integrating ideas from protection motivation theory (Rogers, 1983) into their model, La Tour and Rotfeld (1997) propose that the perceived probability of the event, as well as individual feelings of being able to engage in the protective action, should generate energy, which in turn should have a positive effect on attitudes and intentions. Fear arousal is proposed to have direct and indirect effects on persuasion. Fear arousal can affect intentions indirectly through attitudes or directly by a seemingly instantaneous process that is followed by a cognitive focus on behavioral intentions. Ruiter and colleagues (2001) integrated ideas from protection motivation theory (Rogers, 1983) and the extended parallel process model (Witte, 1992), emphasized the important role of fear arousal in both threat appraisal and coping appraisal, and specified conditions under which fear arousal can directly influence intentions. Their model assumes that exposure to information about a health risk arouses fear, which may generate fear control processes or affect protection motivation independently of threat perceptions (Ruiter et al., 2001). They propose complex and possibly opposing relationships between fear arousal and protection motivation. Fear arousal is assumed to strengthen threat perceptions, affect protection motivation, or induce automatic fear control processes. Fear control processes in turn, are assumed to result in either protection motivation that heightens intentions, or may lead to denial and other forms of avoidant coping. In addition, the impact of fear arousal on protection motivation is proposed to be moderated by the intensity of the emotion and individual characteristics. The ordered protection motivation model (Eppright, Tanner, & Hunt, 1994; Tanner et al., 1991) elaborates ideas from protection motivation theory (Rogers, 1983) by specifying the role that fear arousal can play in the process of threat appraisal. The ordered protection motivation model (Tanner et al., 1991) proposes that when threat appraisal precedes coping appraisal, increased fear arousal will motivate attention to the threat and coping appraisal. The heightened levels of threat and coping appraisal in turn are assumed to lead to greater adaptive behavioral intentions. Fear arousal in the absence of efficacy information is assumed to cause maladaptive responses that reduce the probability of threat perceptions without removing the actual danger. Dual process theories Even though all major fear appeal theories, as described in the previous sections, assume that fear arousal and/or cognitive processes mediate persuasion, none of these theories have made predictions about information processing, and measures of information processing (e.g., cognitive responses) have been virtually absent in research on fear-arousing communications. This

22

Chapter 1

situation changed with the advent of dual-process theories of persuasion (Chaiken, 1980; Eagly & Chaiken, 1993; Petty & Cacioppo, 1986), which were also applied to account for the impact of fear-arousing communications on persuasion (e.g., Gleicher & Petty, 1992; Liberman & Chaiken, 1992). With the introduction of dual-process theories the emphasis in research on fear appeals shifted from studying the effects of communication variables on attitudes and behavior, to studying the processes by which these effects are brought about. More specifically, attention was drawn to studying the effects of fear appeals on message processing. Dual-process theories of persuasion, like the elaboration likelihood model (Petty & Cacioppo, 1986) and the heuristic-systematic model (Chaiken, 1980; Eagly & Chaiken, 1993), focus on information processing as well as persuasive outcomes. Dual-process models differentiate between two distinct modes of information processing, which differ in the extent to which individuals are assumed to engage in effortful processing of the information contained in a message. Although the elaboration likelihood model and the heuristicsystematic model are different in some ways, their general conceptualization of these two modes of processing is rather similar. The first mode of information processing is called systematic processing in the heuristic-systematic model (e.g., Chaiken, 1980), and is named central route in the elaboration likelihood model (e.g., Petty & Cacioppo, 1986). This processing mode involves effortful scrutiny of message arguments and other relevant information. In contrast, the second mode of information processing (which is called heuristic processing in the heuristic-systematic model, and named peripheral route in the elaboration-likelihood model) involves use of heuristics, which are simple rules of inference such as experts are always right. The peripheral route is defined more broadly, and involves an additional variety of low-effort mechanisms such as conditioning and social identification. These mechanisms are based on peripheral cues, stimuli that are capable of affecting persuasion but are not related to message content, for example the source of a communication.1 According to dual process theories (Chaiken, 1980; Petty & Cacioppo, 1986), individuals will base their attitudes about a persuasive message either on thoughts provoked by the message content or on factors unrelated to the content, such as cues or heuristics associated with the message. Which mode of processing will be used depends on individuals motivation and ability to carefully process a persuasive message. According to dual-process theories, processing motivation and ability determine the elaboration likelihood of a message (i.e., scrutiny of issue-relevant arguments contained in the message). At low levels of motivation and ability, heuristic processing prevails, while at increasing levels of motivation and ability systematic processing becomes increasingly dominant, and the impact of heuristic mechanisms fades.

Introduction

23

The elaboration likelihood model proposes several factors that can influence individuals motivation and ability to process information systematically (Petty & Cacioppo, 1986). The ability to process information systematically is assumed to depend on, among others; the amount of distraction from a message, message repetition, prior knowledge, and message comprehensibility. Motivation to process information systematically is proposed to depend on among others: personal relevance, need for cognition, and personal responsibility. One of the most important factors proposed by dual process theories to influence motivation for systematic processing is the personal relevance of a persuasive communication. Personal relevance refers to the importance of a communication for an individual (Petty & Cacioppo, 1986). Higher levels of personal relevance are assumed to increase processing motivation and, consequently, elaboration likelihood (see Johnson & Eagly, 1989). This concept of personal relevance is related to the concept of personal vulnerability/susceptibility used by most fear appeal theories (e.g., Rogers, 1983; Witte, 1992). Both concepts refer to the importance of a persuasive message for an individual. When an individual is both motivated and able to carefully process a persuasive message, systematic processing will occur (Petty & Cacioppo, 1986). Persuasion then is assumed to depend on the quality of the arguments contained in the message: strong supporting arguments will lead to more persuasion than weak supporting arguments. When an individual is not motivated or unable to process a persuasive message, heuristic processing can occur. If that is the case, persuasion will not depend on intensive processing of the message content, and the quality of the arguments used in the persuasive message will have no effect on the individuals attitudes. Instead, any kind of cognitive, affective or social cue or heuristic may produce attitude change (e.g., number of arguments used in an action recommendation or emotions evoked by the message). Whereas dual-process theories initially only focused on accuracy-motivation (i.e., the motivation to hold correct attitudes), as described previously, later alternative motives were incorporated into dual-process theories, which were elaborated into the multiple-motive heuristic-systematic model (Chaiken, Liberman, and Eagly, 1989; Chen & Chaiken, 1999). This model proposes a range of motives that systematic and heuristic processing may serve in addition to accuracy-motivation, including defense motivation. Defense motivation is the desire to form or to defend particular attitudinal positions. Defense-motivated individuals strive to confirm the validity of a preferred position and to denounce non-preferred ones, and this is assumed to result in biased message processing. Thus, defense-motivated individuals will process information in a way that best supports individual beliefs. Applying dual-process theories to fear-arousing communications, fear

24

Chapter 1

arousal is assumed to have two effects, namely act as a motivator to induce recipients to engage in intensive and thoughtful message processing, and induce defense motivation, which will lead to biased message processing (Gleicher & Petty, 1992; Liberman & Chaiken, 1992). Whereas the unbiased or accuracy-motivated perceiver assesses the validity of attitude-relevant information in the interest of achieving a well-founded position, the processing goal of defense-motivated recipients is to confirm the validity of a preferred position and disconfirm the validity of non-preferred positions. Thus, defensemotivated individuals will process and perceive information in ways that best support their own beliefs. This implies that information that is congruent with a particular preferred position will be judged as more valid than information that is incongruent with this position (e.g., Ditto & Lopez, 1992; Pyszczynski & Greenberg, 1987). Studies that assessed the impact of vulnerability and severity on the processing of fear-arousing communications have found evidence for the biased processing of health threat information. These studies in particular have shown that individuals are more critical of evidence that is highly health threatening than of less threatening evidence (Ditto & Lopez, 1992; Janis & Terwilliger, 1962; Kunda, 1987; Liberman & Chaiken, 1992; Reed & Aspinwall, 1998; Sherman, Nelson, & Steele, 2000). Thus, these studies support the assumptions of biased processing through defense motivation. A stage model of processing of fear-arousing communications The fear appeal theories and the dual-process theories of attitude change described in the previous sections have provided two distinct theoretical frameworks to explain the persuasive impact of fear-arousing communications. These two different outlooks on the workings of fear appeals have been integrated and extended in one theoretical framework, the stage model of processing of fear-arousing communications (Das et al., 2003; Stroebe, 2000). The stage model integrates ideas derived from dual-process theories (e.g., Chaiken, 1980), with those of earlier theories of fear-arousing communications (Leventhal, 1970; Rogers, 1983; Witte, 1992) into a theoretical framework that emphasizes the importance of severity of and, in particular, vulnerability to a health risk in promoting the acceptance of fear-arousing communications. In addition, the stage model elaborates the processing of fear-arousing communications by specifying the cognitive processes that occur, depending on perceived severity and vulnerability. In line with most fear appeal theories (e.g., Rogers, 1983; Witte, 1992), the stage model assumes that individuals exposed to a fear-arousing communication engage in two types of appraisal, namely appraisal of the health risk, and appraisal of coping strategies available for reducing or

Introduction

25

eliminating the health risk. The model further observes that these two appraisal processes parallel the two components of fear-arousing communications, namely the fear appeal, which emphasizes the severity of and vulnerability to the health risk, and the action recommendation, which provides information on how to avoid the health risk. Expanding upon the differentiation between appraisal of threat and appraisal of coping (e.g., Rogers, 1983; Witte, 1992), the stage model makes specific differential predictions for the appraisal processes at each of these two stages, based on assumptions derived from dual-process theories of attitude change (e.g., Chaiken, 1980). Appraisal of the magnitude of a threat, based on the information in a fear appeal about the severity of and personal vulnerability to a risk, is assumed to determine both individuals processing mode (i.e., depth of processing) and their processing goal (i.e., accuracy or defense motivation), which is visually represented in Figure 1.4. According to the stage model, if a health risk is trivial and individuals do not feel vulnerable to the health risk (i.e., low vulnerability and low severity), they are unlikely to be motivated to invest much effort in thinking about the contents of the communication and they might rely on heuristic processing modes.
Figure 1.4. The impact of severity and vulnerability on processing mode and goal

Severity
Low Low
Heuristic processing Accuracy motivation

High
Systematic processing Accuracy motivation Systematic processing Defense motivation

In contrast, if individuals feel vulnerable to a minor health risk (i.e., high vulnerability and low severity), the feeling of vulnerability should be sufficient motivation to invest effort into systematically processing the contents of the communication. This is in line with the assumption from dual-process theories that personal relevance, which is a similar concept to perceived vulnerability, is an important motivator for systematic processing (e.g., Eagly & Chaiken, 1993). Similarly, if individuals do not feel vulnerable, but a health risk is depicted as severe (i.e., low vulnerability and high severity), individuals are likely to invest effort in processing the contents of a communication describing a severe risk, because it is useful to be well informed about a serious health risk, even if the danger is not immanent. In other words, the stage model proposes that

Vulnerability

High

Systematic processing Accuracy motivation

26

Chapter 1

systematic processing can also occur in the absence of high personal relevance. In this respect the stage models assumption about information processing differs from the dual-process theories that state that personal relevance is an important precondition for the motivation for systematic processing (e.g., Eagly & Chaiken, 1993). The situation the stage model elaborates on most is the condition in which both vulnerability and severity are high. According to the stage model, if individuals are exposed to information about a severe health risk and if they feel vulnerable to this risk (i.e., high vulnerability and high severity) they will feel personally threatened. In this situation, individuals will especially experience threat to their self-definitional belief of being healthy, which will arouse defense motivation as well as the motivation to carefully scrutinize the information given. The stage model assumes that processing will be systematic instead of heuristic because any communication describing a serious personal health threat is likely to require a thorough evaluation (Chaiken et al., 1989). According to the stage model, defense motivation can manifest itself not only in avoidance reactions, as previous fear appeal models have proposed (e.g., Witte, 1992), but more likely is evident in systematic processing that is biased. The direction of the bias aroused by defense motivation is proposed to vary with the type of appraisal. The stage model proposes a negative processing bias in the appraisal of the threat and a positive processing bias in the appraisal of coping strategies. In appraising the health risk (Stage 1), defense-motivated individuals will attempt to minimize the threat by critically looking at the content of a fear appeal. They will try to criticize and downplay the information in order to reduce the threat. Defense-motivated individuals will engage in a biased search for inconsistencies and logical errors, and their evaluation of evidence will consequently be biased in the direction of their preferred conclusion (Ditto & Lopez, 1992; Liberman & Chaiken, 1992; Sherman et al., 2000). If this strategy proves unsuccessful, because even biased processing is constrained by evidence and rules of inference (Kunda, 1987), individuals will have to accept that they are personally at risk. In this case, the subsequent processing of a recommendation will be biased as well (Stage 2), but in the opposite direction. Under defense motivation, an action recommendation is proposed to be maximized. Because an action recommendation contains a possible solution to the health risk, it can moderate the negative emotions and cognitions evoked by the threat appraisal. At this stage, the processing goal of defense-motivated individuals will, thus, be to find the protective action effective, because then these individuals can feel save. Even though the easiest way would be to accept the recommendation without scrutinizing the arguments contained in the message, it is not likely

Introduction

27

that heuristic processing strategies will prevail when serious personal health risks are involved (e.g., Chaiken et al., 1989). The processing will instead involve attempts to make the recommendation appear highly effective by means of a biased search for arguments supporting the effectiveness of the protective action, and through biased evaluation of these arguments. In other words, defense motivation will lead to a positive bias in the processing of the action recommendation, and consequently heighten the motivation to accept a solution to a particular threat, regardless of the quality of the arguments supporting this recommendation. This prediction of the stage model is contrary to predictions of most fear appeal theories that assume an interaction between perceptions of threat and efficacy (e.g., Hovland et al., 1953; Rogers, 1983; Witte, 1992). Most fear appeal theories propose that high perceptions of threat will only lead to persuasion if the recommended action is effective. The stage model, however, assumes that defense-motivated individuals will be motivated to perceive any recommendation as effective, as long as it is at least somewhat plausible, because engaging in a recommended action will make these individuals feel save. The stage model makes some more specific predictions with respect to the effectiveness of a fear-arousing communication. Whereas other fear appeal theories often hold the same predictions for all measures of persuasion, the stage model makes the important distinction between measures of persuasion that assess attitudes and those that measure behavioral intentions or behavior. Attitudes about a protective action are assumed to be based mainly on an evaluation of the provided information. Individuals are proposed to have more positive attitudes towards a protective action if the recommendation is supported by strong arguments rather than weak arguments, and, because of more elaboration, when a health risk is described as more serious. Yet, when it comes to measures of persuasion that have behavioral implications (i.e., intentions and behavior), the stage model assumes that individuals will only be motivated to engage in the protective action if they actually feel vulnerable. After all, why should one invest effort into avoiding a risk, if one does not feel personally at risk? The only direct test of the predictions of the stage model can be found in a series of experiments reported by Das, de Wit and Stroebe (2003). These studies assessed the effects of the severity of and vulnerability to a health risk on the processing and acceptance of an action recommendation, which was supported by high or low quality arguments. In these experiments, the fear appeal described the health consequences of stress as either severe or mild and the recommended protective action advocated a stress management training, which was either supported by strong or weak arguments. The assignment to different stress vulnerability conditions was based on self-

28

Chapter 1

measures of perceived vulnerability in Experiments 1 and 2, but was manipulated in Experiment 3. Consistent with predictions from the stage model, these experiments revealed a positive bias in the processing of the action recommendation under high vulnerability: Vulnerable respondents reported more positive thoughts about the action recommendation than non-vulnerable respondents did, regardless of the quality of arguments supporting the recommendation. Furthermore, vulnerability proved to be the only determinant of engaging in the recommended action: Vulnerable respondents had higher intentions and reported more behavior change than non-vulnerable respondents. However, findings were less consistent for predictions concerning attitudes towards the recommendation. In particular, the impact of argument quality on attitudes towards the recommended action varied across experiments, thus weakening the case for systematic processing. Attitudes were affected by both vulnerability and argument quality in Experiment 1 and 2, with respondents having more positive attitudes when they felt vulnerable than when they did not, and when the recommendation was supported by strong arguments rather than weak arguments. However, this pattern was not replicated in Experiment 3, in which argument quality failed to have any effect on attitudes, and an interaction-effect between vulnerability and severity was found. Vulnerable respondents were found to have more positive attitudes towards the recommendation, than non-vulnerable respondents, but only when the negative consequences of stress were depicted as severe. Further shortcomings of these earlier experiments are that: (1) only one study experimentally manipulated vulnerability; (2) the experiments only focused on the processing of the action recommendation, and no attention was given to the processing of the fear appeal; (3) no attempt was made to analyze additional factors that might influence the impact of fear-arousing communications on persuasion. Overview The present thesis will further examine the processing of fear-arousing communications from the perspective of the stage model of the processing of fear-arousing communications. Experiments will be reported that examine the processing of both the fear appeal and the recommended action, and the subsequent effects on persuasion. In these experiments severity of and vulnerability to different health risks will be manipulated/varied, as will be aspects of the recommended action (e.g., argument quality, source expertise, and response costs), and aspects outside the communication (e.g., personality factors). The aim of this thesis is to provide more insight into the conditions under which fear-arousing communications are (in)effective, and, more

Introduction

29

importantly, to specify how and why they are (in)effective. In Chapters 2 through 4, the processing of the action recommendation is examined, using an experimental design that manipulates respondents vulnerability to Repetitive Strain Injury (RSI) (i.e., mouse-arm) through falsefeedback on a bogus RSI-test, after which respondents are presented with a message depicting the consequences of RSI as severe. This is followed by a message describing an action recommendation (i.e., stress management training), which is supported by either strong or weak arguments. According to the stage model, exposure to the severe negative consequences of RSI should induce systematic processing in all respondents, and the quality of the arguments used in the recommendation should, therefore, affect attitudes towards and cognitive responses about this recommendation. In addition, making respondents feel vulnerable to these negative consequences of RSI should arouse defense motivation. Defensemotivated individuals are assumed to be motivated to accept an action recommendation. They should, therefore, have more positive thoughts about the recommendation, regardless of argument quality. Attitudes toward the recommendation are assumed to remain unbiased by defense motivation, and should be determined purely by argument quality. Regarding intentions to engage in the recommendation, and actual signing up for it, it is expected that only individuals who feel vulnerable should want to do so. The experiment reported in Chapter 2 additionally examines the role of the expertise of the source to which the recommendation is attributed in the processing and acceptance of fear-arousing communications. Source expertise can act as a heuristic cue (e.g., Eagly & Chaiken, 1993) and is manipulated to further examine depth of information processing by attributing the action recommendation either to a professor in medicine or to a housewife. Since the stage model assumes that with a severe health risk all individuals should process the content of the recommended action systematically, and not merely rely on heuristic cues to evaluate the efficacy of the recommendation, the source to which a recommendation is attributed should have little impact on attitudes. In the study reported in Chapter 3 the experiment presented in Chapter 2 is replicated and extended. In this study, the impact of the costs of engaging in the proposed protective action is examined by manipulating the amount of effort that is required (response costs). Because of the stage models assumption of a positive bias in the processing of action recommendations, the effort required to engage in the protective action should have less impact on intention and behavior when individuals feel vulnerable to the health risk. In Chapter 4, the focus moves away from the influence of components of a fear-arousing communication, towards the impact of individual differences of recipients on the effects of a fear-arousing communication. The experiment

30

Chapter 1

presented in Chapter 4 explores the role of two personality factors, dispositional optimism and neuroticism, on the processing and acceptance of fear-arousing communications. It is expected that these personality factors should have some effect on processing and acceptance of fear-arousing communications, but these effects are assumed to not overrule the expected effects of manipulated vulnerability and argument quality. Respondents high on neuroticism exposed to a fear-arousing communication are expected to react with more negative emotions, but because of their known avoidant or inactive ways of coping (Costa & McCrae, 1987), neuroticism should not have much of an effect on any of the other dependent variables, especially those that have behavioral implications. In contrast, optimistic respondents are expected to experience less negative emotions, and, because of their general positive outlook on life and known active and problem-focused coping (Scheier & Carver, 1985), optimists should have more positive reactions to the action recommendation, affecting cognitive responses, attitudes, as well as intentions and behavior. Chapter 5 will examine the processing of both the fear appeal and the action recommendation, attempting to shed more light on the processes that lead to acceptance of fear-arousing communications. Three experiments are reported that use a slightly different design to the one used in the experiments reported in the previous chapters, and focus on different health risks (e.g., alcohol consumption and hypoglycemia). In addition, in these experiments the severity of the consequences is manipulated (low/high), instead of holding high severity constant. The first experiment focuses on the processing of a fear appeal. This study varies vulnerability to alcohol consumption based on respondents selfmeasures of perceived vulnerability. Respondents read a message depicting the negative consequences of alcohol consumption as either mild or severe, and no protective action is specifically recommended in detail. In accordance with the assumptions of the stage model, a fear appeal should generate thoughts that reveal attempts to minimize the threat, which would indicate a negative processing bias. Because defense motivation is assumed to be aroused especially when respondents feel vulnerable and when the consequences of the health risk are depicted as severe, respondents who feel vulnerable and are exposed to the severe negative consequences of alcohol consumption should experience more negative emotions and have more minimizing thoughts than respondents in all other conditions. The second experiment replicates and extends Experiment 1. Instead of relying on respondents self-measures, respondents vulnerability to hypoglycemia (i.e., low blood glucose levels) is manipulated by bogus feedback on a hypoglycemia test. This is again followed by a message depicting the negative consequences of hypoglycemia as either mild or severe, and no

Introduction

31

protective action is specifically offered. In addition, Experiment 2 also includes as an extra factor a manipulation of message source, but in this experiment the focus is on source credibility (instead of source expertise as in Chapter 2), by attributing the severity message either to a scientific journal or to a popular women magazine. Source credibility is manipulated to provide an explicit test of the depth of processing expectations. When the likelihood of systematic processing is low, source credibility can operate as a heuristic cue and should, hence, affect the persuasive impact of a message mainly in heuristic processing (i.e., when both vulnerability and severity are low). Lastly, the third experiment focuses on both the processing of the fear appeal and the subsequent action recommendation. The experiment is similar to Experiment 2, only this time a protective action is specifically recommended, in the form of stress management training. This recommendation is supported by weak or strong arguments. In Chapter 6, a shift is made from original experiments to the body of previous empirical research on fear appeals, by looking at how the results of the experiments presented in this thesis relate to findings in earlier empirical studies on fear-arousing communications. An extensive meta-analysis of fear appeal experiments published to date will be reported in this chapter. In this meta-analysis the effects of both vulnerability and severity as well as other communication variables on the processing and acceptance of fear-arousing communications are examined, attempting to find more general support for the assumptions derived from the stage model. Finally, Chapter 7 summarizes and discusses the findings reported in the previous chapters, as well as dwells on their theoretical and practical implications. All empirical chapters in this thesis (Chapters 2 through 6) were written as separate papers that have been published or have been submitted for publication in scientific journals. Therefore, there is some similarity and repetition between chapters, especially in the introduction and method sections. Because these chapters consist of separate papers they can be read independently.

In the remainder of this chapter, the names specified by the heuristic-systematic model

will be used for the two modes of information processing. Thus, where the terms systematic and heuristic processing are used one could also read central route and peripheral route processing.

Chapter 2: The impact of fear appeals on processing and accepting action recommendations: the key role of vulnerability1

Fear appeals are still widely used in health education, even though the mechanisms mediating their impact are not well understood. The experiment presented in this article assesses predictions derived from a recently developed stage model of fear-arousing communications (Das, de Wit & Stroebe, 2003; Stroebe, 2000), a theoretical framework which builds on previous fear appeal theories (Leventhal, 1970; Rogers, 1983; Witte, 1992), cognitive stress theory (Lazarus & Folkman, 1984), as well as on dual process theories of attitude change (Chaiken, 1980; Chaiken, Liberman & Eagly, 1989; Petty & Cacioppo, 1986). To test our model we studied the impact of a single strong feararousing communication on the acceptance of an action recommendation. To examine some of the processes underlying changes in attitude, intention and behavior, we manipulated respondents vulnerability to a severe health risk, as well as two factors presumed to influence the acceptance of a recommended protective action, namely the expertise of the communicator and the strength of the arguments contained in the action recommendation. Theoretical approaches to the study of fear-arousing communications Most research on fear appeals has been conducted before the advent of dual process theories of attitude change. Initially, this research was theoretically guided by reinforcement theory (Hovland, Janis & Kelley, 1953). Later cognitive theories were applied, such as the parallel response model (Leventhal, 1970), protection motivation theory (Rogers, 1975, 1983), and more recently the extended parallel process model (Witte, 1992). These later theories abandoned the notion central to reinforcement theory that fear is a necessary antecedent of behavioral adaptation to danger. According to Leventhal (1970), appraisal of an environmental threat can give rise to two independent processes, namely danger control and fear control, with danger control involving the instrumental actions the individual performs in order to reduce the threat. Actions in the service of fear control, such as avoidance actions (including defenses) as well as attempts to control the emotional response (e.g., drinking alcohol) frequently have no effect on

This chapter has been published as: De Hoog, N., Stroebe, W., & de Wit, J.B.F. (2005).

The impact of fear appeals on processing and acceptance of action recommendations. Personality and Social Psychology Bulletin, 31(1), 24-33.

34

Chapter 2

danger. Whereas protection motivation theory focused exclusively on elaborating Leventhal's danger control process, Wittes extended parallel process model reemphasized the role of fear by elaborating conditions under which fear control is likely to affect danger control. However, even though research on fear appeals resulted in evidence that high fear messages produce more change than low fear messages (e.g., Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000), these empirical studies have not succeeded in completely clarifying the theoretical assumptions about underlying mechanisms (Eagly & Chaiken, 1993, pp. 442-446). With the advent of dual process theories of attitude change (e.g., Chaiken, 1980; Petty & Cacioppo, 1986), research began to focus on the processing of the arguments contained in a communication as well as on the outcome of persuasion. According to dual process theories, fear arousal can have two effects, namely (1) act as a motivator to induce recipients to engage in intensive and thoughtful message processing, and (2) induce defense motivation, which will lead to biased message processing (Gleicher & Petty, 1992; Liberman & Chaiken, 1992). In contrast to the unbiased or accuracy-motivated perceiver who assesses the validity of attitude-relevant information in the interest of achieving a valid position, the processing goal of defense-motivated recipients is to confirm the validity of a preferred position and disconfirm the validity of non-preferred positions. Thus, defense-motivated individuals will process information in ways that best support their own beliefs. This implies that information that is congruent with a particular preferred position will be judged as more valid than information that is incongruent with this position (Ditto & Lopez, 1992; Lord, Ross & Lepper, 1979; Pyszczynski & Greenberg, 1987). A stage model of processing of fear-arousing communications Integrating ideas from dual process theories of persuasion, previous fear appeal theories and cognitive stress theory (Lazarus & Folkman, 1984), we developed a stage model of processing of fear-arousing communications (Das et al., 2003; Stroebe, 2000). In line with cognitive stress theory, our model assumes that individuals exposed to fear-arousing communications engage in two types of appraisal, namely (1) appraisal of the threat, and (2) appraisal of coping strategies available for reducing or eliminating the threat. These two appraisal processes parallel the two components of fear-arousing communications, namely the fear appeal, which emphasizes the severity of the health risk and individual vulnerability, and the action recommendation, which provides information on how to avoid the health risk. Expanding upon the differentiation between appraisal of threat and appraisal of recommendations (e.g., Leventhal, 1970; Witte, 1992), our stage

the key role of vulnerability

35

model focuses on the appraisal processes at each of these stages. These appraisal processes are most visible if we examine and assess appraisal of the threat and appraisal of the recommended action both separately and sequentially. Based on dual process theories of attitude change (e.g., Chaiken et al., 1989), we assume that the appraisal of the magnitude of threat implied by the fear appeal determines both the processing mode (i.e., depth of processing) and the processing goal (i.e., accuracy or defense motivation; see Table 2.1).
Table 2.1. The impact of severity and vulnerability on processing mode and goal

Severity
Low Low
Shallow processing Accuracy motivation

High
Shallow processing Accuracy motivation Deep processing Defense motivation

Thus, if an individual is exposed to a severe health risk and feels vulnerable, this should seriously threaten the individuals self-definitional belief that he/she is healthy, and arouse defense motivation as well as the motivation for deep information processing. We assume that processing will be deep instead of shallow, because any communication depicting a serious personal health threat is likely to require a thorough evaluation (Chaiken et al., 1989). According to our stage model, defense motivation induced by high threat perceptions manifests itself not in avoidance reactions as previous fear appeal models have proposed (e.g., Witte, 1992), but in biased deep processing. The direction of the bias aroused by defense motivation will vary with the type of appraisal (i.e., negative bias in appraising threat; positive bias in appraising coping strategies). In appraising the health threat (Stage 1), individuals will attempt to minimize the threat by critically looking at the content of a fear appeal. They will engage in a biased search for inconsistencies and logical errors, and their evaluation of evidence will be biased in the direction of the preferred conclusion (Ditto & Lopez, 1992; Liberman & Chaiken, 1992; Sherman, Nelson & Steele, 2000). If this strategy proves unsuccessful because even biased processing is constrained by evidence and rules of inference (Kunda, 1987), individuals will have to accept that they are personally at risk. In this case, the processing of the recommendation (Stage 2) will be biased as well.

Vulnerability

High

Deep processing Accuracy motivation

36

Chapter 2

At this stage, the processing goal of defense-motivated individuals will be to find the protective action effective, because then they can feel safe. Although the easiest way to feel reassured would be to accept the recommendation without scrutinizing the arguments contained in the message, it is not likely that shallow processing strategies will prevail when serious health risks are involved. We, therefore, propose that vulnerability to a severe health risk motivates individuals to process the action recommendation in a manner that is both biased and deep. The processing will involve attempts to make the recommendation appear highly effective by means of a biased search for arguments supporting the effectiveness of the protective action, and through biased evaluation of these arguments. This type of processing is likely to result in increased persuasion regardless of the quality of the arguments supporting the recommendation. If an individual does not feel vulnerable, a health risk, no matter how severe, is unlikely to arouse defense motivation. Yet, even non-vulnerable individuals are likely to invest effort in processing the contents of a communication depicting a severe risk, because it may be useful to be informed about a serious health risk, even if the danger is not immanent. In this respect, our stage model differs from dual process theories that state that personal relevance is a precondition for deep information processing. The appraisal processes are less elaborate under conditions of low severity, because if a health risk is trivial and the individual does not feel at risk, he/she is unlikely to invest great effort in thinking about the arguments contained in the communication. Even if the individual does feel vulnerable to a trivial health risk, there is no reason to be defensive. For these reasons the current study focuses on reactions following exposure to a message depicting severe health consequences. An important distinction with respect to persuasion that is emphasized by our stage model is that between outcome measures that rely on attitudes and those that assess behavioral intentions or behavior (Fishbein & Ajzen, 1975). In deriving predictions about the impact of action recommendations one has to distinguish between the impact of fear appeals on attitude towards the recommendation and on intention to act on it. For individuals who do not feel vulnerable, a severe health risk will lead to a positive attitude towards a protective action, if the recommendation is supported by strong arguments and to a negative (or less positive) attitude if it is supported by weak arguments. Yet, individuals will only be motivated to engage in the protective action, if they feel vulnerable. After all, why should one invest effort into avoiding a health risk, if one does not feel personally at risk? How does our stage model differ from earlier theories, such as Leventhals parallel response model (1970), Wittes extended parallel process model

the key role of vulnerability

37

(1992) or Rogers protection motivation theory (1983)? We share with these theories the assumption that the emotional response to a threat is an independent (i.e., parallel) response to the threat and not a mediator of the instrumental action. Furthermore, we build on these theories with regard to key factors for threat perceptions (i.e., severity and susceptibility) and efficacy perceptions. However, our theoretical reasoning also differs from that of these models. Whereas Leventhal and Witte emphasize the distinction between two parallel modes of coping, namely problem-focused (danger control) versus emotion-focused coping (fear control) our stages refer to the distinction between primary (threat) and secondary (coping) appraisal. Furthermore, we relate these two types of appraisal to modes and motives of information processing, whereas none of the earlier theories make assumptions about information processing. Finally, neither Leventhal, Witte or Rogers make differential predictions regarding outcome measures. Empirical evidence Most of the studies, which assessed the impact of vulnerability on the processing of fear-arousing communications, have focused exclusively on the processing of fear appeals (Stage 1). There is now ample evidence that people are more critical of evidence, which is highly health-threatening, than of less threatening evidence (Ditto & Lopez, 1992; Janis & Terwilliger, 1962; Kunda, 1987; Liberman & Chaiken, 1992; Reed & Aspinwall, 1998; Sherman et al., 2000). Much less research has been conducted on the processing of action recommendations (Stage 2), and the few studies on this topic either failed to manipulate vulnerability experimentally (Jepson & Chaiken, 1990) or failed to include measures of cognitive processing (Gleicher & Petty, 1992). The only direct tests of the predictions from our stage model are three experiments by Das et al. (2003). These studies assessed the effect of the severity of, and vulnerability to a threat on the processing and acceptance of an action recommendation, which was supported by high or low quality arguments. Consistent with predictions, these studies revealed a positive bias in the processing of the action recommendation under high vulnerability: Vulnerable respondents reported more positive thoughts about the recommendation than non-vulnerable respondents. Furthermore, vulnerability proved to be the only determinant of individuals intention to engage in the recommended action. Findings were less consistent for attitudes. In particular, impact of argument quality on attitudes varied across experiments, thus weakening the case for deep processing. A further shortcoming of this research program was that only one study experimentally manipulated vulnerability. Therefore, the present study was conducted in order to clarify inconsistencies and provide further support for the assumptions of our stage model.

38

Chapter 2

Design overview Extending the research by Das and colleagues (2003), the present study focused on the processing and acceptance of the action recommendation (Stage 2). We used the health consequences of RSI (Repetitive Strain Injury; mouse arm) for the fear appeal. Respondents vulnerability to RSI was manipulated, as was argument quality in the action recommendation. To allow a further test of whether respondents processing of the message was deep or shallow, we also manipulated a second factor that traditionally has been used in dual-process research to identify depth of processing, namely source expertise. Source expertise operates as a heuristic cue and should affect the impact of a message mainly for shallow processing. Severity of the health consequences was not manipulated but kept constant and high in all conditions. We expected that a severe health risk would induce deep processing and that argument quality would therefore affect attitudes and cognitive responses. Since we expected that with a severe health risk all individuals would process the content of the recommendation and not rely on heuristic cues, our source expertise manipulation should have little impact on attitudes. We further expected that inducing vulnerability to a severe health risk would arouse defense motivation. Since individuals who feel vulnerable to a serious health risk should be motivated to accept the action recommendation, they should overestimate the effectiveness of the recommendation and have more positive thoughts about the action, regardless of argument quality. We therefore predicted a main effect of vulnerability on cognitive responses and an interaction of argument quality and vulnerability on attitude, with argument quality having a stronger impact on low rather than high vulnerable individuals. With regard to respondents intentions and with regard to their actions, we predicted only a main effect of vulnerability. Since individuals who do not feel vulnerable, should not seek to engage in the recommended action, however effective it might be there should be no effect of argument quality on intentions of low vulnerability individuals. Under high vulnerability, respondents might be more willing to adopt a recommendation supported by strong rather than weak arguments, but their defense motivation is likely to mitigate against such an effect. The fact that Das et al. (2003) found only a vulnerability main effect on intention in all three studies supports this assumption. Experiment 2.1 Method Participants and design The experiment was based on data from 124 (39 male, 85 female) students of Utrecht University, who participated in the experiment for payment of 4. Four

the key role of vulnerability

39

respondents who expressed suspicion of the vulnerability manipulation, and two respondents who were already suffering from RSI were not included in this analysis. Respondents were run individually and were randomly assigned in equal numbers to the conditions of a 2(vulnerability: high or low) x 2(argument quality: strong or weak) x 2(source expertise: high or low) between-subjects factorial design. Independent variables Vulnerability. Respondents were given false feedback after completing a bogus RSI test. Half the respondents were told they were very vulnerable to RSI (Very vulnerable: The probability of you developing RSI is very high) and the other half were told they were not very vulnerable to RSI (Not very vulnerable: The probability of you developing RSI is very low). Argument quality. The quality of the arguments supporting the action recommendation was manipulated by presenting a recommendation, which contained either six strong or six weak supporting arguments. Following the procedure described by Petty and Cacioppo (1986), these arguments were selected from a pool of arguments that had been rated as weak or strong in a pilot study. Examples of strong supporting arguments are Research has shown that people who participated in stress management training developed RSI less often than people who did not participate, Stress management training makes you less vulnerable to RSI related complaints, and After participating in the training 90% of the participants were able to cope better with high work pressures and developed more efficient ways of working. Weak supporting arguments included Stress management training is a lot of fun and good for social contacts, If the training had not been effective it would never have been offered by the institute that developed the training, and If you participate in stress management training at least you do not have to blame yourself when you do develop RSI. Source expertise. This was manipulated by attributing the action recommendation either to a professor of medicine (high expertise) or a housewife (low expertise). Procedure The experiment was run on personal computers. After a brief introduction, respondents received a communication that manipulated one important aspect of the fear appeal, namely their vulnerability to the health risk. They were told there was a new test that could measure vulnerability to RSI. This test required them to shift a computer mouse repeatedly following various patterns displayed on the screen. A computer program would then assess their vulnerability to RSI from various parameters of their movements (e.g.,

40

Chapter 2

pressure, speed). Respondents then received false feedback in the form of a thermometer appearing on their computer screens in which the red virtual mercury rose either almost all the way to the top of the thermometer, indicating high vulnerability or it stayed low on the thermometer, indicating low vulnerability. Next, a message about the health consequences of RSI was presented. This message was a 300-word text in which the consequences of RSI were described in detail. It was said that RSI is a progressive condition with very serious and long lasting physical consequences. The severity of these consequences was high for all respondents (e.g., constant pain in and numbness and swelling of limbs). Subsequently, respondents received a communication, which manipulated argument quality and source of the action recommendation. Respondents read a fictitious letter (that was said to have recently been printed in a newspaper) promoting stress management training to reduce vulnerability to RSI, which contained either six strong or six weak supporting arguments. This letter was attributed to either a high or a low expertise source. Respondents attitude, intention, and behavior towards the recommendation were then assessed, as were perceived threat, fear, and cognitive responses (i.e., the thoughts they had about the letter promoting stress management training). Before leaving, respondents were thoroughly debriefed. Manipulation checks Vulnerability was assessed by three items on 7-point rating scales [how vulnerable respondents perceived themselves to be, 1= not vulnerable; 7= very vulnerable, how high they thought their risk for RSI was, 1= very low; 7= very high, and the probability of them experiencing negative health consequences due to RSI, 1= very low; 7= very high (= .85)]. Argument quality was assessed by asking respondents to indicate on 7-point rating scales ranging from 1(= not at all) to 7(= very) how strong, persuasive and meaningful they found the arguments used in the recommendation (= .89). Source expertise was assessed by two items [How much of an expert do you think the author of the letter is? and How accomplished do you think the author of the letter is? ((= .85)], again with 7-point rating scales ranging from 1(= not at all) to 7(= very). Dependent variables Measures of persuasion. Attitude towards stress management training was measured by having respondents indicate on five semantic differential items how useful, good, important, interesting, and sound they found the training (= .84). Intention to participate was measured by three items with 7-point rating scales ranging from 1(= certainly not) to 7(= certainly) [Do you have

the key role of vulnerability

41

the intention to, do you want to and are you going to participate in stress management training? (= .83)]. Two types of behavior were measured with one item each and yes/no response options. Respondents could ask for more information about stress management training and could sign up for it. It was pointed out, however, that these commitments were not binding. Perceived threat. Cognitive perceptions of threat were assessed with two items on 7-point rating scales [How threatening are the described health consequences of RSI?, 1= not threatening; 7= very threatening; and How disturbing are the described health consequences of RSI?, 1= not disturbing; 7= very disturbing (= .91)]. Fear. Fear was assessed by asking respondents how fearful they felt after reading the health consequences of RSI with a single item (When thinking of the described health consequences of RSI, how fearful do you feel?, 1= not fearful; 7= very fearful). Cognitive responses. To measure the cognitive processing of the recommendation respondents were asked to write down all their thoughts concerning the recommendation. Results Preliminary analyses showed that there were no gender effects on any of the dependent variables. Therefore, three-way analyses of variance (ANOVAs) that included vulnerability, argument quality and source credibility manipulations were executed for all dependent measures, unless stated otherwise. Manipulation checks Vulnerability. A test of the manipulation of vulnerability revealed a main effect of vulnerability, F(1,116)= 96.90, p < .001 (2 = .46). Respondents perceived their vulnerability as higher in the high vulnerability condition (M = 4.54) than in the low vulnerability condition (M = 2.54). No other effects were found, all F < 1. Argument quality. The manipulation of argument quality was effective. Respondents, who received strong arguments, perceived the arguments as stronger (M = 4.67) than did those who received weak arguments (M = 3.52), F(1,116)= 28.43, p < .001 (2 = .20). No other effects were found, all F < 1. Source expertise. The ANOVA revealed a main effect of source, F(1,116)= 16.30, p < .001 (2 = .12). Respondents perceived the source as more of an expert in the high expertise condition (M = 4.60) than in the low expertise condition (M = 3.60). There were also main effects of argument quality (F(1,116)= 7.70, p < .01; (2 = .06) and vulnerability (F(1,116)= 5.94, p < .05; (2 = .05). Respondents perceived the source as more of an expert in the

42

Chapter 2

strong argument condition (M = 4.43) than in the weak argument condition (M = 3.74). They also perceived the source as more of an expert when their vulnerability was high (M = 4.39) rather than low (M = 3.78), which suggests biased processing. Dependent variables Attitude. There was a main effect of argument quality, F(1,116)= 9.14, p < .01 (2 = .08). Respondents had more positive attitudes towards the training when it was supported by strong (M = 5.05) rather than by weak arguments (M = 4.50). The main effect of source expertise did not approach significance, F(1,116)= 1.24, p = .27, neither did the main effect of vulnerability, F(1,116)= 1.46, p = .23. Interaction effects were also not significant, all F < 1 (Table 2.2).
Table 2.2. Means (SDs) of attitude and intention toward the recommendation by vulnerability, argument quality and source expertise Low vulnerability Attitude Low source expertise Weak arguments Strong arguments High source expertise Weak arguments Strong arguments Intention Low source expertise Weak arguments Strong arguments High source expertise Weak arguments Strong arguments 3.18 (1.24) 2.27 (.97) 3.18 (1.38) 3.96 (1.41) 2.52 (1.51) 2.56 (1.01) 3.60 (1.24) 3.42 (1.67) 4.57 (1.06) 5.04 (.76) 4.47 (1.16) 5.42 (.82) 4.34 (1.31) 4.72 (.69) 4.61 (.88) 5.03 (1.08) High vulnerability

Intention. There was a vulnerability main effect, F(1,116)= 14.53, p < .001 (2 = .11). Respondents had greater intention to participate in the training when their vulnerability was high (M = 3.54) than when their vulnerability was low (M = 2.63). The interaction effect of argument quality and vulnerability did not reach significance, F(1,116)= 2.36, p = .13, neither did any other main or interaction effect, all F < 1 (Table 2.2). Behavior. Logistic regression analysis was used to measure impact of the independent variables on the two measures of behavior. This analysis showed a main effect of vulnerability on requests for information (Wald(1)= 8.28, p <

the key role of vulnerability

43

.01). Vulnerable respondents more often requested information about the recommendation (69.4%) than non-vulnerable respondents (43.5%). No other effects were found. The same pattern emerged for actual signing up (Wald(1)= 12.98, p < .001). Vulnerable respondents signed up more often (37.1%) than non-vulnerable respondents (8.1%). No other effects were found. Perceived threat. A main effect of vulnerability was found, F(1,116)= 7.26, p < .01 (2 = .06). Vulnerable respondents found the text containing the health consequences of RSI more threatening (M = 5.10) than non-vulnerable respondents (M = 4.50). No other effects were found, F < 1. Fear. Vulnerable respondents were also more fearful about the health consequences of RSI (M = 5.02) than were non-vulnerable respondents (M = 4.16), F(1,116)= 11.88, p < .001 (2 = .09). No other effects were found. Cognitive responses. Two independent raters scored the number of thoughts about the recommendation in each of two categories: positive thoughts (Range: 0-3; Kappa= .89) and negative thoughts about stress management training (Range: 0-3; Kappa= .92). Examples of positive thoughts are: I think stress management training is very interesting, and Its a great initiative, more people should participate in it. Examples of negative thoughts are: Im not participating just so I dont have to blame myself if I do get RSI, thats ridiculous, and I dont think stress management training is a very effective way to prevent RSI. Both positive/total thoughts and negative/total thoughts resulted in positive and negative thought ratios (Range: 0-1). Respondents reported an average of 2.42 (SD = 1.05) thoughts in total (Range: 1-5). A 3-factor ANOVA on total number of thoughts about the recommendation did not result in any significant effects. Analysis of positive and negative thoughts separately resulted in main effects of vulnerability on positive thoughts (F(1,116)= 4.47, p < .05; 2 = .04) and on negative thoughts (F(1,116)= 6.811, p < .01; (2 = .06). Vulnerable respondents reported more positive thoughts about the recommendation (M = .59) than non-vulnerable respondents (M = .41). In addition, vulnerable respondents reported less negative thoughts about the recommendation (M = .10) than non-vulnerable respondents (M = .27). On negative thoughts there also was a main effect of argument quality, F(1,116)= 8.98, p < .01 (2 = .07). Respondents had more negative thoughts about the action recommendation when the recommendation was supported by weak (M = .28) rather than strong arguments (M = .09). No further effects were found. Mediation Hierarchical regression-analyses were performed to assess a possible mediation of the effects of argument quality and vulnerability on attitude, intention, and behavior, respectively, by any of the other dependent variables in this study.

44

Chapter 2

Following the procedures specified by Baron and Kenny (1986), variables were only entered as possible mediators if three conditions were met: (1) the independent variable affected the mediator, (2) the independent variable affected the dependent variable, (3) the mediator affected the dependent variable. Which variables met conditions 1 and 2 can be derived from the analysis of variance described above, which variables also met condition 3 can be found in the correlation table (Table 2.3).
Table 2.3. Correlations between all dependent variables Variable 1. Perceived threat 2. Fear 3. Positive thoughts 4. Negative thoughts 5. Attitude 6. Intention 7. Request for information 8. Subscription
* p < .05. ** p < .01.

1 --.53** .20* -.07 .32** .34** .17 .15

2 --.17 -.21* .22* .26** .12 .14

---.27** .20* .30** .15 .17 ---.49** -.17 -.17 -.15 --.43** .17 .16 --.51*** .54*** --.47**

With respect to the effect of argument quality on attitude, only negative thoughts about the recommendation qualified as a mediator. Hierarchical regression-analysis showed that adding negative thoughts to the model, regressing attitude from argument quality, reduced the effect of argument quality on attitude to non-significance (see Table 2.4). A Sobel test (Baron & Kenny, 1986) of mediation was significant (Z = 2.72; p < .01), indicating that negative thoughts mediated the effect of argument quality on attitude. With respect to the effect of vulnerability on intention, positive thoughts about the recommendation, perceived threat and fear qualified as mediators. When fear was added to the model, regressing intention from vulnerability, the effect of fear on intention appeared non-significant ( = .16, ns), and, therefore, fear did not operate as a mediator. Separately adding perceived threat and positive thoughts to the regression equation substantially reduced the effect of vulnerability on intention (Table 2.4). Tests of mediation were significant for both perceived threat (Z = 2.17; p < .05) and positive thoughts (Z = 2.18; p < .05), indicating that the effect of vulnerability on intention was partially mediated by perceived threat and positive thoughts. For the effect of vulnerability on behavior, only intention qualified as a possible mediator. Entering vulnerability and intention into separate

the key role of vulnerability

45

hierarchical regressions with the two measures of behavior as criterion, substantially reduced the vulnerability effects on both behavior measures (Table 2.4). Tests of mediation were significant both for requesting information (Z = 3.26; p < .001) and for signing up for the stress management training (Z = 3.30; p < .001), showing that the effect of vulnerability on behavior was mediated by intention.
Table 2.4. Hierarchical regression-analyses predicting attitudes, intentions, and behavior Dependent variable: attitude Step 1 Argument quality Step 2 Argument quality Negative thoughts Dependent variable: Intention Step 1 Vulnerability Step 2a Vulnerability Threat Step 2b Vulnerability Positive thoughts Step 3 Vulnerability Threat Positive thoughts Dependent variable: behavior Step 1 Vulnerability Step 2 Vulnerability Intention
* p < .05. ** p < .01. *** p < .001.

.27** .15 ns .47*** .33*** .26** .28** .28** .24** .23* .24** .19* Information R .11***

R .07**

.28*** R .11***

.18***

.16***

.22*** Subscription .35** .19* .26*** .48*** .31*** R .07***

.28** .13 ns .47***

Discussion In general, the pattern of findings observed in this study is consistent with the assumptions derived from our stage model. As predicted, respondents held a more favorable attitude towards the training when the recommendation was

46

Chapter 2

supported by strong rather than weak arguments. When supported by strong arguments, respondents produced fewer negative thoughts about the training, and this difference in negative thoughts mediated the effect of argument quality on attitude. This pattern supports our assumption that under high severity conditions information in the action recommendation is processed deeply rather than shallow, that is to say, the contents of the information is processed rather than relying on shallow heuristics. The failure of source expertise to affect attitude towards stress management training, even though the manipulation check clearly indicated that the source manipulation had been successful, is also consistent with this assumption. The one finding which is inconsistent with our theoretical predictions is the failure of vulnerability to modify the main effect of argument quality on attitude. Based on the assumption that vulnerability to a serious health risk would induce defense motivation, we had assumed that vulnerable individuals would evaluate stress management training more positively than would individuals who did not feel vulnerable. The pattern of findings that Das et al. (2003) observed in their first two experiments supported this assumption. However, in the one study in which they manipulated vulnerability, vulnerability also failed to have any effect on attitude. We therefore suspect that measured vulnerability may have had a spurious relation with the attitude measure. The failure of our strong vulnerability manipulation to influence respondents attitudes towards the recommendation suggests that respondents were able to disregard their vulnerability to the health risk in evaluating the action recommendation. The mediation analysis, which indicated that the impact of argument quality on attitude was mediated by negative thoughts about the recommendation and that positive thoughts about the recommendation played no part in mediating the argument quality effect on attitude, provides further support for the assumption that respondents were unbiased (i.e., not defensive) in their evaluation of the recommended action. When we move from attitudes to intentions and behavior, vulnerability becomes the important determinant. In line with our predictions, respondents formed a stronger intention to attend the training when they felt vulnerable to RSI than when they did not. As in the studies of Das and colleagues (2003), this main effect of vulnerability was not moderated by argument quality. The fact that the impact of vulnerability was not tempered by argument quality in the high vulnerability condition is suggestive of defense motivation. Furthermore, if intentions had been formed on a totally informational basis, respondents should have weighed the threat of the health risk against the effectiveness of the recommendation. However, even though perceived threat partly mediated the effect of vulnerability on intention, perceived effectiveness, as manipulated by argument quality, did not. Finally, vulnerable respondents

the key role of vulnerability

47

reacted with more positive and fewer negative thoughts to the training, regardless of argument quality, and their positive thoughts were a significant mediator of the impact of vulnerability on intentions. Taken together, this pattern is consistent with the assumption that the impact of vulnerability on intention is a result of defense motivation. It is interesting to note that, in line with our theory as well as the parallel response model (Leventhal, 1970) and protection motivation theory (Rogers, 1983), fear did not play a role in mediating the impact of vulnerability on intention. Even though individuals, who were made to feel vulnerable perceived the health risk as more threatening and experienced more fear, it was the threat perception rather than the affect of fear, which appeared to have motivated them to form the intention to engage in the recommendation. Fear did not show a significant association with intention, once vulnerability was controlled for. Since our threat measure consisted of two items, whereas our fear measure only of one, the lower reliability of the fear measure could be an alternative explanation for the failure of fear to mediate the effect of vulnerability on intention. However, the fact that the size of the vulnerability effect on fear was descriptively greater than the effect of vulnerability on threat tends to argue against this interpretation. Finally, vulnerability was also the main determinant of the behaviors assessed in this experiment. Vulnerable respondents were not only more likely to request further information about the training, they were also much more likely to sign up for it than respondents who did not feel vulnerable. These findings with regard to the impact of vulnerability on intention and behavior are also consistent with the results reported by Das and colleagues (2003). In all three studies, vulnerability was the only determinant of intention and behavior, with the vulnerability effect on behavior being mediated by intention. There are thus two lines of influence, namely one from argument quality via negative thoughts to attitude, and another from vulnerability via perceived threat and positive thoughts to intention and behavior. The impact of defense motivation elicited by the vulnerability manipulation appears to have been restricted to measures that have behavioral implications: When asked whether they intended or wanted to perform the protective action, vulnerable respondents appear to have based their decisions not (only) on the presented information about the effectiveness of the recommendation, but they were mainly influenced by concerns about their own vulnerability. Respondents who felt vulnerable, perceived the health consequences as more threatening, and this partially mediated the impact of vulnerability on intention. Whereas this could still be considered purely the result of informational effect, the fact that individuals who felt vulnerable to RSI also produced more positive thoughts about the recommendation, regardless of argument quality and that these

48

Chapter 2

positive thoughts partially mediated the effect of vulnerability on intention suggests a defensive process: these individuals adjusted their beliefs about the effectiveness of the protective action and were trying to persuade themselves that they should engage in the recommendation to eliminate their risk. This raises the interesting question of whether threatened individuals would adopt any protective action, however ineffective. The fact that intention correlated significantly with attitude towards the recommendation indicates that there are clear limits to this kind of self-deception and that the perceived effectiveness of an action plays an important role in the decision whether or not to accept the health recommendation, as earlier studies have shown (e.g., Maddux & Rogers, 1983; Rogers & Mewborn, 1976). Only when individuals have decided that the action recommendation could provide some protection against the health threat, do they persuade themselves by selectively emphasizing the positive aspects of the recommended action. There are a number of studies, in particular studies performed before 1970, the results of which are quite opposite to those we report. They either found that higher fear messages resulted in less persuasion (Goldstein, 1959; Janis & Feshbach, 1953; Janis & Terwilliger, 1962; Leventhal, Singer & Jones, 1965) or that non-vulnerable respondents were more persuaded than vulnerable respondents (e.g., Berkowitz & Cottingham, 1960). However, there are several methodological problems with these older studies that might explain these seemingly conflicting results, namely 1) most of these studies did not differentiate between vulnerability and severity within their manipulation of fear level; 2) they assumed that people belonged to a vulnerable group instead of assessing whether or not they felt vulnerable; 3) they used preexisting differences in vulnerability instead of manipulating it; 4) they did not provide detailed or clear recommendations on how to change behavior and 5) their outcome measures were not always very consistent or carefully operated. The coherent pattern of findings, which emerged from the present study as well as from the earlier research by Das et al. (2003), has important practical implications. Whereas the emphasis of health education campaigns has frequently been on depicting the severity of health consequences, as well as on stressing the response efficacy of the recommended action, we have found that, although these factors affected attitudes, they failed to have much of an impact on intention and behavior. Intention and behavior were solely determined by vulnerability. This suggests that, however severe a health risk, and however effective the protection offered by the recommendation, unless we can persuade individuals that they are vulnerable to the health risk, they are unlikely to take protective action.

Chapter 3: The impact of fear appeals on processing and accepting action recommendations: the role of response costs2

Fear-arousing communications are widely used in health education campaigns. It is assumed that the more one succeeds in making individuals concerned about the consequences of their health impairing behavior, the higher will be the likelihood that they accept the recommended action. However, even though research on fear appeals resulted in evidence that high fear messages produce more behavior change than low fear messages (Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000), these empirical studies have not succeeded in fully elucidating the processes that mediate the persuasive impact of fear appeals. Furthermore, little is known about the influence of characteristics of the protective action offered on persuading individuals to change their health impairing behavior. The experiment presented in this article assesses how individuals process fear-arousing communications and explores the role of different aspects of the action recommendation. We test predictions derived from the stage model of processing of fear-arousing communications (Das, de Wit & Stroebe, 2003; De Hoog, Stroebe & de Wit, 2005a; Stroebe, 2000), a theoretical framework that builds on previous fear appeal theories (Leventhal, 1970; Rogers, 1983; Witte, 1992), as well as on dual-process theories of attitude change (Chaiken, 1980; Petty & Cacioppo, 1986). In this experiment we will demonstrate the impact of the manipulation of vulnerability to a health risk on the processing and acceptance of a recommended protective action, of which two characteristics are varied: the quality of the arguments supporting the recommendation as well as the required effort to perform the recommendation (i.e., response costs). Theoretical approaches to fear-arousing communications Initially, research on fear-arousing communications was theoretically guided by reinforcement theory (Hovland, Janis & Kelley, 1953). Later, more cognitive theories were applied, such as the parallel response model (Leventhal, 1970), protection motivation theory (Rogers, 1983), and the extended parallel process model (Witte, 1992). Even though most of these more recent models assume that cognitive processes mediate persuasion, they make no predictions

This chapter is based on: De Hoog, N., Stroebe, W., & de Wit, J.B.F. Fear appeals,

processing of action recommendations and persuasion: the role of response costs. Manuscript submitted for publication.

52

Chapter 3

concerning information processing, and measures of information processing (e.g., cognitive responses) have been virtually absent in research on feararousing communications. This situation changed with the advent of dualprocess theories of attitude change (Chaiken, 1980; Petty & Cacioppo, 1986), which soon were used to account for the impact of fear-arousing communications on persuasion (e.g., Gleicher & Petty, 1992; Liberman & Chaiken, 1992). According to dual-process theories, fear arousal can have two effects, namely act as a motivator to induce intensive and thoughtful message processing, and induce defense motivation, which will lead to biased message processing. Whereas the unbiased or accuracy-motivated perceiver assesses the validity of attitude-relevant information in the interest of achieving a valid position, the processing goal of defense-motivated recipients is to confirm the validity of a preferred position and disconfirm the validity of non-preferred positions. Thus, defense-motivated individuals will process and perceive information in ways that best support their own beliefs. This implies that information that is congruent with a particular preferred position will be judged as more valid than information that is incongruent with this position (Ditto & Lopez, 1992; Lord, Ross & Lepper, 1979; Pyszczynski & Greenberg, 1987). The stage model of processing of fear-arousing communications The stage model of processing of fear-arousing communications (Das et al., 2003; De Hoog et al., 2005a; Stroebe, 2000) integrates ideas from dualprocess theories (e.g., Chaiken, 1980) with those of theories of fear-arousing communications (Leventhal, 1970; Rogers, 1983; Witte, 1992). In line with most fear appeal theories, our model assumes that individuals exposed to feararousing communications engage in two types of appraisal, namely appraisal of the threat, and appraisal of coping strategies available for reducing the threat. Expanding upon this differentiation, our stage model makes differential predictions for the appraisal processes at each of these two stages. Determinants of defense motivation. According to our stage model, if individuals are exposed to information about a severe health risk and if they feel vulnerable to this health risk, this appraisal will seriously threaten their self-definitional belief that they are healthy, and arouse defense motivation as well as the motivation to carefully scrutinize the information given. We assume that processing will be systematic, because any communication describing a serious personal health threat is likely to require a thorough evaluation (Chaiken, Liberman & Eagly, 1989). According to our model, defense motivation manifests itself not in avoidance reactions as previous fear appeal models have proposed (e.g., Witte, 1992), but in biased systematic processing. The direction of the bias aroused by defense motivation will vary with the type

the role of response costs

53

of appraisal. Direction of defense-motivated bias. In appraising the health threat (Stage 1), defense-motivated individuals will attempt to minimize the threat by critically looking at the content of a fear appeal. They will engage in a biased search for inconsistencies and logical errors, and their evaluation of evidence will be biased in the direction of the preferred conclusion (Ditto & Lopez, 1992; Liberman & Chaiken, 1992; Sherman, Nelson & Steele, 2000). If this strategy proves unsuccessful, because even biased processing is constrained by evidence and rules of inference (Kunda, 1987), individuals will have to accept that they are personally at risk. In this case, the subsequent processing of the action recommendation (Stage 2) will be biased as well, but in the opposite direction. Under defense motivation, an action recommendation is proposed to be maximized. Because an action recommendation contains a possible solution to the health risk, it can moderate the negative emotions and cognitions evoked by the threat appraisal. At this stage, the processing goal of defense-motivated individuals will be to find the protective action effective, because then they can feel save. The processing will involve a biased search for, and evaluation of, aspects of the action recommendation supporting the effectiveness of the protective action. The positive bias in the processing of the action recommendation will heighten the motivation to accept and engage in a protective action offering a solution to a particular threat. Defense-motivated individuals will, therefore, disregard the quality of the arguments supporting the recommendation or the costs involved with engaging in it, as long as they have at least have some conviction that the recommended action offers some protection against the threat. Determinants of accuracy motivation. According to our model, defense motivation will only be aroused, if individuals feel vulnerable and if the health risk is at least moderately serious. If the health risk is minor, there is no reason to be defensive, even though the feeling of vulnerability should arouse sufficient motivation to invest effort into systematically processing the contents of the communication. Similarly, if individuals do not feel vulnerable, a health risk, no matter how severe, is equally unlikely to arouse defense motivation. Yet, even non-vulnerable individuals are likely to invest effort in processing a communication describing a severe health risk, because it is useful to be informed, even if the danger is not immanent. In this respect, our model differs from dual-process theories that stress that personal relevance is a precondition for systematic processing. Of course, if a health risk is trivial and individuals do not feel vulnerable, they are unlikely to invest much effort in thinking about the contents of the communication and they might rely on heuristic processing modes.

54

Chapter 3

Evaluation versus action. An important distinction with respect to persuasion emphasized by our model is that between outcome measures that rely on attitudes and those that assess behavioral intentions or behavior. Attitudes about a protective action are based mainly on evaluation of the provided information: Individuals will have more positive attitudes toward a protective action if the recommendation is supported by strong arguments rather than weak arguments. Yet, individuals will only be motivated to engage in the protective action if they feel vulnerable. After all, why should one invest effort into avoiding a health risk, if one does not feel personally at risk? Furthermore, because feeling vulnerable arouses defense-motivated positive processing of the action recommendation, vulnerable individuals will be motivated to engage in a protective action even when this action is supported by weak arguments or when there as some costs involved in implementing the protective action. Empirical evidence Studies conducted by Das et al. (2003) and De Hoog et al. (2005a) provide direct tests of some of the predictions derived from our model. These studies assessed the effects of vulnerability to and severity of a health risk on the processing and acceptance of an action recommendation, which was supported by high or low quality arguments. Consistent with predictions, these studies revealed a positive bias in the processing of the action recommendation under high vulnerability: Vulnerable respondents reported more positive thoughts about the recommendation than non-vulnerable respondents, regardless of argument quality. Furthermore, whereas attitudes were mainly determined by argument quality, vulnerability proved to be the only determinant of intention to engage in the recommendation. Thus, attitudes toward the recommendation remained unbiased by vulnerability and were only determined by argument quality. Yet, vulnerable individuals were willing to engage in the recommendation regardless of the quality of the arguments supporting it. We have previously established that argument quality of the recommendation has no impact on intention, although it affects attitudes, but little is known about the impact of other characteristics of the action recommendation, particularly the impact of response costs on defensemotivated respondents willingness to accept a recommended action. Therefore, the present experiment was designed to examine whether vulnerable individuals were willing to engage in a recommendation regardless of cost. Design overview This experiment manipulated vulnerability to a severe health risk, the quality of the arguments supporting a recommended action and the costs of that action

the role of response costs

55

in a 2 x 2 x 2 factorial design. Respondents vulnerability to Repetitive Strain Injury (RSI) was manipulated, after which they read a message depicting the potential severe health consequences of RSI. Subsequently, an action recommendation (stress management training) was offered. This recommendation was supported by either low or high quality arguments. In addition, the amount of effort required to implement the action recommendation (i.e., response cost) was manipulated. We expected that a severe health risk would induce systematic processing and that argument quality would therefore affect attitudes and cognitive responses. We further expected that inducing vulnerability to the health risk would arouse defense motivation. Individuals, who feel vulnerable to a serious health risk, should be motivated to accept the action recommendation. They should, therefore, overestimate the effectiveness of the recommended action and have more positive thoughts about the action, regardless of argument quality or response costs. In line with previous findings, we expected attitude to remain unbiased by defense motivation and determined purely by argument quality. For intention to engage in stress management training, however, we expected a main effect of vulnerability: Only individuals who feel vulnerable should want to change their behavior. Finally, regarding the impact of response costs on intentions and behavior, we expected that the effort required to engage in the protective action should have less impact on intention and behavior, the more individuals feel vulnerable to the health risk (i.e., interaction of response costs and vulnerability on intention and behavior). Because vulnerable individuals are proposed to be defense-motivated to engage in a protective action, regardless of the costs involved. Experiment 3.1 Method Participants and design Respondents were 120 (32 male, 88 female) students of Utrecht University, who participated in the experiment for payment of 4. Three respondents who expressed suspicion of the vulnerability manipulation were not included in this analysis. Respondents were randomly assigned in equal numbers to the conditions of a 2(vulnerability) x 2(argument quality) x 2(response costs) between-subjects factorial design. Independent variables Vulnerability. Respondents were given false feedback after completing a bogus RSI test. Half the respondents were told they were very vulnerable to

56

Chapter 3

RSI (Very vulnerable: The probability of you developing RSI is very high) and the other half were told they were not very vulnerable to RSI (Not very vulnerable: The probability of you developing RSI is very low). Argument quality. The quality of the arguments supporting the action recommendation was manipulated by presenting a recommendation, which contained either six strong or six weak supporting arguments. Following the procedure described by Petty and Cacioppo (1986), these arguments were selected from a pool of arguments that had been rated as weak or strong in a pilot study. Examples of strong supporting arguments are: Research has shown that people who participate in stress management training develop RSI less often than people who do not participate, Stress management training makes you less vulnerable to RSI related complaints, and After participating in the training 90% of the participants were able to cope better with high work pressures and developed more efficient ways of working. Weak supporting arguments included Stress management training is a lot of fun and good for social contacts, If the training had not been effective it would never have been offered by the institute that developed the training, and If you participate in stress management training at least you do not have to blame yourself when you do develop RSI. Response costs. The effort required to engage in the recommended action was manipulated by stating that the stress management training was given in the city where respondents attended university or in a city approximately 35 miles away. The conditions were rated as low and high in required effort in a pilot study. Procedure The experiment was run on personal computers. After a brief introduction, respondents received a communication that manipulated their vulnerability to the health risk. They were told about a new test that could measure vulnerability to RSI. This test required them to shift a computer mouse repeatedly following various patterns displayed on the screen. A computer program would then assess their vulnerability to RSI from various parameters of their movements (i.e., pressure, speed). Respondents received false feedback in the form of a thermometer appearing on their computer screens in which the red virtual mercury rose either almost all the way to the top of the thermometer, indicating high vulnerability, or it stayed low on the thermometer, indicating low vulnerability. Next, a message about the health consequences of RSI was presented. This message was a 300-word text in which the consequences of RSI were described in detail. It was stated that RSI is a progressive condition with very serious and long lasting physical consequences. The severity of these consequences was high for all respondents

the role of response costs

57

(i.e., constant pain in and numbness and swelling of limbs). Subsequently, respondents received a communication, which manipulated argument quality and response costs of the action recommendation. Respondents read a fictitious letter promoting stress management training to reduce vulnerability to RSI, which was said to have recently been printed in a newspaper. This message contained either six strong or six weak supporting arguments. It was further stated that the training was either in the city where respondents attended university or in a city 35 miles away (one hour by car or public transport). Respondents attitude, intention, and behavior toward the recommendation were then assessed, as were negative affect, perceived threat, and cognitive responses (i.e., the thoughts they had about the letter promoting stress management training). Before leaving, respondents were thoroughly debriefed. Manipulation checks Vulnerability was assessed by three items on 7-point rating scales [how vulnerable respondents perceived themselves to be, 1= not vulnerable; 7= very vulnerable, how high they thought their risk for RSI was, 1= very low; 7= very high, and the probability of them experiencing negative health consequences due to RSI, 1= very low; 7= very high (= .90)]. Argument quality was assessed by asking respondents to indicate on 7-point rating scales ranging from 1(not at all) to 7(very) how strong, persuasive, and meaningful they found the arguments used in the recommendation (= .86). Response costs were assessed by two items on 7-point rating scales asking how much effort respondents thought participating in the recommendation would require (= .82). Dependent variables Negative affect. Negative affect was measured with six semantic differential items with 7-point rating scales (fearless-fearful, calm-restless, comfortableuncomfortable, good-bad, relaxed-tense and optimistic-pessimistic; (= .87). Perceived threat. Cognitive perceptions of threat were assessed with two items on 7-point rating scales [How threatening are the described health consequences of RSI?, 1= not threatening; 7= very threatening; and How disturbing are the described health consequences of RSI?, 1= not disturbing; 7= very disturbing (= .80)]. Cognitive responses. To measure the cognitive processing of the recommendation respondents were asked to write down all their thoughts concerning the recommendation. Measures of persuasion. Attitude toward stress management training was measured by having respondents indicate on five semantic differential items

58

Chapter 3

with 7-point rating scales how useful, good, important, interesting, and effective they found the training (= .83). In line with tradition in research on fear-arousing communications, the attitude measure assesses attitude toward the object (i.e., stress management training) rather than attitude toward the behavior (i.e., participating in stress management training). However, to assess whether effects are similar for attitude toward the behavior, we also included a measure of individuals attitude toward participating. Attitude toward participating in the stress management training was measured by having respondents indicate on five semantic differential items with 7-point rating scales how useful, good, important, interesting, and effective they found participating in stress management training (= .90). Intention to participate was measured by three items with 7-point rating scales ranging from 1(certainly not) to 7(certainly) [Do you have the intention to, do you want to, and are you going to participate in stress management training? (= .92)]. Two types of behavior were measured, with one item each and yes/no response options. Respondents could ask for more information about stress management training and could sign up for it. It was pointed out, however, that these commitments were not binding. Results Three-way analyses of variance (ANOVAs) that included vulnerability, argument quality, and response costs manipulations as independent factors were executed for all dependent measures, unless stated otherwise. Manipulation checks Vulnerability. A test of the manipulation of vulnerability revealed a vulnerability main effect, F(1,112)= 63.52, p < .001 (2 = .36). Respondents perceived their vulnerability as higher in the high vulnerability condition (M = 4.33, SD = 1.19) than in the low vulnerability condition (M = 2.64, SD = 1.09). No other effects were found. Argument quality. The manipulation of argument quality was effective. Respondents, the strong arguments were perceived as stronger (M = 4.60, SD = .94) than the weak arguments (M = 3.70, SD = 1.09), F(1,112)= 22.13, p < .001 (2 = .17). No other effects were found. Response costs. The manipulation of response costs was also effective, F(1,112)= 44.61, p < .001 (2 = .29). Respondents perceived the high response cost condition as requiring more effort (M = 5.32, SD = .85) than the low response cost condition (M = 4.13, SD = 1.05). No other effects were found.

the role of response costs

59

Dependent variables Negative affect. There was a main effect of vulnerability, F(1,112) = 16.94, p < .001 (2 = .13). Vulnerable respondents experienced more negative affect (M = 4.17, SD = .69) than non-vulnerable respondents (M = 3.58, SD = .85). No other effects were found. Perceived threat. A main effect of vulnerability was found, F(1,112)= 10.67, p < .001 (2 = .09). Vulnerable respondents found the health consequences of RSI more threatening (M = 4.84, SD = .85) than non-vulnerable respondents (M= 4.26, SD = 1.09). No other effects were found. Cognitive responses. Two independent raters scored the number of thoughts about the recommendation in each of three categories: positive thoughts (Range: 0-4; Kappa= .90), negative thoughts (Range: 0-3; Kappa= .93) and neutral thoughts (Range: 0-1; Kappa= .93) about stress management training. Because only few respondents reported neutral thoughts (5.8%), no analyses were performed on this variable. Examples of positive thoughts are: I think stress management training is very interesting, and Its a great initiative, more people should participate in it. Examples of negative thoughts are: Im not participating just so I dont have to blame myself if I do get RSI, thats ridiculous, and I dont think stress management training is a very effective way to prevent RSI. Respondents reported an average of 2.48 (SD = .90) thoughts in total (Range: 1-5). A 3-factor ANOVA on total number of thoughts about the recommendation did not result in any significant effects. Positive and negative thoughts were divided by total number of thoughts resulting in positive and negative thought ratios (Range:0-1). Analysis of positive and negative thoughts separately resulted in main effects of vulnerability on positive thoughts, F(1,112)= 19.96, p < .001 (2 = .15), and on negative thoughts, F(1,112)= 26.81, p < .001 (2 = .19). Vulnerable respondents reported more positive thoughts about the recommendation (M = .67, SD = .39) than non-vulnerable respondents (M = .37, SD = .35). In addition, vulnerable respondents reported less negative thoughts (M = .28, SD = .38) than non-vulnerable respondents (M = .62, SD = .35). On negative thoughts there also was a main effect of argument quality, F(1,112)= 8.54, p < .01 (2 = .07). Respondents had more negative thoughts about the recommendation when it was supported by weak (M = .54, SD = .40) rather than strong arguments (M = .35, SD = .38). No further effects were found. Attitude toward the recommendation. There was a main effect of argument quality F(1,112)= 26.05, p < .001 (2 = .19). Respondents had more positive attitudes toward the recommendation when it was supported by strong (M = 5.04, SD = .79) rather than by weak arguments (M = 4.32, SD = .73). No other effects were found (Table 3.1).

60

Chapter 3

Table 3.1. Means (SDs) of attitudes and intention by vulnerability, argument quality, and response cost Vulnerability Low Attitude toward the recommendation Low effort Weak arguments Strong arguments High effort Weak arguments Strong arguments Attitude toward participating Low effort Weak arguments Strong arguments High effort Weak arguments Strong arguments Intention Low effort Weak arguments Strong arguments High effort Weak arguments Strong arguments 2.42 (1.08) 2.16 (.81) 3.50 (1.20) 3.62 (1.37) 3.04 (.87) 2.82 (1.10) 3.62 (.83) 3.42 (1.40) 4.05 (1.05) 4.61 (.60) 4.23 (.93) 4.77 (.95) 4.32 (.76) 4.93 (1.09) 4.21 (.92) 4.73 (1.26) 4.45 (.44) 4.81 (.49) 4.31 (.79) 5.00 (.82) 4.28 (.83) 5.28 (.83) 4.24 (.85) 5.07 (.96) High

Attitude toward participating. There was a main effect of argument quality F(1,112)= 10.11, p < .01 (2 = .08). Respondents had more positive attitudes toward participating in the recommendation when it was supported by strong (M = 4.76, SD = .99) rather than by weak arguments (M = 4.20, SD = .90). No other effects were found (Table 3.1). Intention. There was a vulnerability main effect, F(1,112) = 21.26, p < .001 (2 = .16). Respondents had higher intentions to participate in the recommendation when their vulnerability was high (M = 3.54, SD = 1.19) than when their vulnerability was low (M = 2.61, SD = 1.01) (Table 3.1). In addition, there was a marginally significant interaction-effect between vulnerability and response costs, F(1,112)= 2.86, p = .09 (2 = .03). Contrastanalyses showed that respondents had higher intentions to participate in the recommendation when response costs were low than when response costs were high, but only under low vulnerability (M = 2.93, SD = .93 vs. M = 2.29, SD = .95, p < .05). Vulnerable respondents intention was independent of

the role of response costs

61

required effort (M = 3.52, SD = 1.14 vs. M = 3.56, SD = 1.26). Behavior. Logistic regression-analysis was used to assess the impact of the independent variables on the two measures of behavior. This analysis showed a main effect of vulnerability on requests for information (Wald(1)=7.54, p < .01). Vulnerable respondents more often requested information about the recommendation (41.7%) than non-vulnerable respondents (18.3%). No other effects were found (Table 3). The same pattern emerged for actual signing up (Wald(1)= 6.92, p < .01). Vulnerable respondents signed up more often (30%) than non-vulnerable respondents (10%). No other effects were found. Mediation Hierarchical regression-analyses were performed to assess the possible mediation of the effects of argument quality and vulnerability on attitudes, intention, and behavior, by any of the other dependent variables in this study. Following procedures specified by Baron and Kenny (1986), variables were only entered as possible mediators if three conditions were met: (1) the independent variable affected the mediator, (2) the independent variable affected the dependent variable, and (3) the mediator affected the dependent variable. Which variables met conditions 1 and 2 can be derived from the analyses of variance described above, which variables also met condition 3 can be found in the correlation matrix (Table 3.2).
Table 3.2. Correlations between all dependent variables Variable 1. Perceived threat 2. Negative affect 3. Positive thoughts 4. Negative thoughts 5. Attitude recommendation 6. Attitude participating 7. Intention 8. Request for information 9. Subscription
* p < .05. ** p < .01.

1 --.52** .13

2 --.22*

---

-.17 .24** .33** .34** .33** .32**

-.27** .26** .31** .44** .32** .35**

-.61** .31** .28** .36** .21* .14

---.33** -.23* -.20* -.03 -.02 --.73** .34** .24** .23** --.51** .36** .35** --.71** .67** --.76**

62

Chapter 3

Table 3.3. Hierarchical regression-analyses predicting attitudes, intentions, and behavior Dependent variable: attitude recommendation Step 1 Argument quality Step 2 Argument quality Negative thoughts Dependent variable: attitude participating Step 1 Argument quality Step 2 Argument quality Negative thoughts Dependent variable: Intention Step 1 Vulnerability Step 2a Vulnerability Positive thoughts Step 2b Vulnerability Negative affect Step 3 Vulnerability Positive thoughts Negative affect Dependent variable: behavior Step 1 Vulnerability Step 2 Vulnerability Intention
* p < .05. ** p < .01. *** p < .001.

.43*** .36*** -.21*

R .19***

.22***

.29** .24* .14 .39*** .29** .24** .27** .34*** .18* .21* .33*** Information .26** .03 .72*** .51*** R .07**

.08**

.08** R .15***

.20***

.25***

.29*** Subscription .25** R .06** .01 .67*** .45***

With respect to the effect of argument quality on attitude toward the recommendation, only negative thoughts about the recommendation qualified as a possible mediator. Hierarchical regression-analysis showed that adding negative thoughts to the model that regressed attitude from argument quality, did not reduce the effect of argument quality on attitude to non-significance (Table 3.3). However, the effect of argument quality was reduced and a Sobel-

the role of response costs

63

test of mediation (Baron & Kenny, 1986) was significant (Z = 2.74, p < .001). Thus, negative thoughts partly mediated the effect of argument quality on attitude toward the recommendation. Negative thoughts were also a possible mediator for the effect of argument quality on attitude toward participating. However, when negative thoughts were added to the model that regressed attitude toward participating from argument quality, the effect of negative thoughts was non-significant ( = .14, ns), indicating that negative thoughts about the recommendation did not mediate the effect of argument quality on attitude toward participating. Regarding the effect of vulnerability on intention, the following dependent variables qualified as possible mediators: negative affect, perceived threat, positive thoughts and negative thoughts about the recommendation. When negative thoughts were added to the model that regressed intention from vulnerability, the effect of negative thoughts was non-significant ( = .10, ns), indicating that negative thoughts about the recommendation did not operate as a mediator. Separately adding negative affect, perceived threat and positive thoughts about the recommendation to the regression equation substantially reduced the effect of vulnerability on intention (Table 3.3). When adding all possible mediators simultaneously into the regressionanalysis, perceived threat no longer was significantly related to intention. The combined effect of negative affect and positive thoughts reduced the effect of vulnerability on intention substantially. Tests of mediation were significant for both negative affect (Z = 3.27, p < .001) and positive thoughts (Z = 3.20, p < .001). Hence, the effect of vulnerability on intention was partly mediated by negative affect and positive thoughts about the recommendation. Finally, the effect of vulnerability on behavior was mediated by intention. Entering vulnerability and intention into hierarchical regressions with the two measures of behavior as criterion separately, reduced the vulnerability effect on both behavior measures to non-significance (Table 3.3). Tests of mediation were significant both for requesting information (Z = 4.26, p < .001) and for signing up (Z = 4.18, p < .001). Discussion The pattern of findings observed in the present study is consistent with predictions derived from our stage model. Vulnerable respondents presented with a message depicting severe health consequences experienced more negative emotions and perceived the consequences as more threatening than non-vulnerable respondents. In addition, vulnerable respondents had more positive thoughts about the action recommendation, regardless of argument quality and response costs. For both attitude measures we found the expected

64

Chapter 3

main effect of argument quality. Respondents held more favorable attitudes toward both the recommendation and toward participating when the recommendation was supported by strong rather than weak arguments, regardless of vulnerability and response costs. This suggests that attitudes were mainly based on objective evaluation of the information. Although main effects were the same for both attitude measures, mediation processes were not. Whereas the effect of argument quality on attitude toward the recommendation was partly mediated by negative thoughts about the recommendation, the effect of argument quality on attitude toward participating was not. In other words, negative thoughts about the recommendation aroused by weak arguments, had more of an impact on what respondents thought about the recommendation than on what they thought about participating. The main effect of argument quality on attitudes found in this experiment, as well as in our previous experiment (De Hoog et al., 2005a) is contrary to the findings in studies on personal relevance and message processing. In many studies an interaction-effect between personal relevance and argument quality on attitudes was found, indicating that only under high personal relevance did respondents base their attitudes on the arguments in the message (e.g., Howard-Pitney, Borgida, & Omoto, 1986; Petty & Cacioppo, & Goldman, 1981; Petty & Cacioppo, 1984). However, the difference between those experiments and our experiments is that we do not vary vulnerability through the same range. In many of the studies on personal relevance the issue in the message in the low relevance condition is really irrelevant to the respondents. In contrast, in our experiments even in the low vulnerability condition, one can still get the disease, but it is simply less likely, and, hence, the message is still somewhat relevant. In addition, because in our experiments the message depicts a health risk with severe consequences, the severity of the topic also heightens motivation to process the message systematically. When it came to persuasion outcomes that had behavioral implications, vulnerability was the main determinant. Vulnerable respondents had higher intentions to participate in the recommendation than non-vulnerable respondents and subscribed to the recommendation more often. More specifically, vulnerable respondents had more positive thoughts about the recommended action, regardless of argument quality and response costs, and in combination with the negative affect aroused by the fear appeal this resulted in higher intentions to engage in the protective action as well as more actual behavior. This supports our expectation that vulnerability determines intention and behavior. The effects of response costs in this experiment concur with our assumptions. We had expected that the effort required to engage in the

the role of response costs

65

protective action should have little impact on intention and behavior, when respondents feel vulnerable to the depicted health consequences. We found a marginally significant interaction-effect between response costs and vulnerability on intention, indicating that, whereas for non-vulnerable respondents intention became lower when a recommendation high in required effort was involved, for vulnerable respondents the effort required in following the recommendation had no effect on their intention to engage in the recommendation or on their actual behavior. This confirms our assumption that vulnerable respondents form a strong intention to perform a recommended protective action, regardless of response costs. We did not find this interactioneffect between vulnerability and response costs on actual behavior. However, this is not surprising, since non-vulnerable respondents intentions were already low, and, therefore, their intentions did not lead to actual behavior. In fact, the number of non-vulnerable respondents subscribing to the recommendation was so few that it is statistically unlikely to find any differentiation effect for response costs. Previous research on the effect of response costs on persuasion has found main effects of response costs on persuasion, with increases in response costs resulting in decreases in attitude, intention and behavior change (see Floyd, Prentice-Dunn, & Rogers, 2000; Milne, Sheeran, & Orbell, 2000). However, these studies did not look at the specific impact of varying response costs for vulnerable individuals. This experiment clearly shows that because of their defense-motivated positive processing bias of the action recommendation, vulnerable respondents were motivated to engage in a protective action regardless of response costs. For non-vulnerable respondents, the results in previous studies do correspond with findings in this experiment in which nonvulnerable respondents had lower intentions when response costs were high than when they were low. The results from this study replicate and extend the findings from our previous experiments (Das et al., 2003; De Hoog et al., 2005a). In these experiments, as in the present one, we found a positive bias in the processing of the action recommendation, with vulnerable respondents having more positive thoughts about the recommendation, regardless of argument quality. Additionally, vulnerability has consistently been the only determinant of intention and behavior, and this effect has consistently been mediated by positive thoughts about the action recommendation and negative affect aroused by the fear appeal. The study presented here extends these findings by demonstrating that defense-motivated vulnerable individuals do not only disregard response efficacy (i.e., argument quality), but also response costs in deciding to engage in actions that protect them against the deleterious effects of a serious health risk.

66

Chapter 3

Of course there can be limitations to the willingness of defense-motivated individuals to engage in an offered protective action. These individuals should at least have some conviction that the action they intend to engage in offers some protection against a health risk. In a similar way, even though it was found that vulnerable respondents are willing to put more effort into engaging in a recommendation, it is doubtful they would be prepared to follow a certain action at all costs, just as they are unlikely to participate in an action that is obviously ineffective or pointless. The coherent pattern of findings that emerged from the present study and our earlier experiments has important theoretical as well as practical implications. Whereas the emphasis of both health education campaigns and empirical research has frequently been on depicting and assessing the effects of the severity of health consequences, and on stressing the response efficacy of the recommended action, we have found that, although these factors affect attitudes, they fail to have much of an impact on intention and behavior. Intention and behavior are solely determined by vulnerability. This suggests that, however severe a health risk, and however effective the protection offered by the recommendation, unless we can persuade individuals that they are vulnerable to the health risk, they are unlikely to take protective action.

Chapter 4: The impact of fear appeals on processing and accepting action recommendations: the role of dispositional optimism and neuroticism3

Research on the impact of fear-arousing communications has been mainly concerned with health related appraisals and has focused on how aspects of the communication influence message processing and persuasion. Less attention has been paid to the potential role of personality variables, even though it is theoretically plausible that traits such as dispositional optimism or neuroticism might influence threat appraisal and persuasion. Therefore, the experiment presented in this article assessed the potential role of these two personality factors in a study of the impact of the manipulation of vulnerability to a health risk, on the processing and acceptance of a recommended protective action, of which the quality of the arguments supporting the recommendation are varied. The impact of fear-arousing communications Fear-arousing communications involve exposing individuals to the negative consequences of health behaviors, attempting to make individuals concerned into changing their health impairing behaviors. In over fifty years of fear appeal research several fear appeal theories have been proposed, including the drivereduction model (Hovland, Janis & Kelley, 1953), protection motivation theory (Rogers, 1985), the parallel response model (Leventhal, 1970) and the extended parallel process model (Witte, 1992). A more recent model for studying fear-arousing communications is the stage model of processing of fear-arousing communications (Das, de Wit & Stroebe, 2003; De Hoog, Stroebe & de Wit, 2005a; Stroebe, 2000), a theoretical framework that builds on previous fear appeal theories (Leventhal, 1970; Rogers, 1983; Witte, 1992), as well as on dual-process theories of attitude change (Chaiken, 1980; Petty & Cacioppo, 1986). In line with most fear appeal theories, the stage model assumes that individuals exposed to fear-arousing communications engage in two types of appraisal, namely appraisal of the threat, and appraisal of coping strategies available for reducing the threat. If individuals are exposed to information about a severe health risk and if they feel vulnerable to this risk, this will threaten their self-definitional belief that they are healthy, and arouse defense motivation as well as the motivation to carefully scrutinize the information
3

This chapter is based on: De Hoog, N., Stroebe, W., & de Wit, J.B.F. The impact of fear

appeals on processing and acceptance of action recommendations: the role of dispositional optimism and neuroticism. Manuscript submitted for publication.

70

Chapter 4

given. According to the stage model, defense motivation manifests itself in biased systematic processing. The direction of the bias will vary with the type of appraisal. In appraising the health threat, defense-motivated individuals will attempt to minimize the threat by critically looking at the content of a fear appeal. They will engage in a biased search for inconsistencies and logical errors, and their evaluation of evidence will be biased in the direction of the preferred conclusion (Ditto & Lopez, 1992; Sherman, Nelson & Steele, 2000). If this strategy proves unsuccessful, because even biased processing is constrained by evidence and rules of inference (Kunda, 1987), individuals will have to accept that they are at risk. In this case, the processing of the action recommendation will be biased as well, but in a positive direction. The processing goal of defensemotivated individuals will be to find the recommended protective action effective, because then they can feel save. The processing will involve a biased search for, and evaluation of, arguments supporting the effectiveness of the protective action (see Das et al., 2003 and De Hoog et al., 2005a for a more extensive description of this model). The role of personality variables Most research on the impact of fear-arousing communications has focused on how aspects of the communication influence persuasion. There has been relatively little attention to the potential influence of personality variables, even though many researchers have proposed that personality variables might influence the effectiveness of fear-arousing communications (Leventhal, 1970; Witte, 1992). Over the years, several personality factors have been studied in relation to fear-arousing communications (Dabbs & Leventhal, 1966; Krisher, Darley & Darley, 1973; Sherman et al., 2000). However, the only personality factors that have been researched more extensively in the fear appeal literature are trait anxiety and self-esteem, but these studies resulted in mixed findings (Dabbs & Leventhal, 1966; Jepson & Chaiken, 1990; Krisher et al., 1973). Very few fear appeal studies have looked at the effects of dispositional optimism or neuroticism, with the exception of a study by Aspinwall and Brunhart (1996). They showed that dispositional optimism is adaptive in confronting threats to well-being. Optimistic respondents paid more attention to threatening health information than respondents low on optimism. Dispositional optimism refers to a general positive outlook on life (Scheier & Carver, 1985), and has to be distinguished from defensive optimism, which is mainly displayed in the so-called optimistic bias, when making estimations of ones risks (Weinstein, 1982). Neuroticism is a trait also known as negative affectivity (Watson & Clark, 1984), and refers to a broad, stable dimension of personality associated with increased emotional distress and chronic negative

the role of personality factors

71

emotions, as well as cognitive and behavioral characteristics (Eysenck & Eysenck, 1975). Research has linked dispositional optimism to better physical health and better psychological well-being. Optimists report taking better care of themselves when they have minor as well as life-threatening illnesses. In addition, optimism has been associated with faster recovery after surgery and higher perceptions of quality of life (Lin & Peterson, 1990; Scheier & Carver, 1992). Dispositional optimism has also been shown to be related to active and problem-focused coping and other adaptive behaviors (Aspinwall & Taylor, 1992; Scheier & Carver, 1992). For example, Scheier and Carver (1992) found that optimistic individuals relied more on active and problem-focused coping than pessimistic individuals. It seems that the positive expectancies of optimists lead them to more active and effective problem-solving (Scheier & Carver, 1985; 1992). Neuroticism has been associated with somatic complaints, as well as more life dissatisfaction and some mood disorders, including depression (Costa & McCrae, 1987). Research on neuroticism has also shown that neurotics take longer to recover from sickness and surgery than non-neurotics (Clark, Watson & Mineka, 1994; Watson & Clark, 1984). However, it has also been found that individuals high on neuroticism are not unhealthier, but do report more health complaints than others. Thus, neuroticism seems to mainly affect subjective rather than objective health (Costa & McCrae, 1987; Williams, OBrien & Colder, 2004). Whereas dispositional optimism has been associated with more active forms of coping, neuroticism has mainly been linked to avoidant or less active (Bolger, 1990; Costa & McCrae, 1987). Because of the limited and unclear empirical evidence of the role of personality factors in the effectiveness of fear-arousing communications, and because of findings that both optimism and neuroticism have an effect on several aspects of health and behavior, the present research focused on the impact of dispositional optimism and neuroticism on persuasion through feararousing communications. Design overview and hypothesis This experiment assessed dispositional optimism and neuroticism, and manipulated vulnerability to a severe health risk as well as the quality of the arguments supporting a recommended action. Respondents were asked to fill in optimism and neuroticism scales after which their vulnerability to Repetitive Strain Injury (RSI) was manipulated. Respondents then read a message depicting the potential health consequences of RSI. Subsequently, an action recommendation was offered, which was supported by either low or high quality arguments. Deriving predictions from the stage model of processing of fear-arousing

72

Chapter 4

communications (Das et al., 2003; Stroebe, 2000), we expected vulnerable respondents exposed to a message depicting negative health consequences to experience more negative affect than non-vulnerable respondents. We further expected that inducing vulnerability to a health risk would arouse defense motivation. Individuals, who feel vulnerable to a serious health risk, should be motivated to accept the action recommendation. They should, therefore, overestimate the effectiveness of the recommendation and have more positive and less negative thoughts about the recommendation, regardless of argument quality. In line with previous findings (De Hoog et al., 2005a), we expected attitudes to remain unbiased by defense motivation and to be determined purely by argument quality. We expected attitudes to be more positive when strong arguments were used than when weak arguments were used. In addition, we expected weak arguments to generate more negative thoughts about the recommendation than strong arguments. For intention and behavior, we expected a main effect of vulnerability: Only individuals who feel vulnerable should want to change their behavior. We had differential expectations for optimism and neuroticism. We expected that respondents high on neuroticism exposed to a fear-arousing communication would react with more negative emotions, but because of their known inactive coping, we expected neuroticism to have no effect on any of the other dependent variables, especially those that have behavioral implications (e.g., intention and behavior). In contrast, we expected respondents high in optimism to experience less negative emotions, and because of their general positive outlook on life and known active coping, we expected optimists to have a more positive reaction to the action recommendation. We expected respondents high in optimism to have more positive and less negative thoughts about the recommendation. In addition, we expected respondents high on optimism to have more positive attitudes, higher intentions, and more behavior than respondents low in optimism. Finally, we expected personality factors to have direct effects on outcome measures and these effects should not interact with or overrule the expected effects of vulnerability and argument quality. Experiment 4.1 Method Participants and design Respondents were 279 (87 male, 192 female) students of Utrecht University, who participated in the experiment for payment of 4. Nine respondents, who expressed suspicion of the vulnerability manipulation, were not included in this analysis. The experiment had a 2(vulnerability) x 2(argument quality) x

the role of personality factors

73

2(dispositional optimism) x 2(neuroticism) design. Independent variables Vulnerability. Respondents were given false feedback after completing a bogus RSI test. Half the respondents were told they were very vulnerable to RSI (Very vulnerable: The probability of you developing RSI is very high) and the other half were told they were not very vulnerable (Not very vulnerable: The probability of you developing RSI is very low). Argument quality. The quality of the arguments supporting the action recommendation was manipulated by presenting a recommendation, which contained either six strong or six weak supporting arguments. Following the procedure described by Petty and Cacioppo (1986), these arguments were selected from a pool of arguments that had been rated as weak or strong in a pilot study. Examples of strong supporting arguments are: Research has shown that people who participate in stress management training develop RSI less often than people who do not participate and After participating in the training 90% of the participants were able to cope better with high work pressures and developed more efficient ways of working. Weak supporting arguments included Stress management training is a lot of fun and good for social contacts and If you participate in stress management training at least you do not have to blame yourself when you do develop RSI. Dispositional optimism. Respondents completed the Dutch version of the Life Orientation Test (LOT; Scheier & Carver, 1985; Vinck, Wels, Arickx & Vinck, 1998), an 8-item scale that measures dispositional optimism. Examples of items are: I always look on the bright side of life and In uncertain times I usually expect the best (1= strongly agree, 5= strongly disagree; = .76). The LOT has been shown to have good discriminant validity with respect to a variety of constructs including neuroticism (Scheier & Carver, 1985; Scheier, Carver & Bridges, 1994). Using a median-split procedure respondents were divided into respondents high and low in optimism (Md= 31, SD= 4.34). Neuroticism. Respondents were asked to fill in the neuroticism sub-scale of the Dutch translation of the Eysenck Personality Questionnaire, a 16-item scale that measures neuroticism (EPQ; Eysenck & Eysenck, 1975; Van Kampen, 1993). Examples of items are: I worry about the horrible things that might happen to me and I get easily upset when other people criticize me (1= strongly agree, 5= strongly disagree; = .92). Respondents were divided into high and low in neuroticism by means of a median-split procedure (Md= 54, SD= 17.12). Procedure The experiment was run on personal computers. After a brief introduction, respondents were asked to fill in the optimism and neuroticism scales.

74

Chapter 4

Respondents then received a communication that manipulated their vulnerability to the health risk. They were told about a new test that could measure vulnerability to RSI. This test required them to shift a computer mouse repeatedly following various patterns displayed on the screen. They were led to believe a computer program would then assess their vulnerability to RSI from various parameters of their movements (i.e., pressure, speed). Respondents received false feedback in the form of a thermometer appearing on their computer screens in which the red virtual mercury rose either almost all the way to the top of the thermometer, indicating high vulnerability, or it stayed low on the thermometer, indicating low vulnerability. Next, a message about the health consequences of RSI was presented. This message was a 300word text in which the consequences of RSI were described in detail. It was stated that RSI is a progressive condition with very serious and long lasting physical consequences. The severity of these consequences was high for all respondents (i.e., constant pain in and numbness and swelling of limbs). Subsequently, respondents read a fictitious letter promoting stress management training to reduce vulnerability to RSI, which was said to have recently been printed in a newspaper. This message contained either six strong or six weak supporting arguments. Respondents negative affect and cognitive responses (i.e., their thoughts about stress management training) were then assessed, as were manipulation checks, and lastly attitude, intention, and behavior toward the recommendation. Before leaving, respondents were thoroughly debriefed. Manipulation checks Vulnerability was assessed by three items on 7-point rating scales [how vulnerable respondents perceived themselves to be, 1= not vulnerable; 7= very vulnerable, how high they thought their risk for RSI was, 1= very low; 7= very high, and the probability of them experiencing negative health consequences due to RSI, 1= very low; 7= very high (= .85)]. Argument quality was assessed by asking respondents to indicate on 7-point rating scales ranging from 1(not at all) to 7(very) how strong, persuasive, and meaningful they found the arguments used in the recommendation (= .91). Dependent variables Negative affect. Negative affect was measured with six semantic differential items with 7-point rating scales (fearless-fearful, calm-restless, comfortableuncomfortable, good-bad, relaxed-tense and optimistic-pessimistic; = .90). Cognitive responses. To measure the cognitive processing of the recommendation respondents were asked to write down all their thoughts concerning the recommendation. Measures of persuasion. Attitude toward stress management training was

the role of personality factors

75

measured by having respondents indicate on four semantic differential items with 7-point rating scales how useful, good, important, and effective they found the training ( = .87). Intention to participate was measured by three items with 7-point rating scales ranging from 1(certainly not) to 7(certainly) [Do you have the intention to, do you want to, and are you going to participate in stress management training? ( = .89)]. Two types of behavior were measured, with one item each and yes/no response options. Respondents could ask for more information about the training and could sign up for it. It was pointed out, however, that these commitments were not binding. Results Preliminary analyses showed that there were no gender effects on any of the dependent variables. Therefore, four-way analyses of variance (ANOVAs) that included vulnerability and argument quality manipulations, and optimism and neuroticism as independent factors were executed for all dependent measures, unless stated otherwise. Manipulation checks Vulnerability. A test of the manipulation of vulnerability revealed a vulnerability main effect, F(1, 263)= 115.93, p < .001 (2 = .31). Respondents perceived their vulnerability as higher in the high vulnerability condition (M = 4.07, SD = 1.16) than in the low vulnerability condition (M = 2.41, SD = 1.12). No other effects were found. Argument quality. The manipulation of argument quality was effective. Respondents, who received strong arguments, perceived the arguments as stronger (M = 4.62, SD = 1.05) than did those who received weak arguments (M = 3.62, SD = 1.10), F(1, 263)= 42.62, p < .001 (2 = .14). No other effects were found.
Table 4.1. Means (SDs) of negative affect by vulnerability, argument quality, optimism, and neuroticism Low vulnerability Low optimism Weak arguments Low neurotic High neurotic Strog arguments Low neurotic High neurotic 3.33 (1.23) 3.89 (1.18) 3.41 (.89) 3.76 (1.19) 4.42 (.81) 4.69 (.99) 4.22 (.89) 4.52 (.53) 3.72 (.51) 3.97 (.68) 3.42 (1.11) 3.53 (.92) 4.63 (.87) 4.53 (.84) 3.95 (1.30) 3.99 (1.21) High optimism High vulnerability Low optimism High optimism

76

Chapter 4

Dependent variables Negative affect. There was a main effect of vulnerability, F(1, 263)= 30.78, p < .001 (2 = .11). Vulnerable respondents experienced more negative affect (M = 4.35, SD = .99) than non-vulnerable respondents (M = 3.64, SD = 1.00). In addition, a main effect of optimism was found, F(1, 263)= 5.05, p < .05 (2 = .02), as well as a marginally significant main effect of neuroticism, F(1, 263)= 2.82, p = .094 (2 = .01) (Table 4.1). Optimistic respondents (M = 3.81, SD = 1.09) experienced less negative affect than non-optimistic respondents (M = 4.20, SD = .98). Respondents low on neuroticism experienced less negative affect (M = 3.82, SD = 1.06) than respondents high on neuroticism (M = 4.18, SD = 1.02). No other effects were found. Cognitive responses. Two independent raters scored the number of thoughts about the recommendation in each of three categories: positive thoughts (Range: 0-5; Kappa= .90), negative thoughts (Range: 0-4; Kappa= .93), and neutral thoughts (Range: 0-1; Kappa= .93) about the recommendation. Because only few respondents listed neutral thoughts, no analyses were performed on neutral thoughts. An example of positive thoughts is: I think its a great initiative, more people should participate in it. An example of negative thoughts is: I dont think stress management training is a very effective way to prevent RSI. Respondents reported an average of 2.36 (SD = 1.06) thoughts in total. A 4-factor ANOVA on total number of thoughts about the recommendation did not result in any significant effects. Positive and negative thoughts were divided by total number of thoughts resulting in positive and negative thought ratios. Analysis of positive thought ratios resulted in a main effect of vulnerability, F(1,263)= 16.26, p < .001 (2 = .06). Vulnerable respondents reported more positive thoughts about the recommendation (M = .66, SD = .37) than nonvulnerable respondents (M = .47, SD = .42). There also was a marginally significant main effect of optimism, F(1, 263)= 3.30, p = .074 (2 = .01). Optimistic respondents had more positive thoughts (M = .61, SD = .39) than non-optimistic respondents (M = .52, SD = .41). No other effects were found. Analysis of negative thoughts resulted in main effects of vulnerability, F(1, 263)= 7.60, p < .01 (2 = .03), argument quality, F(1, 263)= 6.37, p < .01 (2 = .02) and optimism, F(1, 263)= 4.34, p < .05 (2 = .02). Vulnerable respondents reported less negative thoughts (M = .34, SD = .36) than nonvulnerable respondents (M = .43, SD = .40). Further, respondents had more negative thoughts about the recommendation when it was supported by weak (M = .48, SD = .41) rather than strong arguments (M = .29, SD = .34), and optimistic respondents had less negative thoughts (M = .34, SD = .36) than non-optimistic respondents (M = .44, SD = .40). No other effects were found. Attitude. There was a main effect of argument quality F(1,263)= 38.41, p < .001 (2 = .13). Respondents had more positive attitudes toward the

the role of personality factors

77

recommendation when it was supported by strong (M = 5.17, SD = .85) rather than by weak arguments (M = 4.34, SD = 1.04) (Table 4.2). No other effects were found. Intention. There was a vulnerability main effect, F(1,263)= 59.07, p < .001 2 ( = .18). Vulnerable respondents had higher intentions to participate in the recommendation (M = 3.59, SD = 1.28) than non-vulnerable respondents (M = 2.39, SD = 1.10). In addition, there was a marginally significant effect of optimism, F(1,263)= 3.43, p = .07 (2 = .01). Optimistic respondents (M = 3.10, SD = 1.37) had higher intentions than non-optimistic respondents (M = 2.86, SD = 1.29). No other effects were found (Table 4.2).
Table 4.2. Mean (SDs) attitude and intention by vulnerability, argument quality, optimism, and neuroticism Low vulnerability Attitude Weak arguments Low neuroticism High neuroticism Strong arguments Low neuroticism High neuroticism Intention Weak arguments Low neuroticism High neuroticism Strong arguments Low neuroticism High neuroticism 1.96 (.95) 2.10 (.97) 2.72 (1.26) 2.38 (1.02) 3.31 (1.32) 3.54 91.32) 3.65 91.36) 3.93 91.63) 2.06 (.98) 2.48 (1.04) 2.45 (1.23) 2.67 (.94) 3.67 (.93) 3.50 (1.24) 3.60 (1.33) 3.62 (1.21) 4.68 (.80) 5.07 (.93) 5.47 (.72) 4.97 (.89) 5.08 (.54) 5.36 (.88) 5.02 (.80) 5.46 (1.15) 4.25 (1.11) 4.32 (1.01) 4.48 (1.16) 4.60 (1.06) 4.18 (.93) 4.15 (1.00) 4.43 (1.17) 4.33 (.90) High vulnerability Low optimism High optimism Low optimism High optimism

Behavior. Logistic regression-analysis was used to assess the impact of the independent variables on the two measures of behavior. This analysis showed main effects of vulnerability (Wald(1)= 31.32, p < .001) and optimism (Wald(1)= 3.98, p < .05) on request for information. Vulnerable respondents more often requested information about the recommendation (63.3%) than non-vulnerable respondents (29.3%). In addition, optimistic respondents (51%) more often requested information than non-optimistic respondents (40.9%). No other effects were found. For actual signing up only a main effect of vulnerability was found (Wald(1)= 17.46, p < .001). Vulnerable respondents signed up more often (27.3%) than non-vulnerable respondents (7.1%).

78

Chapter 4

Discussion The pattern of findings observed in the present study is consistent with our predictions. As expected, respondents high in dispositional optimism experienced less negative affect and had more positive and less negative thoughts about the recommendation than respondents low in dispositional optimism. Furthermore, only dispositional optimism and not neuroticism had an effect on persuasion. Optimistic respondents had higher intentions and requested information more often than non-optimistic respondents. These results correspond with research findings that optimistic individuals generally have more active and problem-focused ways of coping, whereas neurotic individuals are more likely to apply avoidant or inactive coping (Costa & McCrae, 1987; Scheier et al., 1986). We have to point out, however, that many of the effects found for dispositional optimism were only marginally significant. Furthermore, although optimistic respondents had more positive thoughts about the recommendation, and also showed more behavioral intentions and requested more information, no effect was found for attitudes. However, this result does correspond with findings in previous studies that attitudes are mainly based on the objective information at hand (Das et al., 2003; De Hoog et al., 2005). The only effect for neuroticism that was found was on negative affect, and this effect was only marginally significant. As expected, respondents high in neuroticism experienced more negative emotions than respondents low in neuroticism. This is not surprising because individuals high on neuroticism are known to react with more negative emotions (Watson & Clark, 1984). As expected, no effects for neuroticism on any of the measures of persuasion were found. This confirms the notion that individuals high on neuroticism tend to react to stressors with more negative affect, but they do not attempt to actively cope with the problem (Costa & McCrae, 1987). Neither being optimistic nor being neurotic had an effect on perceptions of vulnerability to RSI. Feelings of vulnerability were only determined by the vulnerability manipulation. This clearly demonstrates the difference between dispositional optimism, as measured in this experiment, and the well-known optimistic bias (Weinstein, 1982), which makes individuals estimate their risk for negative consequences as lower than is the case. Perceptions of vulnerability of respondents high in neuroticism correspond with research findings that although individuals high in neuroticism are generally more negative and report more health complaints, they are realistic in estimations of their own risks (Williams et al., 2004). The pattern of findings for vulnerability and argument quality is also consistent with our predictions. Vulnerable respondents presented with a message depicting severe health consequences experienced more negative

the role of personality factors

79

emotions than non-vulnerable respondents. In addition, vulnerable respondents had more positive and less negative thoughts about the recommendation, regardless of argument quality. For attitudes, we found a main effect of argument quality. Respondents held more favorable attitudes toward the recommendation, when it was supported by strong rather than weak arguments. This suggests that attitudes were mainly based on objective evaluation of the information. When it came to persuasion outcomes that had behavioral implications, however, vulnerability was the main determinant. Vulnerable respondents had higher intentions to participate in the recommendation, requested more information, and subscribed to the recommendation more often. Finally, as expected, none of the effects of optimism or neuroticism interacted with or overruled any of the effects of vulnerability and argument quality on any of the dependent variables. In other words, in this study that included optimism and neuroticism as extra factors, we still found all the expected effects of vulnerability and argument quality. This is not surprising, because many of the effects of optimism, and the one effect of neuroticism, were only marginally significant. These findings seem to imply that although personality factors, or at least optimism and neuroticism marginally, have some effect on how respondents perceive and respond to fear-arousing communications, the main way to get people to change their behavior is through feelings of vulnerability aroused by a health communication. The results from this study replicate and extend the findings from previous experiments testing predictions from the stage model (Das et al., 2003; De Hoog et al., 2005a). In these experiments, as in the present one, a positive bias in the processing of the action recommendation was found, with vulnerable respondents having more positive thoughts about the recommendation, regardless of argument quality. Additionally, vulnerability has consistently been the main determinant of intention and behavior. The study presented here extends these findings by demonstrating that, although neuroticism and dispositional optimism have some effect on the way respondents perceive a threat and a recommended action, feelings of vulnerability resulting from a health communication is the most important element in getting people to change their behavior.

Chapter 5: Getting the whole picture: How biased processing leads to persuasion4

Fear-arousing communications are widely used in health education campaigns. It is assumed that the more one succeeds in making individuals concerned about the consequences of their health impairing behavior, the greater will be the probability that they will accept the recommended action. However, even though research on fear appeals resulted in evidence that high fear messages produce more behavior change than low fear messages (Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000), these empirical studies have not succeeded in fully elucidating the processes, which mediate the persuasive impact of fear appeals. The experiments presented in this article assess how individuals process fear-arousing communications. We test predictions derived from the stage model of processing of fear-arousing communications (Das, de Wit & Stroebe, 2003; De Hoog, Stroebe & de Wit, 2005a; Stroebe, 2000), a theoretical framework that builds on previous fear appeal theories (Leventhal, 1970; Rogers, 1983; Witte, 1992), as well as on dual-process theories of attitude change (Chaiken, 1980; Chaiken, Liberman & Eagly, 1989; Petty & Cacioppo, 1986). In three experiments, we will illustrate how fear-arousing communications work, by showing the impact of manipulations of vulnerability to and severity of a health risk on the processing of both the fear appeal and the subsequent recommended protective action, and by establishing how these processes relate to attitudes, intention, and behavior. Theoretical perspectives in the study of fear-arousing communications Research on the impact of fear-arousing communications can be divided into three distinct phases. The first phase began with the classic study of Janis and Feshbach (1953) and was theoretically guided by the drive-reduction model of fear appeals (Hovland, Janis & Kelley, 1953). According to this model, informing recipients about a health risk arouses fear, which is reduced by the recipients rehearsal of the communicators recommendations. When a response reduces fear, it is reinforced and becomes part of ones permanent response repertory. The drive model, therefore, suggests that higher fear should result in more persuasion, but only if the recommended action is perceived as effective in avoiding the danger. Extensions of the drive-reduction model were later suggested by Janis (1967) and McGuire (1968).
4

This chapter is based on: De Hoog, N., Stroebe, W., & de Wit, J.B.F. (2005). The

processing of fear-arousing communications: How biased processing lead to persuasion. Manuscript submitted for publication.

82

Chapter 5

Because part of the empirical evidence was inconsistent with these drive models, Leventhal (1970) developed the parallel response model, which introduced threat appraisal as the important mediator between environmental threat and action, and abandoned the notion that emotional arousal is a necessary antecedent of the adaptation to danger. This model was the beginning of the second (more cognitive) phase of research on fear-arousing communications. According to the parallel response model, an environmental threat is cognitively evaluated by the individual and this appraisal can give rise to two parallel or independent processes, namely danger control and fear control. Danger control involves the decision to act as well as instrumental actions to reduce or eliminate the threat. Actions in the service of fear control, such as avoidance actions (including defenses) as well as attempts to control the emotional response (e.g., drinking alcohol), frequently have no effect on the actual danger. The important contribution of the parallel response model is the central role given to cognitive appraisal processes and the differentiation between emotional and cognitive responses to fear-arousing communications. A weakness of this model is that it does not specify the processes of cognitive appraisal that precede action tendencies. This task was completed by later models that focused exclusively on elaborating Leventhals danger control process, namely the health belief model (Rosenstock, 1974) and protection motivation theory (Rogers, 1983). More recently, Wittes extended parallel process model (1992) reemphasized the role of fear by elaborating conditions under which fear control is likely to affect danger control. According to the extended parallel process model, perceived efficacy of the recommended action will determine whether individuals, who believe that they are susceptible to a serious threat, will engage in danger or fear control. Individuals will mainly engage in danger control when they perceive the recommended action as effective in reducing the threat, and they will mainly engage in fear control when they perceive the recommended action as ineffective or when they feel unable to perform the recommended action. The one assumption shared by all Phase 1 and Phase 2 models is that level of threat interacts with the perceived efficacy of a recommended action in their impact on attitudes, intention and behavior. These models suggest that even the most vulnerable individuals will not adopt protective actions they perceive as ineffective in averting negative consequences. However, there seems to be little empirical evidence for any such interaction. In their recent meta-analysis, Witte and Allen (2000) found no significant interaction between threat and efficacy. Furthermore, even though the cognitive models developed in Phase 2, and even the drive-reduction models of Phase 1, assumed that cognitive processes affect fear arousal and/or mediate persuasion, they made no predictions about information processing, and measures of information processing (e.g., cognitive responses) have been virtually absent in research

Getting the whole picture

83

on fear-arousing communications. This situation changed in Phase 3, with the advent of dual-process theories of attitude change (Chaiken, 1980; Petty & Cacioppo, 1986), which were soon applied to account for the impact of fear-arousing communications on persuasion (e.g., Gleicher & Petty, 1992; Liberman & Chaiken, 1992). Dualprocess theories focus on the processing of the arguments contained in a communication as well as on the outcomes of persuasion. According to dualprocess theories, fear arousal can have two effects, namely act as a motivator to induce recipients to engage in intensive and thoughtful message processing, and induce defense motivation, which will lead to biased message processing. Whereas the unbiased or accuracy-motivated perceiver assesses the validity of attitude-relevant information in the interest of achieving a valid position, the processing goal of defense-motivated recipients is to confirm the validity of a preferred position and disconfirm the validity of non-preferred positions. Thus, defense-motivated individuals will process and perceive information in ways that best support their own beliefs. This implies that information that is congruent with a particular preferred position will be judged as more valid than information that is incongruent with this position (Ditto & Lopez, 1992; Lord, Ross & Lepper, 1979; Pyszczynski & Greenberg, 1987). The stage model of processing of fear-arousing communications The stage model of processing of fear-arousing communications (Das et al., 2003; De Hoog et al., 2005a; Stroebe, 2000) integrates ideas from dualprocess theories (e.g., Chaiken, 1980) with those of earlier theories of feararousing communications (Leventhal, 1970; Rogers, 1983; Witte, 1992). In line with most Phase 2 theories, our model assumes that individuals exposed to fear-arousing communications engage in two types of appraisal, namely appraisal of the threat, and appraisal of coping strategies available for reducing or eliminating the threat. These two appraisal processes parallel the two components of fear-arousing communications, namely the fear appeal, which emphasizes the severity of and vulnerability to the health risk, and the action recommendation, which provides information on how to avoid the health risk. Expanding upon the differentiation between appraisal of threat and appraisal of recommendations (e.g., Leventhal, 1970; Witte, 1992), our stage model makes differential predictions for the appraisal processes at each of these two stages. Threat appraisal implied by the fear appeal is assumed to determine both the processing mode (i.e., depth of processing) and the processing goal (i.e., accuracy or defense motivation). According to our stage model, if individuals are exposed to information about a severe health risk and feel vulnerable, this will seriously threaten their self-definitional belief that they are healthy, and arouse defense motivation as well as the motivation to carefully scrutinize the information given. We assume

84

Chapter 5

that processing will be systematic instead of heuristic, because any communication describing a serious personal health threat is likely to require a thorough evaluation (Chaiken et al., 1989). According to our stage model, defense motivation induced by high threat perceptions manifests itself not only in avoidance reactions as previous fear appeal models have proposed (e.g., Witte, 1992), but in systematic processing, that is biased. The direction of the bias aroused by defense motivation will vary with the type of appraisal. In appraising the health threat (Stage 1), defense-motivated individuals will attempt to minimize the threat by critically looking at the content of a fear appeal. They will try to criticize and downplay the information in order to reduce the threat. Defense-motivated individuals will engage in a biased search for inconsistencies and logical errors, and their evaluation of evidence will be biased in the direction of the preferred conclusion (Ditto & Lopez, 1992; Liberman & Chaiken, 1992; Sherman, Nelson & Steele, 2000). If this strategy proves unsuccessful, because even biased processing is constrained by evidence and rules of inference (Kunda, 1987), individuals will have to accept that they are personally at risk. In this case, the subsequent processing of the recommendation will be biased as well (Stage 2). At this stage, the processing goal of defense-motivated individuals will be to find the protective action effective, because then they can feel save. Even though the easiest way would be to accept the recommendation without scrutinizing the arguments contained in the message, it is not likely that heuristic processing strategies will prevail when serious health risks are involved (Chaiken et al., 1989). The processing will involve attempts to make the recommendation appear highly effective by means of a biased search for arguments supporting the effectiveness of the protective action, and through biased evaluation of these arguments. According to our stage model, if an individual feels vulnerable but the health risk is minor, there is no reason to be defensive, even though the feeling of vulnerability should be sufficient motivation to invest effort into systematically processing the contents of the communication. Similarly, if an individual does not feel vulnerable, a health risk, no matter how severe, is equally unlikely to arouse defense motivation. Yet, even non-vulnerable individuals are likely to invest effort in processing the contents of a communication describing a severe risk, because it is useful to be well informed about a serious health risk, even if the danger is not immanent. In other words, our stage model proposes that systematic processing can also occur in the absence of high personal relevance. In this respect, our stage model differs from dual-process theories that state that personal relevance is an important precondition for systematic processing. Of course, if a health risk is trivial and an individual does not feel at risk, he/she is unlikely to invest much effort in thinking about the contents of the communication and he/she might rely on heuristic processing modes.

Getting the whole picture

85

An important distinction with respect to persuasion that is emphasized by our model is that between outcome measures that rely on attitudes and those that assess behavioral intentions or behavior (Fishbein & Ajzen, 1975). Whereas previous fear appeal theories have often made the same predictions for attitudes, intentions and behavior, our stage model makes differential predictions for attitudes on the one hand and intention and behavior on the other. Attitudes about a protective action are based mainly on evaluation of the provided information. Individuals will, hence, have more positive attitudes towards a protective action if the recommendation is supported by strong arguments rather than weak arguments, and when a health risk is described as more serious. Yet, individuals will only be motivated to engage in the protective action if they feel vulnerable. After all, why should one invest effort into avoiding a health risk, if one does not feel personally at risk? Empirical evidence Most of the studies, which assessed the impact of vulnerability and severity on the processing of fear-arousing communications, have focused exclusively on the processing of fear appeals (Stage 1). There is now ample evidence that people are more critical of evidence that is highly health-threatening, than of less threatening evidence (Ditto & Lopez, 1992; Janis & Terwilliger, 1962; Kunda, 1987; Liberman & Chaiken, 1992; Reed & Aspinwall, 1998; Sherman et al., 2000). However, the designs of these studies did not allow the analysis of those variables, which are important according to our stage model. None of these studies focused on attitude vs. intentions as measures of persuasion and either vulnerability or cognitive responses, which indicate processing mode and goal, were not measured explicitly. Much less research has been conducted with respect to the processing of action recommendations (Stage 2), and the few studies on this topic either failed to manipulate vulnerability experimentally (Jepson & Chaiken, 1990) or failed to include measures of cognitive processing (Gleicher & Petty, 1992). More importantly, none of these studies explained how processing of a fear appeal relates to processing of the action recommendation, and how defense motivation can have a positive effect on persuasion. Studies conducted by Das et al. (2003) and De Hoog et al. (2005a) provide direct tests of some of the predictions from our stage model. These studies assessed the effects of the severity of, and vulnerability to a health risk on the processing and acceptance of an action recommendation (Stage 2), which was supported by high or low quality arguments. Consistent with predictions, these studies revealed a positive bias in the processing of the action recommendation under high vulnerability: Vulnerable respondents reported more positive thoughts about the recommendation than non-vulnerable respondents, regardless of argument quality. Furthermore, vulnerability proved to be the

86

Chapter 5

only determinant of individuals intention to engage in the recommended action. Attitudes were mainly determined by argument quality. A shortcoming of these studies is that they only focused on the processing of the action recommendation. The present studies were conducted in order to: (1) examine the processing of the fear appeal; (2) investigate the processing of the action recommendation; and (3) determine how processing of both parts of feararousing communications affect each other and lead to persuasion, and thereby providing further support for our model. Three experiments will be reported which varied/manipulated vulnerability to and severity of a health risk. Experiments 5.1 and 5.2 assessed the impact of vulnerability and severity only on the processing of the fear appeal. Experiment 5.3, in addition, studied the impact of these variables on both the processing of the fear appeal and processing of the action recommendation. Experiment 5.1 This experiment focused on the processing of the fear appeal. The topic of the study was the negative health consequences of alcohol consumption. Respondents reported their average alcohol consumption and indicated how vulnerable they perceived themselves to be with regard to alcohol related health consequences. Subsequently, they read a message depicting the health consequences of alcohol consumption as either high or low in severity. We expected exposure to a message depicting negative health consequences of alcohol consumption to arouse negative affect, and the high severity message should arouse more negative affect than the low severity message. In addition, we expected vulnerable respondents to experience more negative affect than non-vulnerable respondents. With regard to respondents cognitive responses we expected the fear appeal to generate several types of thoughts (e.g., about vulnerability, the severity of the consequences) of which we were particularly interested in thoughts that revealed attempts to minimize the threat, which would indicate a negative processing bias. We predicted that the high severity condition should generate more minimizing thoughts than the low severity condition. Moreover, vulnerable respondents should have more minimizing thoughts than nonvulnerable respondents. Because we expected that defense motivation would be aroused especially when respondents feel vulnerable and when the consequences of the health risk are depicted as severe, respondents in the condition of high vulnerability and severity should experience more negative emotions and have more minimizing thoughts than respondents in the other three conditions. With regard to attitude towards reducing alcohol consumption, we predicted

Getting the whole picture

87

only an effect of the severity manipulation, with attitudes being more positive when consequences are described as more serious. With regard to intention to cut back on drinking, we expected a main effect of vulnerability: only individuals who feel vulnerable should want to change their behavior. Method Participants and design Respondents were 104 (29 male, 75 female) students of Utrecht University, who participated in the experiment as part of course requirements. Four respondents who did not read the severity message were not included in the analysis. The experiment had a 2(vulnerability) x 2(severity) design. Independent variables Vulnerability. Perceived vulnerability was assessed by having respondents rate their vulnerability on three items with 7-point rating scales [how vulnerable respondents perceived themselves to be, 1= not vulnerable; 7= very vulnerable, how high they thought their risk for developing alcohol related complaints was, 1= very low; 7= very high, and the probability of them experiencing negative health consequences due to their alcohol consumption, 1= very low; 7= very high (= .94)]. Using a median split procedure respondents were divided into a high and a low vulnerable group (Md = 1.17, SD = 1.08).2 Severity. Respondents read a 600-word message, which described the health consequences of alcohol consumption in detail. Physical, as well as neurological effects were discussed and respondents were provided with numerous examples of negative consequences. The text was divided into four sections; the first two sections described the effects of alcohol on bodily organs and on the brain. The consequences were much more serious in the high severity (e.g., liver failure, brain damage) than in the low severity condition (e.g., stomachaches, reaction speed). The third and fourth section for the low severity condition consisted of hangover and weight gaining information, and for the high severity condition consisted of information about cancer risk and death. These two messages were rated high and low in severity, respectively, in a pilot study. Procedure The experiment was run on personal computers. After a brief introduction, respondents were asked about their alcohol consumption (how often and how much alcohol they consumed) and their perceived vulnerability for alcohol related health problems. Subsequently, respondents read a message about the negative health consequences of alcohol consumption. This message depicted

88

Chapter 5

health consequences of alcohol consumption as either low or high in severity. After this message, respondents were first queried about their affect and cognitions regarding the severity message, and the effectiveness of the manipulation was checked. Then attitude and intention toward cutting back on drinking were measured. Before leaving, respondents were thoroughly debriefed. Manipulation check Severity was assessed by three items, namely how severe, harmful and serious respondents perceived the described health consequences of alcohol consumption, with 7-point rating scales ranging from 1(not at all) to 7(very) (= .85). Dependent variables Negative affect. Negative affect was measured with six semantic differential items with 7-point rating scales (not fearful-fearful, calm-restless, comfortableuncomfortable, good-bad, relaxed-tense and optimistic-pessimistic; = .91). Cognitive responses. To measure the cognitive processing of the severity message respondents were asked to write down all their thoughts concerning the severity message. Measures of persuasion. Attitude towards cutting back on drinking was measured by having respondents indicate on four semantic differential items with 7-point rating scales how useful, good, important, and effective they found cutting back on alcohol (= .92). Intention to cut back was measured by three items with 7-point rating scales ranging from 1 (certainly not) to 7 (certainly) [Do you have the intention to, do you want to and are you going to cut back on your alcohol consumption? (= .94)]. Results Preliminary analyses showed that there were no gender effects on any of the dependent variables. Therefore, two-way analyses of variance (ANOVAs) that included vulnerability and severity as factors were executed for all dependent measures, unless stated otherwise. In addition, planned contrasts were performed to test whether the condition of high vulnerability and high severity differed significantly from all other conditions on negative affect and cognitive responses. Every condition included 25-27 respondents. Manipulation checks Severity. The manipulation of severity was effective. Respondents who received the message with severe consequences, perceived the consequences as more severe (M = 6.35) than those who received the message low in severity (M = 5.46), F(1,100)= 52.96, p < .001 (2= .35). There was also a

Getting the whole picture

89

marginally significant effect of vulnerability, F(1,100)= 3.56, p = .06 (2= .03). Vulnerable respondents perceived the consequences as less severe (M = 5.80) than non-vulnerable respondents (M = 6.00). No other effects were found. Dependent variables Negative affect. Main effects of both vulnerability, F(1,100)= 11.81, p < .001 (2 = .11) and severity, F(1,100)= 4.93, p < .05 (2 = .05) were found (Table 5.1). Vulnerable respondents experienced more negative affect (M = 3.70) than non-vulnerable respondents (M = 2.90). In addition, the message high in severity elicited more negative emotions (M = 3.57) than the message low in severity (M = 3.03). No other effects were found. Planned contrasts showed, as expected, that vulnerable respondents under high severity experienced more negative affect than respondents in all other conditions, F(1,100)= 12.61, p < .01.
Table 5.1. Means (SDs) of negative affect by vulnerability and severity Low severity Experiment 1 Low vulnerability High vulnerability Experiment 2 Low vulnerability High vulnerability Experiment 3 Low vulnerability High vulnerability 2.91 (1.06) 3.79 (1.12) 3.21 (.93) 4.32 (1.07) 3.27 (1.28) 3.70 (1.09) 3.65 (1.03) 4.36 (1.14) 2.70 (1.23) 3.39 (.98) High severity 3.11 (1.38) 4.00 (1.07)

Cognitive responses. Two independent raters scored the thoughts about the severity message into several categories, including minimizing thoughts (Range: 0-2; Kappa= .92). Examples of minimizing thoughts are: I drink just as much as other students so it cant be too much and These health consequences are exaggerated, alcohol cant be that bad for you. Minimizing thoughts were divided by total number of thoughts, which resulted in a minimizing thought ratio (Range: 0-1). Respondents reported an average of 2.68 (SD= .71) thoughts in total (Range: 1-4). A 3-factor ANOVA on total number of thoughts did not result in any significant effects (Table 5.2). Main effects of both severity, F(1,100)= 9.74, p < .01 (2 = .09) and vulnerability, F(1,100)= 12.45, p < .001 (2 = .11) were found on minimizing thoughts ratio. Respondents more often had minimizing thoughts when the consequences were high in severity (M = .30) than when they were low (M = .13). Moreover, vulnerable respondents more often had minimizing thoughts

90

Chapter 5

(M = .30) than non-vulnerable respondents did (M = .12). No other effects were found. Planned contrasts confirmed our prediction that vulnerable respondents in the high severity condition more often reported minimizing thoughts than respondents in all other conditions, F(1,100)= 22.91, p < .001. Attitude. No effects were found. Respondents were generally positive about cutting back on drinking (M = 5.39). Intention. There was only a main effect of vulnerability, F(1,100)= 8.12, p < .01 (2 = .08). Vulnerable respondent had greater intention to cut back (M = 3.51) than non-vulnerable respondents (M = 2.56).
Table 5.2. Means (SDs) of minimizing thoughts by vulnerability and severity Low vulnerability Experiment 1 Low severity High severity Experiment 2 Low severity High severity Experiment 3 Low severity High severity .04 (.12) .14 (.27) .22 (.36) .49 (.43) .04 (.13) .21 (.31) .26 (.31) .58 (.35) .07 (.16) .17 (.26) .19 (.26) .40 (.28) High vulnerability

Discussion The results of this study confirmed our expectation that vulnerability to a severe health risk induces defense motivation, which is evident in a negative bias in the processing of the fear appeal. Vulnerable respondents in the high severity condition experienced more negative emotions than non-vulnerable respondents in the same condition and all respondents in the low severity conditions. In addition, vulnerable respondents in the high severity condition reported more often thoughts that minimized the severity than did respondents in all other conditions. Another sign for a negative processing bias was found in a marginally significant effect of vulnerability on the severity manipulation check, indicating that vulnerable respondents perceived the negative consequences as less severe than non-vulnerable respondents. No effects on attitude were found. With a mean attitude of 5.39 on a 7-point scale, respondents generally had a positive attitude towards reducing alcohol consumption, irrespective of whether or not the consequences of drinking had been depicted as serious. This could be a ceiling effect or it might indicate that the severity information does not affect an individuals attitude toward some protective action, unless this protective action has been specifically recommended in a communication, which was not the case in this study.

Getting the whole picture However, vulnerable respondents did have a greater intention to cut back on drinking than non-vulnerable respondents did. This supports our expectation that vulnerability determines intention, and presumably action. Experiment 5.2

91

Because perceived vulnerability could be correlated with other variables that are more directly related to an individuals attitude towards reducing alcohol consumption or the intention to cut back on drinking, using self-reported vulnerability to assign respondents to conditions of high and low vulnerability precludes one from drawing strong conclusions about the role of vulnerability in the processing of fear-arousing communications. In a conceptual replication of Experiment 5.1, Experiment 5.2 therefore manipulated vulnerability to test whether the findings of Experiment 5.1 could be reproduced when using manipulated vulnerability instead of relying on self-measures of vulnerability. In addition, Experiment 5.2 also used a different health risk, namely the health consequences of hypoglycemia (i.e., low blood glucose levels). Furthermore, Experiment 5.2 also manipulated the expertise of the source to which the severity information was attributed, to provide an explicit test of our depth of processing expectations. When the likelihood of thoughtful processing is low, source expertise can operate as a heuristic cue and should then affect the persuasive impact of a message mainly during heuristic processing (i.e., when both vulnerability and severity are low). Method Participants and design Respondents were 123 (36 male, 87 female) students at Utrecht University, who participated in the experiment for payment of 3. Four respondents who expressed suspicion of the vulnerability manipulation were not included in the analysis. Respondents were run individually and were randomly assigned in equal numbers to the conditions of a 2(vulnerability) x 2(severity) x 2(source expertise) between-subjects factorial design. Independent variables Vulnerability. Respondents were given false feedback after completing Doctor Harpers health test for hypoglycemia, a 55-item questionnaire used to indicate problems with blood glucose levels. Half the respondents were told they were very vulnerable to hypoglycemia (Very vulnerable: The probability of you becoming hypoglycemic is very high) and the other half were told they were not very vulnerable to hypoglycemia (Not very vulnerable: The probability of you becoming hypoglycemic is very low). Severity. Respondents read a 350-word message, which described the

92

Chapter 5

health consequences of hypoglycemia in detail. It described the impact of low glucose levels on several processes in the body and in the brain as well as numerous physical and psychological negative consequences. The severity of these consequences was either high (e.g., convulsions, organ failure, depression) or low (e.g., headaches, mood swings, fatigue). These two messages were rated high and low on severity, respectively, in a pilot study. Source expertise. Source expertise was manipulated by attributing the severity message to either a scientific magazine (high expertise) or a popular womens magazine (low expertise). Procedure The experiment was run on personal computers. After a brief introduction, respondents received a communication that manipulated their vulnerability to the health risk. They were told there was a test that could measure vulnerability to hypoglycemia. Respondents were asked to fill out this questionnaire and then received false feedback on their computer screens. Half the respondents were told their score on the test indicated they were very vulnerable to hypoglycemia and the other half was told their score meant they were not very vulnerable to hypoglycemia. Next, a message about the health consequences of hypoglycemia was presented. This message depicted the health consequences of hypoglycemia as either high or low in severity. This message was attributed to either a high or a low expertise source. After this message, respondents were first queried about their affect and cognitions regarding the severity message, and the effectiveness of the manipulations was checked. Then attitude and intention toward being tested for hypoglycemia were measured. Before leaving, respondents were thoroughly debriefed. Manipulation checks Source expertise was assessed by four items [How scientific did you find the message?, How reliable do you think the information in the message is?, How much of an expert do you think the author of the message is? and How accomplished do you think the author of the message is? (= .82)], with 7point rating scales ranging from 1 (not at all) to 7 (very). The manipulation checks of vulnerability and severity were assessed with the same items as used in Experiment 5.1. Dependent variables All the measures of dependent variables were identical to those used in Experiment 5.1.

Getting the whole picture

93

Results Preliminary analyses showed that there were no gender effects on any of the dependent variables. Therefore three-way analyses of variance (ANOVAs) that included vulnerability, severity, and source manipulations as factors were performed for all dependent measures, unless stated otherwise. In addition, the same planned contrast-analyses were carried out as in Experiment 5.1. Manipulation checks Vulnerability. A test of vulnerability revealed a main effect of vulnerability, F(1,115)= 38.80, p < .001 (2 = .25). Respondents perceived their vulnerability as higher in the high vulnerability condition (M = 4.06) than in the low vulnerability condition (M = 2.51). No other effects were found. Severity. The manipulation of severity was effective. Respondents who received the message with consequences of hypoglycemia high in severity, perceived the consequences as more severe (M = 5.45) than did those who received the message with consequences that were low in severity (M = 4.13), F(1,115)= 70.57, p < .001 (2 = .38). No other effects were found. Source. There was a main effect of source, F(1,115)= 28.11, p < .001 (2 = .20). The high credibility source was perceived as more credible (M = 4.85) than the low credibility source (M = 4.09). No other effects were found. Dependent variables Negative affect. Main effects of both vulnerability, F(1,115)= 7.48, p < 01 2 ( = .06) and severity were found, F(1,115)= 6.20, p < .05 (2= .05). Vulnerable respondents experienced more negative affect (M = 4.03) than nonvulnerable respondents (M = 3.46), and the message high in severity elicited more negative emotions (M = 4.00) than the message low in severity (M = 3.49). No other effects were found. Planned contrasts showed, as expected, that vulnerable respondents in the high severity condition experienced significantly more negative affect than all other respondents, F(1,119)= 11.73, p < .01 (Table 5.1). Cognitive responses. Two independent raters scored respondents thoughts about the severity message into several categories, including minimizing thoughts (Range: 0-2; Kappa= .85). Examples of minimizing thoughts are: I do not think the consequences of hypoglycemia are as bad as they described and I am not becoming hypoglycemic, I am just having a very stressful week. Minimizing thoughts were divided by total number of thoughts, which resulted in a minimizing thought ratio (Range: 0-1). Respondents reported an average of 2.08 (SD = .68) thoughts in total (Range: 1-4). A 3-factor ANOVA on total number of thoughts about the severity message did not result in any significant effects.

94

Chapter 5

Main effects of both severity, F(1,119)= 20.19, p < .001 (2= .15) and vulnerability, F(1,119)= 30.80, p < .001 (2= .21) were found on minimizing thoughts. Respondents more often had minimizing thoughts when the consequences were high in severity (M = .39) than when they were low (M = .15). Moreover, vulnerable respondents more often had minimizing thoughts (M = .42) than non-vulnerable respondents did (M = .13). No other effects were found. Planned contrasts confirmed that vulnerable respondents under high severity more often reported minimizing thoughts than respondents in all other conditions, F(1,119)= 43.52, p < .001 (Table 5.2). Attitude. No effects were found. Respondents were generally positive about being tested for hypoglycemia (M = 5.21) (Table 5.3). Intention. There was only a main effect of vulnerability, F(1,115)= 12.76, p < .01 (2= .10). Vulnerable respondents had greater intention to be tested (M = 3.81) than non-vulnerable respondents (M = 2.97) (Table 5.3).
Table 5.3. Means (SDs) of attitude and intention by vulnerability, severity and source expertise Experiment 5.2 Low vulnerability Attitude Low severity Low source High source High severity Low source High source Intention Low severity Low source High source High severity Low source High source 2.85 (1.41) 3.18 (1.48) 3.84 (1.37) 3.76 (1.26) 2.49 (1.17) 3.33 (1.66) 3.65 (.91) 4.00 (1.10) 4.80 (1.08) 5.30 (.77) 5.27 (.79) 5.40 (.75) 5.45 (.89) 5.23 (.72) 4.94 (1.13) 5.37 (.53) High vulnerability

Discussion The results of this second study reproduced the main findings from our first study. We found that with a completely different topic and manipulated instead of self-measures of vulnerability, vulnerability to a severe health risk induces defense motivation. Vulnerable respondents in the high severity condition experienced more negative emotions than non-vulnerable respondents in the same condition and all respondents in the low severity conditions. Moreover, these vulnerable respondents reported minimizing thoughts more often than

Getting the whole picture

95

any other respondents did. There is little support for our expectation that recipients of information about a health risk, which is not severe and to which they do not feel vulnerable, process the information about the health threat less systematically than respondents in other conditions do. Our source expertise manipulation failed to have a significant effect on any dependent variable in this experiment, despite a significant manipulation check indicating that respondents clearly perceived the strong source as stronger than the weak source. However, given the fact that even in the low severity condition, the perceived severity of the health risk was above the midpoint on the scale, these findings are not inconsistent with our theory. We appear to have failed in creating health threats, which are really perceived as being trivial. In this study, similar to Experiment 5.1, we found no effects on attitude, but did find the expected vulnerability main effect on intention. Both vulnerable and non-vulnerable respondents had positive attitudes towards being tested for hypoglycemia, regardless of severity. Vulnerable respondents had a greater intention to be tested for hypoglycemia than non-vulnerable respondents did. Experiment 5.3 In the previous two experiments, no detailed action recommendation was offered and only the fear appeal was presented, which gave us the opportunity to examine the processing of the fear appeal in detail. To test our entire model a third experiment was conducted, which generally used procedures identical to Experiment 5.2, with the difference that an action recommendation was now offered. Respondents were told about a food workshop that could reduce vulnerability to hypoglycemia, and this message was accompanied by either strong or weak supporting arguments. Additionally to our expectations described in Experiment 5.1, we expected attitudes to be more positive when strong arguments were used than when weak arguments were used. In addition, weak arguments should generate more negative thoughts about the recommendation than strong arguments. Furthermore, we predicted vulnerable respondents to have more positive and less negative thoughts about the recommendation than non-vulnerable respondents. Because we expected a positive processing bias especially when respondents felt vulnerable and the consequences were depicted as severe, respondents in the condition of high vulnerability and severity should generate more positive and less negative thoughts about the recommendation than respondents in the other three conditions. To avoid an overly complex design and, more importantly, because no effects were found, source expertise manipulations were dropped.

96

Chapter 5

Method Participants and design Data were collected from 126 (31 male, 95 female) students of Utrecht University, who participated in the experiment for payment of 3. Three respondents who expressed suspicion of the vulnerability manipulation were not included in the analysis. Respondents were run individually and were randomly assigned in equal numbers to the conditions of a 2(vulnerability) x 2(severity) x 2(argument quality) between-subjects factorial design. Independent variables The same vulnerability and severity manipulations were used as in Experiment 5.2. In addition, we manipulated the quality of the arguments in the action recommendation. Argument quality. The quality of the arguments supporting the action recommendation was manipulated by presenting a recommendation, which contained either six strong or six weak supporting arguments. Following the procedure described by Petty and Cacioppo (1986), these arguments were selected from a pool of arguments that had been rated as weak or strong in a pilot study. Examples of strong supporting arguments are Research has shown that people who participate in the hypoglycemia food workshops develop hypoglycemia less often than people who do not participate, The food workshop makes you less vulnerable to blood glucose level problems, and After participating in the training 90% of the participants were able to cope better with their hypoglycemic complaints. Weak supporting arguments included Food workshops are a lot of fun and good for social contacts, If the workshops had not been effective they would never have been offered by the institute that developed the workshops, and If you participate in the food workshop at least you do not have to blame yourself when you do become hypoglycemic. Procedure The procedure was identical to Experiment 5.2 with the exception that after respondents were presented with the severity message they were offered an action recommendation. Respondents received a communication, which manipulated argument quality of the action recommendation. Respondents read a message promoting food workshops to reduce vulnerability to hypoglycemia, which contained either six strong or six weak supporting arguments.

Getting the whole picture

97

Manipulation checks Checks of the manipulation of vulnerability and severity were assessed with the same items as used in Experiment 5.1. Argument quality was assessed by asking respondents to indicate, on 7-point rating scales ranging from 1(not at all) to 7(very) how strong, persuasive, and meaningful they found the arguments in the recommendation (= .92). Dependent variables The dependent variables were identical to those used in Experiment 5.1 and 5.2. In addition, we also measured cognitive responses to the recommendation, and actual behavior. Cognitive responses. To measure the cognitive processing of the recommendation, respondents were asked to write down all their thoughts concerning the action recommendation. Behavior. Two types of behavior were measured with one item each and yes/no response options. Respondents could ask for more information about the food workshop and could sign up for it. It was pointed out, however, that these commitments were not binding. Results Preliminary analyses showed that there were no gender effects on any of the dependent variables. Therefore three-way analyses of variance (ANOVAs) that included vulnerability, severity, and argument quality manipulations as factors were performed for all dependent measures, unless stated otherwise. The same planned contrasts were performed, as in the previous experiments. In addition, planned contrasts were also performed for cognitive responses to the action recommendation, contrasting the high vulnerability/high severity condition with the three other conditions. Manipulation checks Vulnerability. A test of the manipulation of vulnerability revealed a main effect of vulnerability, F(1,118)= 102.23, p < .001 (2 = .46). Respondents perceived their vulnerability as higher in the high vulnerability condition (M = 4.35) than in the low vulnerability condition (M = 2.44). No other effects were found. Severity. The manipulation of severity was effective. Respondents who received the message depicting severe consequences of hypoglycemia, perceived the consequences as more severe (M = 5.60) than did those who received the message with consequences low in severity (M = 4.21), F(1,118)= 85.78, p < .001 (2= .42). There also was a main effect of vulnerability, F(1,118)= 7.68, p < .05. (2 = .06). Non-vulnerable respondents found the

98

Chapter 5

consequences more severe (M = 5.13) than vulnerable respondents (M = 4.69). No other effects were found. Argument quality. The manipulation of argument quality was effective. Respondents who received strong arguments, perceived the arguments as stronger (M = 4.99) than did those who received weak arguments (M = 3.31), F(1,118)= 94.56, p < .001 (2 = .45). No other effects were found. Dependent variables Negative affect. Main effects of both vulnerability, F(1,118)= 28.14, p < .001 (2 = .19) and severity were found, F(1,118)= 4.94, p < .05 (2= .04). Vulnerable respondents experienced more negative affect (M = 4.05) than nonvulnerable respondents (M = 3.06) and the severe message elicited more negative affect (M = 3.76) than the message low in severity (M = 3.36). No other effects were found. Planned contrasts showed, as expected, that vulnerable respondents in the high severity condition experienced more negative affect than respondents in all other conditions, F(1,122)= 21.84, p < .001 (Table 5.1). Cognitive responses: severity message. Two independent raters scored respondents thoughts about the severity message into several categories, including minimizing thoughts (Range: 0-2; Kappa= .87). Minimizing thoughts were divided by total number of thoughts, which resulted in a minimizing thought ratio (Range: 0-1). Respondents reported an average of 2.01 (SD = .79) thoughts in total (Range: 1-5). A 3-factor ANOVA on total number of thoughts about the severity message did not result in any significant effects. Main effects of both severity, F(1,122)= 10.67, p < .01 (2= .08) and vulnerability, F(1,122)= 22.56, p < .001 (2= .16) were found for minimizing thoughts. Respondents more often had minimizing thoughts when the consequences were high in severity (M = .31) than when they were low (M = .13). Moreover, vulnerable respondents more often had minimizing thoughts (M = .35) than non-vulnerable respondents did (M = .09). No other effects were found. Planned contrasts confirmed that vulnerable respondents under high severity more often reported minimizing thoughts than respondents in all other conditions, F(1,122)= 30.34, p < .001 (Table 5.2). Cognitive responses: action recommendation. Two independent raters scored the number of thoughts about the recommendation in each of three categories: positive thoughts (Range: 0-4; Kappa= .89), negative thoughts (Range: 0-4; Kappa= .89), and neutral thoughts (Range: 0-1; Kappa= .92) about the food workshop. Examples of positive thoughts are: I think the food workshop is a great idea and I would love to participate and the food workshop sounds very interesting. Examples of negative thoughts are: I do not think a food workshop can really help you with hypoglycemia and I am not participating just so I do not have to blame myself if I do become

Getting the whole picture

99

hypoglycemic. Because only few respondents (6.4%) listed neutral thoughts, no analyses were performed on neutral thoughts. Both positive and negative thoughts were divided by total number of thoughts, which resulted in positive and negative thought ratios (Range: 0-1). Respondents reported an average of 2.71 (SD = .90) thoughts in total (Range: 1-5). A 3-factor ANOVA on total number of thoughts about the recommendation did not result in any significant effects. Analysis of the positive thoughts ratio resulted in a main effect of vulnerability, F(1,118)= 12.02, p < .01 (2= .09). Vulnerable respondents reported more positive thoughts about the recommendation (M = .81) than non-vulnerable respondents (M = .50). No other effects were found. Planned contrasts confirmed that vulnerable respondents in the high severity condition had more positive thoughts than respondents in all other conditions, F(1,122)= 22.48, p < .001. On the negative thoughts ratio main effects of argument quality, F(1,118)= 5.41, p < . 05 (2 = .04), and vulnerability were found, F(1,118)= 10.08, p < .01 (2 = .08). Respondents had more negative thoughts about the action recommendation when it was supported by weak (M = .41) rather than strong arguments (M = .28) and vulnerable respondents had less negative thoughts (M = .26) than non-vulnerable respondents (M = .44). No other effects were found. Planned contrasts showed that, as expected, vulnerable respondents in the high severity condition had less negative thoughts than respondents in all other conditions, F(1,122) = 13.40, p < .001.
Table 5.4. Means (SDs) of attitude and intention by vulnerability, severity, and argument quality Experiment 5.3 Low vulnerability Attitude Low severity Weak arguments Strong arguments High severity Weak arguments Strong arguments Intention Low severity Weak arguments Strong arguments High severity Weak arguments Strong arguments 2.92 (1.54) 2.21 (1.11) 3.54 (1.30) 4.11 (1.26) 2.47 (.97) 2.79 (1.48) 3.13 (1.03) 3.65 (1.28) 5.08 (1.33) 5.20 (.93) 4.50 (.74) 5.90 (.77) 4.38 (1.33) 5.61 (.78) 4.26 (.78) 5.00 (.81) High vulnerability

100

Chapter 5

Attitude. There were main effects of argument quality, F(1,118)= 26.26, p < .001 (2 = .18) and severity, F(1,118)= 4.38, p < .05 (2 = .04). Respondents had more positive attitudes towards the workshop when it was supported by strong (M = 5.42) rather than by weak arguments (M = 4.56). In addition, respondents had more positive attitudes when the consequences were presented as high in severity (M = 5.16) rather than low in severity (M = 4.82). There was also a three-way interaction between argument quality, severity and vulnerability, F(1,118)= 6.75, p < .05 (2 = .05) (Table 5.4). Simple effect analyses revealed for non-vulnerable respondents a main effect of argument quality, F(1,122)= 6.67, p < .05, as well as a significant interaction, F(1,122)= 4.44, p < .05, indicating that non-vulnerable respondents have more positive attitudes when strong arguments were used instead of weak ones, but only when the consequences were depicted as low in severity. Vulnerable respondents had more positive attitudes when severity was high, F(1,122)= 5.19, p < .05, and when strong arguments were used, F(1,122)= 18.28, p < .001. Intention. There was only a vulnerability main effect, F(1,118)= 20.13, p < .001 (2 = .15). Respondents had greater intention to participate in the training when their vulnerability was high (M = 3.60) than when their vulnerability was low (M = 2.60) (Table 5.4). Behavior. Logistic regression analysis was used to measure the impact of the independent variables on the two measures of behavior. This analysis showed a main effect of vulnerability on requests for information (Wald(1)= 5.61, p < .05). Vulnerable respondents more often requested information about the recommendation (52.4%) than non-vulnerable respondents (31.7%). No other effects were found. For actual signing up only an effect of vulnerability was found (Wald(1)= 7.70, p < .01). Vulnerable respondents signed up more often (25.4%) than non-vulnerable respondents (6.3%). Mediation Hierarchical regression-analyses were performed to assess a possible mediation of the effects of argument quality and vulnerability on attitude, intention, and behavior, by any of the other dependent variables in this study. Following procedures specified by Baron and Kenny (1986), variables were only entered as possible mediators if three conditions were met: (1) the independent variable affected the mediator, (2) the independent variable affected the dependent variable, and (3) the mediator affected the dependent variable. Which variables met conditions 1 and 2 can be derived from the analyses of variance described above, which variables also met condition 3 can be found in the correlation table (Table 5.5).

Getting the whole picture

101

Table 5.5. Correlations between all dependent variables Experiment 5.3 Variable 1. Negative affect 2. Minimizing thoughts 3. Positive thoughts 4. Negative thoughts 5. Attitude 6. Intention 7. Request for information 8. Subscription
* p < .05. ** p < .01.

1 --.09 .21* -.19* .01 .48*** .16 .11

2 --.30** -.11 .00 .20* .02 .18*

---.52** .26** .30** .10 .18* ---.40** -.16 -.15 -.22* --.26** .01 .14 --.55** .55** --.29**

With respect to the effect of argument quality on attitude, only negative thoughts about the recommendation qualified as a mediator. Hierarchical regression-analysis showed that adding negative thoughts to the model that regressed attitude from argument quality, did not reduce the effect of argument quality on attitude to non-significance (see Table 5.6). However, the effect was reduced and a Sobel test (Baron & Kenny, 1986) of mediation was significant (Z = 2.41, p < .01). Thus, negative thoughts partly mediated the effect of argument quality on attitude. With respect to the effect of vulnerability on intention, negative affect, minimizing thoughts about the fear appeal, and positive thoughts about the recommendation qualified as mediators. When minimizing thoughts were added to the model that regressed intention from vulnerability, the effect of minimizing thoughts appeared non-significant ( = .09, ns), and moreover, did not reduce the effect of vulnerability on intention. Therefore, minimizing thoughts did not operate as a mediator. In contrast, separately adding negative affect and positive thoughts about the recommendation to the regression equation substantially reduced the effect of vulnerability on intention (Table 5.6). The combined effect of negative affect and positive thoughts on intentions reduced the effect of vulnerability on intention to non-significance. Tests of mediation were significant for both negative affect (Z = 3.96, p < .001) and positive thoughts (Z = 2.81, p < .01). Hence, the effect of vulnerability on intention was mediated by negative emotions in view of the consequences and positive thoughts about the recommendation. Minimizing thoughts about the fear appeal, negative and positive thoughts about the action recommendation and intention qualified as possible mediators for the effect of vulnerability on subscription, whereas only intention qualified as possible mediator for requesting information. When minimizing thoughts, negative thoughts and positive thoughts were separately added to the model that regressed subscription from vulnerability, the effect of minimizing

102

Chapter 5

thoughts ( = .11, ns), negative thoughts ( = .08, ns), and positive thoughts ( = .13, ns) appeared non-significant, and were unable to reduce the effect of vulnerability on subscription. Therefore, these variables did not operate as mediators. Entering vulnerability and intention into hierarchical regression with the subscription as criterion reduced the vulnerability effects on this behavior measures to non-significance (Table 5.6). The same outcome was found when vulnerability and intention were entered into the regression with requesting information as criterion. Tests of mediation were significant both for requesting information (Z= 3.78, p < .001) and for signing up for the workshop (Z= 3.78, p < .001). Thus, the effect of vulnerability on behavior was mediated by intention.
Table 5.6. Hierarchical regression-analyses predicting attitudes, intentions, and behavior Experiment 5.3 Dependent variable: attitude Step 1 Argument quality Step 2 Argument quality Negative thoughts Dependent variable: Intention Step 1 Vulnerability Step 2a Vulnerability Positive thoughts Step 2b Vulnerability Negative affect Step 3 Vulnerability Negative affect Positive thoughts Dependent variable: behavior Step 1 Vulnerability Step 2 Vulnerability Intention
* p < .05. ** p < .01. *** p < .001.

.40*** .32*** -.32*** .37*** .27** .36*** .20* .39*** .14 ns .34*** .31*** Information .21* .01 ns .54*** .29*** R .04*

R .16***

.25*** R .13***

.24***

.25***

.33*** Subscription .26** .07 ns .52*** .29*** R .06**

Getting the whole picture

103

Discussion This study revealed the same pattern of results with regard to the processing of the fear appeal that we observed in Experiments 5.1 and 5.2. We again found that vulnerable respondents in the high severity condition experienced more negative emotions than respondents in all other conditions. In addition, vulnerable respondents under high severity also reported minimizing thoughts more often than any other respondents did. As in Experiment 5.1, we also found an effect of vulnerability on the manipulation check of severity, indicating that vulnerable respondents perceived the described consequences of hypoglycemia as less severe than non-vulnerable respondents. For attitudes we found the expected main effect of argument quality. Respondents had a more positive attitude towards the action recommendation when strong arguments were used than when weak arguments were used. In addition, we found, as predicted, that severe consequences resulted in more positive attitudes. The three-way interaction effect revealed that whereas vulnerable respondents had more positive attitudes when strong arguments were used and when the consequences were depicted as severe, for nonvulnerable respondents attitudes were mainly determined by argument quality only when the consequences were depicted as low in severity. Another interpretation is that when both argument quality and severity were low, nonvulnerable respondents had the least positive attitudes. Whereas attitude was mainly determined by argument quality, as well as by severity, persuasion measures with behavioral implications (i.e., intention and behavior) were only influenced by vulnerability. Vulnerable respondents had a greater intention to participate in the workshop, more often requested additional information, and signed up for it more often than non-vulnerable respondents. Mediation-analyses showed that these vulnerability effects on intention were mediated by negative emotions aroused by the fear appeal and positive thoughts about the recommended action, whereas the vulnerability effects on behavior were mediated by intention. A final question addressed in this experiment was the impact defensive appraisal at Stage 1 had on appraisal of the subsequent recommended action. On the one hand, one could expect that the more individuals minimize the health threat, the less they should be willing to accept the recommended action. Alternatively, it is also possible that defense motivation is influenced by a disposition and that individuals, who are most likely to minimize the health threat, are also most likely to maximize the efficacy of the recommended action. The pattern of correlations presented in Table 5.5 supports this latter interpretation. Minimizing thoughts about the fear appeal correlate positively with positive thoughts about the action recommendation, and also with intention to sign up for the workshop and actual signing up.

104

Chapter 5

General discussion The pattern of findings observed in the presented studies is consistent with the assumptions derived from our stage model. All three experiments showed that vulnerability to a severe health risk induces defense motivation. Vulnerable respondents presented with a message depicting severe health consequences experienced more negative emotions than respondents in all other conditions did. In addition, vulnerable respondents reported minimizing thoughts about the fear appeal more often than respondents in all other conditions. In Experiments 5.1 and 5.3 we also found that vulnerable respondents perceived the health consequences as less severe than non-vulnerable respondents. However, this effect was only marginally significant in Experiment 5.1, and non-significant in Experiment 5.2. With regard to attitude, the first two experiments in which no detailed recommendation was offered showed no effects at all on attitude, although respondents generally had very positive attitudes. In the third experiment that included a recommended action, respondents held a more favorable attitude towards the training when the recommendation was supported by strong rather than weak arguments, and when the consequences were severe rather than minor. These contrasting findings with respect to attitude can be explained by the fact that in Experiment 5.1 and 5.2 no information about the recommendation was offered, whereas in Experiment 3 detailed information about a recommendation was given, including supporting arguments. This might indicate that an individuals attitude toward a protective action is not affected by the fear appeal, unless the protective action has been specifically recommended. Furthermore, the type of recommendation offered might also explain these conflicting results. In Experiment 5.1, respondents were recommended to lower their alcohol consumption, in Experiment 5.2 to be tested for hypoglycemia and in Experiment 5.3 they were recommended to follow a food workshop. When asked how one feels about moderating alcohol consumption or getting tested for a disease, one might be naturally inclined to think this is a good thing to do, regardless of the severity of the depicted consequences. This might not be the case for participating in a food workshop. In this situation, attitudes are determined by relying on the information given, like severity of the consequences and the arguments supporting the recommendation. Even though these different recommendations for dissimilar health risks had contrasting effects on attitudes, they had the same effect on intentions. In all three experiments vulnerable respondents had higher intentions to follow the action recommendation than non-vulnerable respondents. Vulnerable respondents who were exposed to a message depicting severe health consequences had the most positive thoughts about the recommended action,

Getting the whole picture

105

regardless of argument quality, and in combination with their experienced negative affect aroused by the fear appeal this resulted in greater intention to engage in the offered protective action as well as more actual behavior. This supports our assumption that when it comes to behavioral intentions and behavior, the most important factor is for individuals to feel vulnerable. The reported experiments demonstrate that exposing vulnerable respondents to a severe health risk arouses defense motivation, which manifests itself not in avoidance reactions as previous fear appeal models have proposed (e.g., Witte, 1992), but in biased systematic processing. In other words, vulnerable respondents experienced most negative affect after exposure to the high fear message and this arouses minimizing thoughts. Yet, when it came to processing of a possible solution to the threat, vulnerable respondents had more positive thoughts about the recommended protective action and a stronger intention to change their behavior. Mediation analyses provided more insight into how information processing affected persuasion. In processing the fear appeal, vulnerable respondents experienced more negative affect and their negative processing bias showed itself in more minimizing thoughts. Their processing of the action recommendation resulted in the more positive thoughts about the recommendation, and these thoughts together with the negative emotions elicited by the fear appeal, mediated the effect of vulnerability on intention. Even though minimizing thoughts did not qualify as a mediator, the significant correlation between this variable and positive thoughts about the recommendation and intention clearly indicate that this factor also contributed to increased persuasion. The negative bias in processing the fear appeal, which results in minimizing thoughts, is positively related to positive thoughts aroused by processing of the recommended action. In other words, it is not the case that vulnerability to a severe health risk, which induces defense motivation results in either minimization of the threat or only in maximization of the recommendation, but the negative processing bias in threat appraisal contributes to the positive processing bias in coping appraisal. The results from these studies replicate and extend the findings from our previous experiments (Das et al., 2003; De Hoog et al., 2005a) in which we focused on the processing of the action recommendation after exposure to a severe health risk. In these previous experiments, as in the current third experiment, we found a positive bias in the processing of the action recommendation, with vulnerable respondents in the severe condition having more positive thoughts about the recommendation, regardless of argument quality. Vulnerability was also the only determinant of intention and behavior, which was mediated by thoughts and affect aroused by the fear-arousing communication. The studies presented here have elaborated on these effects by also exploring the processing of the fear appeal and relating processing of

106

Chapter 5

the fear appeal to processing of the action recommendation and giving more insight into the underlying processes that lead to intention and behavior. The studies in this chapter have shed some light on the processes underlying persuasion through fear-arousing communications, but more research is needed in order to fully understand the processes at work. The one clear finding that has emerged from the present studies as well as from our earlier experiments, is the importance of vulnerability on intention and behavior. This finding has important theoretical and practical implications. Whereas the emphasis of both health education campaigns and empirical research has frequently been on depicting and assessing the effects of the severity of health consequences, and on stressing the response efficacy of the recommended action, we have found that, although these factors affect attitudes, they fail to have much of an impact on intention and behavior. Intention and behavior were solely determined by vulnerability. This suggests that, however severe a health risk, and however effective the protection offered by the recommendation, unless we can persuade individuals that they are vulnerable to the health risk, they are unlikely to take protective action.

A median of 1.17 on a 7-point rating scale might be very low, however, this is not

remarkable when one compares perceptions of vulnerability with reported alcohol consumption. Respondents who perceived their vulnerability as 1 also reported a low alcohol consumption (M = 4 glasses a week) compared to respondents who perceived their vulnerability as higher (M = 15 glasses a week).

Chapter 6: The impact of fear appeals on information processing and persuasion: A meta-analysis5

For over fifty years fear-arousing communications have been used in health education campaigns, and their effectiveness has been studied extensively. Fear-arousing communications emphasize the negative consequences of health-impairing behaviors in order to motivate individuals to change these behaviors. Previous meta-analyses have shown that high fear messages in general are more effective than low fear messages in changing attitudes, intentions and behaviors (Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000). However, because these meta-analyses did not differentiate between different fear manipulations (severity of the consequences, personal vulnerability, or both components) nor between different types of fear messages (e.g., scary pictures; written information), such findings do not allow one to identify the specific kinds of information that make high fear messages particularly effective. Furthermore, since these meta-analytic reviews failed to include information processing variables in their analyses, they do not allow one to assess the validity of information processing theories of fear-arousing communications. The present meta-analysis was conducted to address these deficiencies. This meta-analysis will expand on previous meta-analyses in several ways. First, and foremost, we will not restrict ourselves to persuasion, but also assess the impact of fear-arousing communications on the processing of fear-arousing information. Second, we will analyze the independent as well as joint effects of vulnerability and severity information on information processing and persuasion. Third, we will assess the effects of different types of severity and vulnerability manipulations. Theories of fear-arousing communications Classic theories Research on fear-arousing communications started in the 1950s, theoretically guided by the drive-reduction model (Hovland, Janis & Kelley, 1953). The main assumption of the drive reduction model is that fear or emotional tension has the functional properties of a drive. A drive is a motivator for action, and the stronger the drive, the more motivational function it possesses. According to the drive reduction model, exposure to threatening information motivates individuals to search for responses that reduce a threat. Whatever response
5

This chapter is based on: De Hoog, N., Stroebe, W., & de Wit, J.B.F. The impact of

vulnerability to and severity of a health risk on processing and accepting fear-arousing communications: A meta-analysis. Manuscript submitted for publication.

110

Chapter 6

reduces the emotional tension produced by a fear appeal will be reinforced, and will become a habitual response. When a recommended protective action reduces fear, the recommendation is reinforced and can become part of ones permanent response repertoire. However, when a recommendation fails to reduce the fear level, because it is perceived as ineffective or as impossible to carry out, defensive reactions will occur that may serve to reduce fear. Especially high levels of fear are proposed to evoke defensive reactions that could undermine persuasion. Although an increasing level of emotional tension should lead to increased persuasion, at some point the emotional tension will reach a level at which the reassuring recommendation will not sufficiently reduce the tension. Residual emotional tension might give rise to defensive avoidance, causing a decrease in persuasion. This results in a curvilinear relationship between the level of emotional arousal and persuasion. The main contribution of the drive reduction model (Hovland et al., 1953) has been the focus on defensive reactions that individuals may display in reaction to a fear-arousing communication. In addition, a number of issues that are explicitly addressed in later theories are already implied by the drive reduction model. For example, the drive reduction model addresses the role of perceived efficacy of a recommendation and ones ability to carry it out, two concepts that featured prominently in later theories. The drive reduction model was followed by more cognitive models, beginning with the parallel response model (Leventhal, 1970). The parallel response model introduced threat appraisal as a mediator between an environmental threat and individual action, and abandoned the notion that fear is a necessary antecedent of the behavioral response to danger. According to the parallel response model, an environmental threat is cognitively evaluated and this appraisal can give rise to two parallel processes, namely fear control and danger control. Fear control involves responses that attempt to reduce the unpleasant affect evoked by a fear appeal, such as threat denial or avoidance. Action conducted in the service of fear control, therefore, have little or no effect on dealing with the threat. In contrast, danger control involves the selection of responses aimed at reducing or eliminating a threat. In danger control, actions are directed at coping with the threat. Therefore, the acceptance of a recommendation is proposed to depend mainly on the process of danger control. The importance of the parallel response model for the understanding of the processes mediating the impact of fear-arousing communications lies in its movement away from the notion of fear as the central explanatory concept. Instead, the parallel response model recognizes that individuals responses to fear-arousing communications involve adaptive behavior motivated by a desire to avert the anticipated danger and expectations as to how this can be done.

A meta-analysis

111

However, the parallel response model fails to specify the conditions that lead to each of the two processes of fear and danger control. It also does not tell us when and how these two processes interact or how people switch from one process to the other. The model only broadly categorizes emotional and cognitive responding as two distinct reactions to a health risk. Protection motivation theory (Rogers, 1983) elaborated the parallel response models process of danger control, by specifying the processes of cognitive appraisal that precede action tendencies. Rogers identified four message components of fear appeals that were assumed to motivate individuals to implement a protective action: (1) severity of a threat, (2) probability of the events occurrence, (3) response efficacy (i.e., the efficacy of a protective response in averting the threat), (4) and self-efficacy (i.e., the individuals ability to perform the protective response). Protection motivation theory differentiates between threat appraisal and coping appraisal processes. Threat appraisal is an appraisal of vulnerability to, and severity of, a threat and of the rewards associated with maladaptive behavior. Coping appraisal involves the appraisal of response efficacy, self-efficacy, and of the costs of adaptive behavior. These two forms of appraisal are assumed to interact with one another. Protection motivation will be strongest, when the threat is appraised as serious and coping is appraised as effective. Under these conditions, acceptance of the message recommending a protective action is most likely. Protection motivation theory has specified some of the important components of fear-arousing communications. Additionally, protection motivation theory has attempted to specify the cognitive appraisals of each of these components and the relationship between these cognitive appraisals and subsequent protective behavior. The main ideas from the parallel response model (Leventhal, 1970), protection motivation theory (Rogers, 1983), as well as the drive-reduction model (e.g., Hovland et al., 1953), were integrated by the extended parallel process model (Witte, 1992). According to the extended parallel process model, exposure to a fear-arousing communication initiates two appraisal processes, namely threat appraisal and coping appraisal. Individuals will appraise the threat portrayed in a fear-arousing communication, and the more they believe they are vulnerable to a serious threat, the more motivated they will be to engage in coping appraisal. If the threat is perceived as irrelevant or insignificant, the extended parallel process model proposes that there is no motivation to process a fear-arousing communication any further, and individuals will simply ignore the remainder of a communication. In contrast, when a health risk is believed to be serious and relevant, individuals will become scared, and their fear should motivate them to take action to reduce their fear. Perceived efficacy of the recommended action is proposed to determine

112

Chapter 6

whether individuals, who believe that they are vulnerable to a serious risk, will engage in danger control or fear control. Witte (1992) assumes that individuals will mainly engage in danger control when they perceive the recommended action as effective in reducing the threat, and they will mainly engage in fear control when they perceive the recommended action as ineffective, or when they feel unable to perform the recommended action. In that case defense motivation is elicited, which the extended parallel process model defines as individuals using denial, defensive avoidance or reactance to alleviate their fear. In sum, the extended parallel process model proposes that perceived threat determines the magnitude of the response to a fear-arousing communication, whereas perceived efficacy determines the direction of the response. The extended parallel process model has succeeded in integrating ideas of both protection motivation theory (Rogers, 1983) and the parallel response model (Leventhal, 1970), and has extended these ideas by identifying how threat appraisal and coping appraisal relate to each other, and by specifying the role of perceived fear in threat and coping appraisal. However, like the other fear appeal theories described so far, the extended parallel response model does not allow any predictions about information processing. Dual-process theories of fear-arousing communications With the rise of dual-process theories of persuasion in the nineteen eighties (e.g., (Chaiken, 1980; Eagly & Chaiken, 1993; Petty & Cacioppo, 1986) research on attitude change began to focus on the processing of the arguments contained in a communication as well as the outcomes of persuasion. It was, therefore, only a matter of time before dual-process theories were also applied to account for the impact of fear-arousing communications on persuasion (e.g., Gleicher & Petty, 1992; Liberman & Chaiken, 1992). According to dual-process theories, fear arousal is assumed to have two effects, namely act as a motivator to induce individuals to engage in intensive and thoughtful message processing, and to induce defense motivation, which will lead to biased message processing (Gleicher & Petty, 1992; Liberman & Chaiken, 1992). Whereas the unbiased or accuracy-motivated individual assesses the validity of attitude-relevant information in the interest of achieving a well-founded position, the processing goal of defense-motivated individuals is to confirm the validity of a preferred position and disconfirm the validity of non-preferred positions. Thus, defense-motivated individuals will process and perceive information in ways that best supports their own beliefs. This implies that information that is congruent with a particular preferred position will be judged as more valid than information that is incongruent with this position (Ditto & Lopez, 1992; Lord, Ross & Lepper, 1979).

A meta-analysis

113

The stage model of processing of fear-arousing communications Our stage model of processing of fear-arousing communications (Das, de Wit & Stroebe, 2003; De Hoog, Stroebe & de Wit, 2005a; Stroebe, 2000) integrates ideas from dual-process theories (e.g., Chaiken, 1980) with those of earlier theories of fear-arousing communications (Leventhal, 1970; Rogers, 1983; Witte, 1992). Our stage model shares with most previous fear appeal theories the assumption that individuals who are exposed to fear-arousing communications engage in two types of appraisal, namely appraisal of the threat and appraisal of coping strategies available for reducing the threat. These two appraisal processes parallel the two components of fear-arousing communications, namely the fear appeal, which emphasizes the severity of and vulnerability to a health risk, and the recommended protective action, which provides information on how to avoid a health risk.
Table 6.1. The impact of severity and vulnerability on processing mode and goal

Severity
Low Low
Heuristic processing Accuracy motivation

High
Systematic processing Accuracy motivation Systematic processing Defense motivation

Appraisal of the magnitude of a threat, based on the information in a fear appeal about the severity of and personal vulnerability to a risk, is assumed to determine both individual processing mode (i.e., depth of processing) and processing goal (i.e., accuracy or defense motivation; see Table 6.1). According to our stage model, if individuals feel vulnerable to a minor risk, the feeling of vulnerability should be sufficient motivation to invest effort into systematically processing the contents of the communication. Similarly, if individuals do not feel vulnerable, but a risk is depicted as severe, individuals are likely to invest effort in processing the contents of a communication, because it is useful to be well informed about a serious risk, even if the danger is not immanent. In contrast, if a risk is trivial and individuals do not feel vulnerable, they are unlikely to invest much effort in thinking about the contents of the communication and they might rely on heuristic processing modes. The situation we are most interested in is when individuals feel vulnerable to a severe risk. According to the stage model, if individuals are exposed to information about a severe risk and feel vulnerable, this will threaten their self-definitional

Vulnerability

High

Systematic processing Accuracy motivation

114

Chapter 6

belief that they are healthy, and arouse defense motivation as well as the motivation to carefully examine the presented information. We assume that processing will be systematic instead of heuristic, because any communication describing a serious personal threat is likely to require a thorough evaluation (Chaiken, Liberman & Eagly, 1989). We further assume that defense motivation induced by high threat perceptions manifests itself not only in avoidance reactions as previous fear appeal models have proposed (e.g., Witte, 1992), but in systematic processing that is biased. The direction of the bias aroused by defense motivation depends on the type of appraisal: We propose a negative processing bias in appraisal of the threat, and a positive processing bias in the appraisal of the action recommendation. In appraising a threat (Stage 1), defense-motivated individuals will attempt to minimize the threat by critically looking at the content of a fear appeal. Defense-motivated individuals will, for instance, engage in a biased search for inconsistencies, and their evaluation of evidence will be biased in the direction of the preferred conclusion (Ditto & Lopez, 1992; Liberman & Chaiken, 1992; Sherman, Nelson & Steele, 2000). If this strategy proves unsuccessful, because even biased processing is constrained by evidence and rules of inference (Kunda, 1987), individuals will have to accept that they are at risk. In this case, the processing of the action recommendation (Stage 2) will be biased as well, but in a positive direction. The processing goal of defensemotivated individuals will now be to find the recommended protective action effective, because then they can feel save. The processing will involve attempts to make the recommendation appear effective by means of a biased search for, and evaluation of, information supporting the effectiveness of the protective action. In other words, defense motivation will lead to a positive bias in the processing of the action recommendation, and consequently heighten the motivation to accept a solution to a particular threat, regardless of the quality of the arguments supporting this recommendation. This latter prediction of the stage model differs from predictions of most fear appeal theories that assume an interaction between perceptions of threat and efficacy (e.g., Hovland et al., 1953; Rogers, 1983; Witte, 1992). Most fear appeal theories propose that a severe threat will motivate acceptance of the recommended action only if that action is perceived as effective in reducing the threat. The stage model, however, assumes that any increase in threat will increase defense motivation. The stronger the defense motivation, the more individuals will be motivated to perceive any recommendation as effective, as long as it is at least somewhat plausible, because engaging in a recommended action will make these individuals feel save. Even though threat appraisal and coping appraisal are sequential processes, the stage model, in line with cognitive stress theory (e.g., Lazarus & Folkman, 1984), assumes that the outcome of the coping appraisal can lead to a

A meta-analysis

115

reappraisal of the threat. More specifically, the appraisal that the recommended action promises to protect the individual against the health threat, will reduce defense motivation and allow individuals to admit to themselves that the threat is severe and that they are at risk. Thus, we would expect less defensive bias in the evaluation of a threat, if the communication also recommends a viable protective action than if such a recommendation were absent. A further prediction, which differentiates our stage model from most other fear appeal theories, is the assumption that fear appeals will have different effects on different outcome measures. Whereas most other fear appeal theories make the same predictions for all measures of persuasion, our stage model draws a distinction between measures of persuasion that assess attitudes and those that measure behavioral intentions or behavior. We assume that attitudes about a protective action are based mainly on an objective evaluation of the information provided in the communication. Regardless of whether or not individuals feel vulnerable, a severe health risk will lead to a positive attitude towards a protective action if the recommendation is supported by strong arguments and to a negative (or less positive) attitude, if it is supported by weak arguments. Thus, even if individuals living in Europe may not feel at risk to contract some deadly virus that is epidemic only in Africa, they might have a positive attitude towards Africans being inoculated against this virus, if the inoculation is known to be effective. But unless they have plans to travel to Africa, they are unlikely to get themselves inoculated. Because when it comes to measures of persuasion that have behavioral implications (i.e., intention and behavior), individuals will only be motivated to engage in the protective action if they actually feel vulnerable. After all, why should one invest effort into avoiding a risk, if one does not feel personally at risk? Previous meta-analyses of fear appeal studies To date, several meta-analyses have been performed on studies assessing the effects of fear-arousing communications (Boster & Mongeau, 1984; Mongeau, 1998; Sutton, 1980; Witte & Allen, 2000), as well as regarding the effectiveness of protection motivation theory (Floyd, Prentice-Dunn, & Rogers, 2000; Milne, Sheeran, & Orbell, 2000). Boster and Mongeau (1984) examined the influence of different levels of fear in a fear-arousing communication on perceived fear (the manipulation check), attitudes towards the recommendation, and behavior. They found medium strong effects of manipulation of fear level on reported fear (r= .36), and small effects of manipulation of fear level on attitudes (r= .21), and behavior (r= .10). In other words, high fear messages resulted in more fear, and more attitude and behavior change than low fear messages. Mongeaus update (1998) of this meta-analysis resulted in similar effects.

116

Chapter 6

The meta-analysis performed by Sutton (1980) focused on the impact of manipulation of fear level on intentions and behavior, in which z-scores were used as the meta-analytic test. Sutton reported significant positive effects of manipulation of fear level on intentions to engage in a recommended protective action, as well as on behavior. In concurrence with Boster and Mongeau (1984; Mongeau, 1998), Sutton found that higher levels of fear in a fear-arousing communication produced more intention and behavior change than lower levels. Witte and Allen (2000) expanded on these previous meta-analyses by not only focusing on the influence of fear level, but by also examining the mediating effects of perceived fear, perceived severity, perceived vulnerability, perceived response efficacy and perceived self-efficacy on attitudes, intentions and behavior. They found medium to strong effects of manipulation of fear level on perceptions of fear, severity, vulnerability, response efficacy and selfefficacy. In addition, they found small significant effects of perceived fear, perceived severity, perceived vulnerability, perceived response efficacy, and perceived self-efficacy on attitudes (all rs between .12 and .15), intentions (all rs between .13 and .17), and behavior (all rs between .13 and .16). In other words higher perceptions of fear, severity, vulnerability, response efficacy and self-efficacy all resulted in more attitude, intention and behavior change. Two meta-analyses related to fear appeals examined the effects of protection motivation theory components on persuasion (Floyd et al., 2000; Milne et al., 2000). Although these meta-analyses included many studies that did not use fear-arousing communications, or did not include any communications at all, the effects they found for severity of and vulnerability to a threat on measures of persuasion were comparable to those found in fear appeal meta-analyses. Significant positive effects in the small range were found in these meta-analyses for both severity and vulnerability on intention and behavior. In addition, both meta-analyses found significant positive effects in the medium range of response efficacy and self-efficacy on intention and behavior. In general, previous meta-analyses have shown that high fear messages (i.e., fear appeals that use higher levels of manipulated fear appeal components) produce more fear, and influence attitudes, intention and behavior more than low fear messages. However, this provides us with little information about which specific components and types of information in a fear message are most effective in changing attitudes, intentions and behavior, because of: (1) the wide variation in fear levels and fear types across studies; (2) the wide variation across studies in factors that were manipulated in a fear message; and (3) the dominant role of severity information compared to the relatively neglected role of perceived vulnerability in fear-arousing communications. A further limitation of previous meta-analysis has been their

A meta-analysis

117

failure to pay attention to the processing of fear-arousing communications. None of these meta-analyses focused on the effects of components of a feararousing communication on cognitions elicited by a fear message, and on the possible role of these factors in mediating the effects of severity and vulnerability on measures of persuasion. Overview present meta-analysis The present meta-analysis updates and expands on previous meta-analyses by focusing on the processing and acceptance of fear-arousing communications. We will compare the results of this meta-analysis to predictions of previous fear appeal theories, as well as assess support for our stage model of the processing of fear-arousing communications (e.g., De Hoog et al., 2005a). We will examine the separate effects of severity and vulnerability on measures of persuasion, as well as on affect and cognitions. We will also assess the extent to which the effects of vulnerability and severity on persuasion are mediated by affect and cognitions. We will not only examine the overall effects of severity and vulnerability, but also the effects of different levels and different types of severity and vulnerability. In particular, we will compare the effects of studies that manipulated different levels of severity information (e.g., very strong or mild), and compare the effects of studies that manipulated different types of severity information (e.g., fear-arousing images or written information only). For vulnerability, we will examine the use of different types of variations in personal vulnerability by comparing studies that manipulated vulnerability with those that relied on respondents self-measures of perceived vulnerability, and those that compared pre-defined vulnerable and non-vulnerable groups (i.e., smokers and non-smokers). In addition, we will examine the effects of other factors that have been manipulated in fear appeal studies alongside severity and vulnerability (e.g., argument quality, response efficacy, and message source). Furthermore, we will test a number of moderator models by looking at study characteristics than can account for heterogeneity in effect sizes. Method Literature search An extensive search to identify all empirical fear appeal studies was conducted. Firstly, previous reviews and meta-analyses of fear appeal research (Boster & Mongeau, 1984; Floyd et al., 2000; Higbee, 1969; Milne et al., 2000; Mongeau, 1998; Rogers & Prentice-Dunn, 1997; Sutton, 1982; Witte & Allen, 2000) were searched for relevant publications. Secondly, computer databases (e.g., PsychInfo, Picarta, Pubmed) were searched for fear appeal studies using the

118

Chapter 6

following search terms: fear appeal(s), threat appeal(s), fear message(s), threatening message(s), fear communication(s), fear-arousing communication(s), threat communication(s), threatening communication(s), risk communication(s), fear and persuasion, fear and attitude, health threat(s) and persuasion, fear-arousing/fear arousal and persuasion, and fear and persuasive communication. This electronic search was performed in November 2003, repeated and updated in May 2004. Lastly, the references cited in the identified studies were reviewed for relevant publications. In total this literature search resulted in 187 references to publications describing empirical research using a fear appeal. Inclusion criteria Of the 187 publications found that reported empirical fear appeal research, only 105 studies (in 95 publications) could be included in this meta-analysis. Our criteria for inclusion were: (1) the studies had to be written in the English language; (2) the studies had to have been published (i.e. all unpublished doctoral theses were not retrieved); (3) the same data should not have been published repeatedly (i.e., only the first publication of a study was included); (4) the studies had to have as the independent variable a manipulation of severity or a manipulation of vulnerability; (5) the studies should not have confounded factors in a manipulation (e.g., manipulation of both vulnerability and severity in the same condition); and (6) the study had to have as dependent variables at least one outcome measure relevant to our stage model (e.g., attitude, intention, cognitive responses). Coding All studies included in this meta-analysis were coded for the following characteristics: (a) publication year; (b) country of origin; (c) study topic; (d) type of media used; (e) theoretical framework; (f) number of respondents; (g) type of respondents; (h) experimental design; (i) inclusion of a specific recommendation; (j) order of presentation of the fear appeal and recommendation; (k) number of exposures to the fear-arousing communication; (l) inclusion of manipulation checks; (m) use of pre-measures; and (n) time of post-measures. In addition, we coded the independent and dependent variables used in each study. All independent and dependent variables in a study were coded using definitions derived from our stage model. This was done because variable names are sometimes interchangeably used or misused in some, especially older, studies. For example, when a study reported to measure attitudes by asking respondents whether they intended to follow a recommendation, according to our definitions, based on presently commonly accepted operationalisations, that question measured intention and was coded as such.

A meta-analysis

119

When a study included as independent variable a manipulation/variation of severity or vulnerability, the level (i.e., low, mild or high) and type (i.e., written or scary images) of the severity manipulation and the type of vulnerability variation (i.e., manipulation, self-measures or existing vulnerable and non-vulnerable groups) was coded as well. Two judges independently coded the parameters of each study, and resolved discrepancies through discussion. Disagreements were few because of the predefined operationalisations of all relevant dependent and independent variables. Furthermore, although the labels attached to dependent measures sometimes differed across studies, the type of questions for dependent measures used across fear appeal studies tended to be fairly similar (e.g., most studies assessed intentions by asking respondents whether they intended to engage in a certain behavior). Statistical procedures Effect sizes were calculated and analyzed using Wilsons Effect size determination program (Lipsey & Wilson, 2001), and the software program META (Schwarzer, 1989). Effect sizes were extracted from each study for: (1) effects of vulnerability on measures of persuasion (attitudes, intention, and behavior); (2) effects of vulnerability on potential mediating variables (fear, negative affect, perceived threat, minimizing thoughts about the fear appeal, and positive thoughts about the action recommendation); (3) effects of severity on all dependent variables, and (4) effects of argument quality, response efficacy, and source manipulations on all dependent variables. We used the d statistic (Cohen, 1977), the standardized mean difference between groups, as the effect size measure. In most cases the effect size was calculated using the value of the test statistic and the number of participants. In other cases means and SDs, proportions, or p values were used to calculate effect sizes. In some cases test statistics (i.e., F-values) required for calculating effect sizes were extracted from two or three-way analysis of variance (Cortina & Nouri, 2000). We calculated one effect size per construct per study in order to prevent overweighting of effect sizes. When studies included more than one measure of the same construct the effects of these different measures were combined into one average effect size. When studies merely reported the results as nonsignificant, without enough summary statistics to calculate an effect size, the effect was assigned a one-tailed probability of .5, equal to an effect size of zero (Rosenthal, 1991). This procedure is conservative and tends to underestimate the effect, however, disregarding non-significant findings results in an inflated estimate. For each effect to be tested (e.g., the effect of vulnerability on intention) the effect sizes of each study measuring this effect were combined into a mean effect size d. The d values were corrected for small-sample size

120

Chapter 6

bias (Hedges & Olkin, 1985). Homogeneity statistics were calculated to test whether each set of ds shared a common effect size. Both categorical and continuous models were used to test our predicted models, as well as to look for additional moderator variables by use of general study characteristics. Categorical models show whether effect sizes are homogeneous within established subgroups. Moderating effects of categorical variables were tested by dividing the studies into classes based on their characteristics and testing for homogeneity of effect sizes between and within classes. Weighted least squares linear regression was used to test for the moderating effects of continuous variables with each effect size weighted by the reciprocal of its variance. Results Sample characteristics The main study characteristics can be found in Table 6.2. Studies were published between 1954 and 2005, with one study published in the 1950s, 25 studies published in the 1970s, and in all other decades around 18-20% of the studies were published. The majority of the studies were performed in the United States (78%) and most studies used undergraduate students (57%) as respondents. The studies employed a wide range of different topics, but most fell in the category of health and health related behavior (85%). Main topics were smoking (15%), dental hygiene (13%), cancer (7%), and alcohol consumption (6%) A variety of media was used in the fear-arousing communication, including written information, audio, pictures and films. The studies were based on a number of different theoretical frameworks, of which the most popular were protection motivation theory (15%) and the drivereduction model (13%). However, almost half of the studies in this metaanalysis (46%) used more than one theory or were not based on any theory in particular. With regard to method related study characteristics, it was found that the majority of studies (76%) included a specific recommendation in their feararousing communication, which was almost always presented after respondents had been exposed to severity and/or vulnerability information. In addition, respondents were generally only exposed to a message once (96%), and most studies that manipulated factors used manipulation checks (90%). In half of the studies (52%) pre-measures were used, and post-measures were usually answered immediately after exposure to the message, although in 18% of the studies post-measures were also administered a second or third time, and in 11% of the studies post-measures were only assessed at a later time.

A meta-analysis

121

Table 6.2. Summary of study characteristics Variable and class Study year 1950-1959 1960-1969 1970-1979 1980-1989 1990-1999 2000-2005 Country United States Australia Canada Germany Great Britain Israel Netherlands Study topic AIDS/STDs Alcohol consumption Blood donation Caffeine consumption Cancer Crime prevention Dental hygiene Dieting/exercise Driving Drug use Earthquake preparedness Energy crisis Fallout shelters Fictitious disease Green peace Heart disease HMO Hypoglycemia (Life) insurance Mumps Nuclear war 2 2 2 4 1 2 1 2 2 2 3 3 6 1 4 7 3 14 3 7 4 82 1 4 2 4 1 11 1 19 25 20 21 19 N Variable and class Type of media used Written (paper/computer) Tape Film Commercials/PSAs Paper & tape Paper & pictures Paper & film Slides & tape Slides & presentation Theoretical framework Drive reduction model Parallel response model Protection motivation theory Extended parallel response model Dual-process theories Stage model Other Non-specific Type of respondents Undergraduate students High school students General public Other specific groups Inclusion of a specific recommendation yes no Order of presentation Fear appeal-recommendation Recommendation-fear appeal Fear appeal only Number of exposures Once More than once Inclusion of manipulation checks yes no not relevant 83 9 13 101 4 79 1 25 80 25 60 14 20 11 14 3 16 1 6 9 8 48 52 5 16 5 3 2 11 9 2 N

Note Non-specific refers to studies that used no specific theoretical framework as well as studies that used more than one theory; Studies either included no relevant manipulations or compared a fear message with a control group.

122

Chapter 6

Table 6.2. Summary of study characteristics (continued) Variable and class Study topic Obesity Osteoporosis Population control Repetitive strain injury Smoking Stress Tetanus Water conservation 2 2 1 2 16 3 3 1 N Variable and class Use of pre-measures yes no Time of post-measures Immediate Immediate and later Only at a later date 75 19 11 55 50 N

Note Non-specific refers to studies that used no specific theoretical framework as well as studies that used more than one theory; Studies either included no relevant manipulations or compared a fear message with a control group.

The present meta-analysis included 105 studies with a total of 17,919 participants. Most of the studies in this meta-analysis included variations of severity (N= 89), while only less than half of the studies varied vulnerability (N= 43). The proportion of studies that varied both severity and vulnerability in the same study is only 26% (N= 27). In reality the number of studies manipulating both vulnerability and severity in the same experiment is higher (N= 58), but the other studies all confounded manipulations of vulnerability and severity in their design, and were, therefore, excluded from our metaanalysis. Only 21% (N= 19) of the studies that included variations of severity compared the severity manipulation to a control group, while the majority of studies (79%) compared different levels of severity. Studies in this metaanalysis where almost equally divided into studies that used scary images in their severity manipulation (43%), and studies that used only written information (57%). Of the studies that included variations of vulnerability, 56% (N= 24) manipulated vulnerability, 12% (N= 5) relied on self-measures of perceived vulnerability, and 33% (N= 14) compared existing vulnerable and non-vulnerable groups. As for the dependent variables, 44% (N= 46) of the studies reported measures of attitudes, 62% (N= 65) of the studies reported measures of intention, and 46% (N= 48) reported behavioral measures. In addition, studies included a host of additional dependent variables (not shown in Table 3): measures of fear (46%), negative affect (29%), perceived threat (9%), thoughts about the fear appeal (8%), and thoughts about the recommendation (7%).

A meta-analysis

123

Meta-analytic findings In this meta-analysis we examined: (1) the effects of vulnerability; (2) the effects of severity; (3) the effects of vulnerability and severity in studies that varied both factors; (4) the effects of other manipulated factors; and (5) moderator models of study characteristics. Most of the effects found in this meta-analysis were consistent with the predictions of our stage model, and most of the effect sizes were in the small to medium range. Cohen specified criteria for interpreting the magnitude of effect size d as .20, .50 and .80 for small, medium and large effects respectively (Cohen, 1977). 1. Effects of vulnerability The effects of vulnerability were analyzed for: (a) measures of persuasion (attitudes, intention and behavior); (b) potential mediating variables (fear, negative affect, perceived threat, minimizing thoughts, and positive thoughts); (c) mediation of the effects on measures of persuasion by the potential mediating variables; and (d) all variables mentioned by comparing types of vulnerability: manipulated vulnerability, self-measured perceived vulnerability, and existing vulnerable and non-vulnerable groups.
Table 6.4. Effect sizes for vulnerability and severity on attitude, intention and behavior Variable Attitude Vulnerability Severity Intention Vulnerability Severity Behavior Vulnerability Severity 19 41 2424 6077 +.4122*** +.1921*** +.2128 to +.6125 +.0925 to +.2917 106.55**** 145.38**** 32 55 4376 8814 +.3271**** +.2339*** +.1731 to +.4811 +.1597 to +.3082 199.16**** 162.15**** 19 40 2550 6225 -.0825 +.2608**** -.2794 to +.1114 +.1505 to +.3711 109.10**** 168.81**** N Nk d 95% CI Qw

Note N = number of studies; Nk = number of respondents; d = mean effect size; CI = confidence interval; Qw = within class homogeneity statistic; * p < .05, ** p < .01, *** p < .001, **** p < .0001.

Effects of vulnerability on measures of persuasion Effects of vulnerability on measures of persuasion were as expected (see Table 6.4). The effect of vulnerability on attitudes towards the recommendation was non-significant (d= -.08, p = .41), which is consistent with our assumption that attitudes are mainly objective evaluations of the information provided in a communication and are unaffected by feelings of vulnerability. In contrast, positive significant effects in the small range were found for intention (d= .33, p < .0001) and behavior (d= .41, p < .001). This confirms our prediction that greater feelings of vulnerability results in stronger intentions and more

124

Chapter 6

behavior change. For all three effects, the homogeneity statistics were very significant, indicating that the study effect sizes were, as expected, heterogeneous. Effects of vulnerability on potential mediating variables Analyses of the effect of vulnerability on potential mediating variables (i.e., affect and cognitions) were mostly in the expected direction and in the medium range (see Table 6.5). A significant positive effect was found on fear (d= .32, p < .0001) and on negative affect (d= .60, p < .0001). As expected, feeling vulnerable to a depicted risk aroused negative emotions and fear. The effect on negative affect was stronger than the effect on fear, which is not surprising, as fear-arousing communications tend to arouse more general negative emotions rather than actually scaring individuals.
Table 6.5. Effect sizes of vulnerability and severity on potential mediating variables Dependent variable Negative affect Vulnerability Severity Fear Vulnerability Severity Perceived threat Vulnerability Severity Thoughts fear appeal Vulnerability Severity Thoughts recommendation Vulnerability Severity 6 4 935 +.5452**** +.4067 to +.6837 475 +.0383 -.2056 to +.2822 5.33 5.13 6 5 812 +.4907*** +.2470 to +.7345 14.24* 2.57 445 +.4572**** +.2685 to +.6458 8 1036 +.0935 5 505 +.4000** -.2210 to +.4080 +.1205 to +.6795 43.50**** 9.79* 23 3499 +.3230**** +.2150 to +.4310 51.64*** 39 6615 +.5868**** +.4546 to +.7201 223.78**** 13 1747 +.6048**** +.4541 to +.7554 25.31* 26 4373 +.5049**** +.3615 to +.6483 123.11**** N Nk d 95% CI Qw

Note N = number of studies; Nk = number of respondents; d = mean effect size; CI = confidence interval; Qw = homogeneity statistic; * p < .05, ** p < .01, *** p < .001, **** p < .0001.

Effects of vulnerability on perceptions of threat were consistent with our predictions. The effect of vulnerability on perceptions of threat was nonsignificant (d= .09, p = .56). However, this failure to find a significant effect appears to have been due to the fact that the impact of vulnerability on perceived threat was moderated by the inclusion of an action recommendation. When studies followed the health threat with the recommendation of a

A meta-analysis

125

protective action, inducing vulnerability increased the perception of threat (d= .22, p = .18). However, in studies which failed to recommend a protective action, inducing vulnerability reduced perceived threat (d= -.28, p = .33). Unfortunately, both of the effects were non-significant, as was the difference between these effects. However, at least descriptively, these differences are consistent with our assumption that individuals, who are made to feel vulnerable to a health risk defensively minimize the threat, unless a protective action is specifically recommended. Regarding thoughts elicited by the fear-arousing communication, we found significant positive effects for vulnerability on minimizing thoughts about the fear appeal (d= .49, p < .001), and on positive thoughts about the recommendation (d= .54, p < .0001). This confirms the expectation derived from our stage model that minimizing thoughts about a fear appeal and positive thoughts about a recommended action are elicited by defense motivation, which is induced by feelings of vulnerability. Because most studies that assessed minimizing thoughts and offered a specific recommendation assessed minimizing thoughts before the recommendation was offered, no analyses on differences in minimizing thoughts depending on whether or not a recommendation was offered could be performed. As expected, most effect sizes were heterogeneous, as indicated by significant homogeneity statistics. Only the effect on positive thoughts about the recommendation was homogeneous. Mediation of effects of vulnerability on intention and behavior We next assessed possible mediation of the effects of vulnerability on intention and behavior by the effects of vulnerability on affect and cognitions. We expected the effect of vulnerability on intention to be mediated by the negative emotions aroused by the fear appeal and the thoughts about the fear-arousing recommendation. In addition, we expected the effect of vulnerability on behavior to be mediated by intentions. We tested continuous models in which predicted intention and behavior from the effects of vulnerability on fear, negative affect, perceived threat, minimizing thoughts about the fear appeal and positive thoughts about the action recommendation. As can be seen in Table 6.6, both negative affect and minimizing thoughts were significant predictors of intention. The stronger the effect of vulnerability on negative affect and minimizing thoughts about the fear appeal, respectively, the greater the intention to engage in a protective action. When entering both negative affect and minimizing thoughts as predictors at the same time, both predictors were no longer significant. However, there were only three studies in this analysis and, therefore, no conclusions can be drawn from this.

126

Chapter 6

Table 6.6. Continuous models predicting intentions and behavior in studies varying vulnerability Predictor Intention Negative affect Fear Perceived threat Minimizing thoughts Positive thoughts Behavior Negative affect Fear Perceived threat Positive thoughts Intention
* p < .05, ** p < .01, *** p < .001.

N 11 20 7 4 6 8 9 6 5 16

b 1.24*** .56 .34 .94** .43 1.13*** .97** -.31 -.01 .79****

B .81 .30 .46 .89 .28 .82 .76 -.47 -.01 .93

There were not enough studies that measured both minimizing thoughts and behavior, and, therefore, no mediation analyses were performed with minimizing thoughts. Negative affect, fear, and intention were found to be significant predictors of behavior. The stronger the effects of vulnerability on fear, negative affect and intentions, respectively, the more behavior change. When entering intention and fear, and intention and negative affect as predictors at the same time, the effects of fear and negative affect were no longer significant, whereas the effect of intentions was. This confirms our assumption that the effects of vulnerability on behavior are mediated by the effect of vulnerability on intentions. Type of vulnerability Specific analyses of the effects of vulnerability on measures of persuasion, as well as on potential mediating variables, were performed by dividing the studies that varied vulnerability into those studies that: (1) manipulated vulnerability; (2) relied on self-measures of perceived vulnerability; and (3) compared existing vulnerable and non-vulnerable groups (e.g., smokers vs. non-smokers). As can be seen in Table 6.7 the effects on all measures of persuasion differed according to vulnerability type. Effects on attitudes were small and non-significant for manipulated vulnerability (d= .10, p = .20), whereas a significant positive effect was obtained for studies that used self-measures of vulnerability (d= .30, p < .05). A significant negative effect was found for studies that compared existing vulnerable and non-vulnerable groups (d= -.56, p < .0001). The negative effect found for existing groups on attitudes can be

A meta-analysis

127

explained by looking at the type of groups that were compared. Five out of the seven studies that compared vulnerable and non-vulnerable groups used smokers as their respondents. It is not surprising that when asked about their attitudes towards giving up smoking, non-smokers are much more positive about this than smokers, because non-smokers only see the positive aspects of quitting smoking, while smokers also see the negative aspects.
Table 6.7. Effect sizes for vulnerability on all dependent variables by type of vulnerability Variable Attitude Manipulations Self-measures Existing groups Intention Manipulations Self-measures Existing groups Behavior Manipulations Self-measures Existing groups Negative affect Manipulations Self-measures Existing groups Fear Manipulations Self-measures Existing groups Thoughts fear appeal Manipulation and self-measures Existing groups
Note

Qb 38.84****

N 9 3 7

Nk

95% CI -.056 to +.263 +.008 to +.583 -.764 to -.355 +.226 to +.619 +.160 to +.852 -.511 to +.427 +.325 to +.748 -.196 to +.920 -.642 to +.293 +.477 to +.805 +.525 to +1.033 -.040 to +.561 +.174 to +.479 +.069 to +.741 +.100 to +.487

Qw 6.02 1.14 9.33 10.72 6.31 15.44 14.61 3.02 .56 7.17 10.85 2.78 14.89 1.89 4.54

1389 +.1037a 392 769 +.2954a* -.5597b****

7.28* 18 2562 +.4224a**** 6 8 7.40* 14 1838 +.5367a**** 2 3 7.02* 7 3 3 .32 12 1893 +.3262a**** 3 8 12.39*** 3 3 353 459 +.7698**** +.2594** +.554 to +.986 +.075 to +.444 .40 1.45 276 +.4049a* 1330 +.2938a** 966 390 391 +.6410a**** +.7778a**** +.2604b 235 351 +.3616ab -.1742b 714 +.5060a** 1037 -.0265b

Qb = between-classes homogeneity statistic; N = number of studies; Nk = number of respondents; d = mean

effect size; CI = confidence interval; Qw = within classes homogeneity statistic; Differences between mean effect sizes per dependent variable that do not have a subscript in common are significantly different at least p < .05; * p < .05, ** p < .01, *** p < .001, **** p < .0001.

The between class homogeneity statistic was significant, indicating that the effect sizes for type of vulnerability were significantly different. Comparisons of the vulnerability type groups confirmed our prediction that the effects of

128

Chapter 6

manipulated and perceived vulnerability did not differ significantly, while the effects were significantly different between manipulated/perceived vulnerability and studies that used existing groups. For intentions, significant positive effects were found for both studies that manipulated vulnerability (d= .42, p < .0001) and studies that used selfmeasures (d= .51, p < .01), whereas the effect of studies that used existing groups was non-significant (d= -.03, p = .86). The between class homogeneity statistic was significant, and comparisons of the vulnerability type groups indicated, as predicted, that the effects of manipulated and perceived vulnerability did not differ significantly, while effect sizes were significantly different between manipulated/perceived vulnerability and studies using existing groups. Studies that manipulated vulnerability had significant positive effects on behavior (d= .54, p < .0001), while positive yet non-significant effects were found for studies that used self-measures (d= .36, p = .20), and those that used existing groups (d= .17, p = .47). Again, the between class homogeneity statistic was significant, and comparisons of vulnerability types indicated that the effects of manipulated and perceived vulnerability did not differ significantly from each other. By contrast, effect sizes were significantly different between manipulated vulnerability and studies using existing groups, but not between perceived vulnerability and existing groups. However, there were only five studies in the latter two groups combined, and, therefore, no conclusions can be drawn from this result. Within class homogeneity statistics were non-significant for all the effects tested, apart from the effect of existing groups on intention. This implies that, for the most part, dividing vulnerability studies in these three types of subgroups is a good classification as all effect sizes are now homogeneous. More importantly, the significantly different effects obtained for manipulated and perceived vulnerability versus existing vulnerable groups on measures on persuasion shows that targeting existing vulnerable groups is not very effective in influencing intentions and behavior. This confirms our assumption that it is a matter of feeling vulnerable, instead of belonging to a vulnerable or risk group, which is a precondition for intention and behavior change. Effects of vulnerability type on affect and cognitions Effects on fear aroused by the communications were similar for the different vulnerability types, and were all significantly positive in the small range (Table 6.7). However, there were significant differences according to vulnerability type on experienced negative affect. Studies that manipulated vulnerability (d= .64, p < .0001) and studies that used self-measures (d= .78, p < .0001) had significant positive effects that different significantly from the effect of studies using existing groups that was non-significant (d=. 26, p = .09). This

A meta-analysis

129

difference indicates that, although in all three subgroups negative emotions are elicited by exposure to a fear appeal, it seems that when one feels vulnerable instead of when one belongs to a vulnerable group, much more negative emotions are aroused. This is because belonging to a vulnerable or risk group does not necessarily mean that one feels vulnerable, and feeling vulnerable to a risk elicits more negative emotions than being exposed to a risk one does not feel vulnerable to. Positive thoughts about the recommendation and perceptions of threat were mainly measured in studies that manipulated vulnerability and, therefore, no additional analysis on these variables were performed. Furthermore, because there were not enough studies in the perceived vulnerability subgroup, studies that used manipulated and perceived vulnerability were combined and compared to existing groups for thoughts about the fear appeal. Analysis for minimizing thoughts about the fear appeal revealed that there were significant differences between studies that used manipulated/perceived vulnerability (d= .77, p < .001) and studies that used existing groups (d= .26, p < .01). This confirms our expectation that minimizing thoughts are mainly elicited by feelings of vulnerability, and reflects defense motivation. Within class homogeneity statistics were non-significant for all groups on all dependent measures, which implies that dividing vulnerability studies in these three types is a good division as all effect sizes in each subgroup are now homogeneous. 2. Effects of severity The effects of severity were analyzed for: (a) measures of persuasion (attitudes, intention and behavior); (b) potential mediating variables (fear, negative affect, perceived threat, minimizing thoughts and positive thoughts); (c) measures of persuasion by comparing levels of severity: strong vs. mild vs. control; and (d) measures of persuasion by comparing types of severity: scary images and written information. Effects of severity on measures of persuasion Effects of severity on measures of persuasion were in the expected direction (see Table 6.4). Significant positive effects were found for all three persuasion measures: attitudes, (d= .26, p < .0001), intention (d= .23, p < .0001) and behavior (d= .19, p < .001). As expected, severity had positive effects on attitudes, intention, and behavior. In addition, as predicted, the effects of severity on attitudes appeared stronger than of vulnerability, while the effects of severity on intentions and behavior appeared weaker. Effects of severity on potential mediating variables Most effects of severity on potential mediating variables (i.e., affect and cognitions) were in the expected direction and in the medium range (Table

130

Chapter 6

6.5). Significant positive effects were found for severity on fear (d= .59, p < .0001) and negative affect (d= .50, p < .0001). This confirms our prediction that exposure to a fear-arousing communication high in severity elicits more negative emotions than a communication with lower levels of severity. We also found a significant positive effect of severity on perceived threat (d= .40, p < .01), which confirmed our prediction that perceptions of threat is increased when the depicted risk is described as more severe. Inconsistent with our predictions was the significant positive effect of severity on minimizing thoughts about the fear appeal (d= .46, p < .0001). We had assumed that the thoughts about a fear-arousing communication would mainly be elicited by feelings of vulnerability, and not severity. This prediction was confirmed, however, by the non-significant effect of severity on positive thoughts about the action recommendation (d= .04, p = .76), which we initially had predicted to be mainly determined by a positive processing bias induced by feelings of vulnerability. As expected, the study effect sizes for negative affect, fear and perceived threat were heterogeneous, as indicated by significant homogeneity statistics. However, the effects on minimizing thoughts about the fear appeal and positive thoughts about the recommendation were found to be homogeneous. Severity levels Next, effects of severity on measures of persuasion were reanalyzed by dividing the studies into groups that varied different severity levels. We divided studies into those studies that compared: (1) high levels of severity with lower levels; (2) mild levels of severity with lower levels; and (3) any severity level to a control group. A significant positive effect in the small range was found on attitudes for studies that compared higher severity levels to lower levels (d= .27, p < .05) (Table 6.8). The effect for studies that compared mild severity levels to lower levels was small and non-significant (d= .15, p =. 06), while the effect for studies that compared severity levels to a control group was significantly positive and in the medium range (d= .50, p < .0001). However, the between class homogeneity statistic indicated that effects for all three groups did not differ significantly. Comparisons of the severity level groups indicated that the only significant difference was between studies comparing a mild severity level to a lower severity level and studies comparing any severity level to a control group. Regarding intentions, no significant differences were found between groups of studies. Studies comparing higher severity levels to lower levels (d= .23, p < .001), studies comparing mild severity levels to lower levels (d= .24, p < .0001) and studies comparing severity levels to a control group (d= .25, p < .01) had similar positive significant effects on intention. For behavior, the

A meta-analysis

131

effects of studies comparing higher severity levels to lower levels (d= .24, p < .01) and studies comparing mild severity levels to lower levels (d= .21, p < .05) were small and significant, while the effect of studies comparing severity levels to a control group was small and non-significant (d= .14, p = .16). However, no significant differences between these effects were found.
Table 6.8. Effect sizes for severity on measures of persuasion by severity level: higher vs. lower severity, mild vs. lower severity, and severity vs. control Variable Attitude High severity vs. lower Mild severity vs. lower Severity vs. control Intention High severity vs. lower Mild severity vs. lower Severity vs. control Behavior High severity vs. lower Mild severity vs. lower Severity vs. control
Note Qb = between-classes homogeneity statistic; N = number of studies; Nk = number of respondents; d = mean effect size; CI = confidence interval; Qw = within classes homogeneity statistic; Differences between mean effect sizes per dependent variable that do not have a subscript in common are significantly different at least p < .05; * p < .05, ** p < .01, *** p < .001, **** p < .0001.

Qb 5.75

Nk

d +.2671ab* +.1529a

95% CI +.0609 to +.4733 -.0022 to +.30.80

Qw 10.97 20.69 15.52*

12 1396 19 3599 1220 9 .04 19 3019 24 4169 11 1566

+.4952b**** +.2623 to +.7281

+.2314a*** +.2542a**

+.1029 to +.3599 +.0809 to +.4274

8.51 30.18 10.88

+.2375a**** +.1228 to +.3521

.66

15 2045 11 1848 12 1666

+.2433a** +.2134a* +.1371a

+.0669 to +.4198 +.0104 to +.4128 -.0560 to +.3302

30.32* 5.51 7.22

Homogeneity statistics indicated that, apart from the effect on attitudes for studies comparing severity levels to a control group on attitudes, and the effect on behavior for studies comparing high levels of severity to lower levels, the effect sizes of the studies in all groups are homogeneous. To sum up the analysis of severity levels, it seems to make no difference what level of severity is used, as long as some severity information is given. Apart from a stronger effect on attitudes when severity information is compared to a control group than when it is compared to lower levels of severity, no other significant differences were found. All effects of severity were positive and in the small range. This indicates that severity, regardless of its magnitude, positively affects attitudes, intention and behavior change.

132

Chapter 6

Type of severity We next compared effects according to type of severity by dividing studies into studies that used: (1) scary images (pictures or films); and (2) written severity information only. As can be seen in Table 6.9, the between class homogeneity statistics indicate that no significant differences were found on measures of persuasion between studies that used scary images and studies that did not. On attitude significant positive effects were found both for studies that used fear-arousing images (d= .24, p < .01) and studies that used only written severity information (d= .25, p < .001). For intentions similar significant positive effects were found for studies that used fear-arousing images (d= .23, p < .0001) and studies that did not (d= .25, p < .0001). On behavior the effect for studies using fear-arousing images was significantly positive (d= .25, p < .001), whereas the effect for studies that used written information was non-significant (d= .17, p = .06). However, between class homogeneity statistics did not indicate that this difference was significant. Within class homogeneity statistics indicated that dividing studies by severity type made all the effects homogeneous, apart from the effect on attitudes for studies using written severity information and the effect on behavior for studies using scary images. This indicates that severity, regardless of the type of severity information used, has a positive effect on attitudes, intentions and behavior.
Table 6.9. Effect sizes for severity on persuasion by type of severity information Variable Attitude Written Scary images Intention Written Scary images Behavior Written Scary images
Note

Qb .01

N 26 14

Nk 3449 2776 4307 4507 2468 3609

d +.2515*** +.2430** +.2513**** +.2272**** +.1728 +.2515***

95% CI +.1098 to +.3933 +.0599 to +.4216 +.1402 to +.3624 +.1228 to +.3316 -.0039 to +.3494 +.1108 to +.3922

Qw 37.49* 11.46 28.64 21.69 7.20 33.30*

.10 26 29 .47 18 23

Qb = between-classes homogeneity statistic; N = number of studies; Nk = number of respondents; d = mean

effect size; CI = confidence interval; Qw = within classes homogeneity statistic; * p < .05, ** p < .01, *** p < .001, **** p < .0001.

3. Effects of vulnerability and severity in studies that varied both factors So far we have found that both severity and vulnerability generally had small effects on most measures of persuasion (see Table 6.4). For attitudes, the effect for severity seemed stronger (d= .26, p < .0001) than the effect of vulnerability (d= -.08, p = .41). In contrast, the effects of severity on intention

A meta-analysis

133

(d= .23, p < .001 vs. d= .33, p < .0001) and behavior (d= .19, p < .001 vs. d= .41, p < .001) appear weaker than the effects of vulnerability. However, we were also interested in the effects of vulnerability and severity in studies that varied both factors. We examined the relative effects of vulnerability and severity on measures of persuasion by looking at the main effects of severity and vulnerability, as well as the interaction-effect, in studies that manipulated/varied both vulnerability and severity. Because of the finding that studies that used existing vulnerable and non-vulnerable groups had little effect on measures on persuasion, only studies that manipulated vulnerability or used self-measures of perceived vulnerability were included in these analyses. Effects of vulnerability and severity on measures of persuasion in studies that varied both factors are similar to effects found in all studies included in this meta-analysis. For attitudes the effects of both vulnerability (d= .14, p = .09) and severity (d= .16, p = .11) were small and non-significant (Table 6.10). These effects are comparable to the effects found for all studies, although the effects of both factors are non-significant for studies that varied both factors. This is probably due to the smaller number of studies in this latter analysis. The interaction effect between vulnerability and severity on attitudes was non-significant (d= .11, p = .053).
Table 6.10. Effect sizes for severity and vulnerability on persuasion (in studies that varied both factors) Variable Attitude Vulnerability Severity Interaction Intention Vulnerability Severity Interaction Behavior Vulnerability Severity Interaction
Note

N 11 11 11 18 18 18 10 10 10

Nk 1369 1369 1369 2514 2514 2514 1321 1321 1321

d +.1368 +.1586 +.1102 +.3601**** +.2123*** +.0755 +.2715** +.4247* +.0642

95% CI -.0231 to +.2967 -.0359 to +.3531 -.0014 to +.2219 +.1984 to +.5217 +.1068 to +.3177 -.0027 to +.1537 +.0926 to +.4504 +.0968 to +.7526 -.0437 to +.1721

Qw 21.80* 34.60*** 2.11 69.53**** 29.98* 1.53 23.84** 77.64**** 1.02

N = number of studies; Nk = number of respondents; d = mean effect size; CI = confidence interval; Qw =

within classes homogeneity statistic; * p < .05, ** p < .01, *** p < .001, **** p < .0001.

Effects of both vulnerability (d= .36, p < .0001) and severity (d=.21, p < .001) on intentions were small and significant, which is again comparable to the effects found for all studies in this meta-analysis. The interaction-effect

134

Chapter 6

between vulnerability and severity was non-significant (d= .08, p = .06). The effects on behavior were similar to those found on intention. Significant positive effects were found for both vulnerability (d= .27, p < .01) and severity (d= .42, p < .05). Again, this is comparable to the effects found for all studies, but the effect of vulnerability is weaker than the effect found in all studies. No evidence was found for an interaction between the two factors (d= .06, p = .20). 4. Effects of other manipulated independent variables Although studies manipulated a wide range of additional variables alongside severity and vulnerability, only few variables were varied in more than one or two studies. These included manipulations of argument quality, response efficacy, and source of the recommendation, as well as variations in personality factors. Because almost all studies measuring personality factors used different personality factors, and because combining these studies still resulted in too few studies to examine the effects of personality factors, no analyses were carried out for personality factors. Analyses were performed for the effects of argument quality, response efficacy, and source manipulations on measures of persuasion (see Table 6.11), as well as on affect and cognitions.
Table 6.11. Effect sizes for response efficacy, argument quality, and source on measures of persuasion Variable Attitude Argument quality Source Intention Response efficacy Argument quality Source Behavior Response efficacy Argument quality Source 6 6 4 823 881 634 +.0722 -.0689 +.1790 -.0645 to +.2089 -.2011 to + .0633 -.1697 to +.5278 .44 2.92 14.51** 12 7 3 1755 1058 451 +.2093*** +.0324 -.0479 +.1012 to +.3174 -.0882 to +.1529 -.2327 to + .1370 14.27 1.93 1.92 11 7 1831 865 +.5382**** +.1303 +.3731 to +.7032 -.0347 to +.2953 29.14** 8.97 N Nk d 95% CI Qw

Note N = number of studies; Nk = number of respondents; d = mean effect size; CI = confidence interval; Qw = homogeneity statistic; * p < .05, ** p < .01, *** p < .001, **** p < .0001.

Argument quality As expected, argument quality had a significant positive effect on attitudes (d= .54, p < .0001), but the effect was small and non-significant on intention (d= .03, p = .60) and behavior (d= -.07, p = .31). Analysis of interaction-effects on attitudes between argument quality and vulnerability and severity

A meta-analysis

135

respectively revealed small non-significant effects for the argument qualityseverity interaction (d= .12, p = .06) and the argument quality-vulnerability interaction (d= .10, p = .13). The effects on intention of the interactions between argument quality and severity (d= 11, p = .27) and argument quality and vulnerability (d= .10, p = .12) were, as expected, also small and nonsignificant. There were too few studies that varied argument quality and severity and/or vulnerability and measured behavior, therefore, no interaction effects for behavior could be tested. Small and non-significant effects were also found for argument quality on affect and cognitions. These findings confirm our assumption that the quality of arguments used in a recommended action affect attitudes towards this protective action, but does not affect intentions and behavior that are mainly determined by feelings of vulnerability. Response efficacy The concept of argument quality is closely related to that of response efficacy, which is proposed as a major determinant of persuasion in many fear appeal theories (e.g., Rogers, 1983; Witte, 1992). Because of the similarity between these two concepts, we had the same predictions for response efficacy as for argument quality. For response efficacy it was not possible to carry out analyses on attitudes, because only one of the studies that manipulated response efficacy also measured attitudes. A significant positive effect was found for response efficacy on intention (d= .21, p < .001), whereas the effect on behavior was non-significant (d= .07, p = .30). Analysis of interactioneffects on intentions revealed non-significant effects for the response efficacyseverity interaction (d= .09, p = .08) and the response efficacy-vulnerability interaction (d= .10, p = .08). On behavior the effects of interactions between response efficacy and severity (d= .09, p = .22) and response efficacy and vulnerability (d= .14, p = .09) were, as expected, non-significant. Response efficacy also had non-significant effects on affect and cognitions. Source Manipulations of message source had small and non-significant effects on attitudes (d= .13, p = .12), intention (d= -.05, p = .61), and behavior (d= .18, p = .31), as well as on affect and cognitions. There were too few studies that varied message source and severity or vulnerability and, therefore, no interaction effects could be tested. We had expected source manipulations to affect attitudes, however, the studies included in this analysis used different types of source manipulations (e.g., source expertise, source credibility, source attractiveness) and, therefore, no broad conclusions can be drawn from these results.

136

Chapter 6

5. Moderator models of study characteristics The effect sizes for vulnerability, once divided by vulnerability type, were mostly homogeneous, whereas the effect sizes for subgroups of severity were homogeneous for intentions, but remained heterogeneous for some subgroups on attitudes and behavior. Therefore, we examined study characteristics to assess possible moderating effects and achieve homogeneity of effect sizes of severity. No significant differences in effects of severity on measures of persuasion were found when dividing studies according to use of specific recommendations, use of pre-measures, time of post-measures, or type of respondents. In addition, none of these moderating models were effective in creating homogeneity of effect sizes on all three measures of persuasion. We applied these same moderator models to the effects of vulnerability, but there were too few studies in the different groups defined by study characteristics to effectively test these models. We did test one model for the effects of vulnerability, and that was for study year. It was found that the more recent the study, the stronger the effect of vulnerability on intention (b= .02, p < .001) and behavior (b= .01, p < .05). Discussion Findings of this meta-analysis largely confirmed our expectations. We found that both vulnerability and severity had significant effects on most measures of persuasion. Positive effects of severity on attitudes were found, whereas the effect of vulnerability on attitudes was very small and non-significant. Both vulnerability and severity had positive effects on intentions and behavior, with the effects of vulnerability appearing slightly stronger than those of severity. These results are in line with our predictions that vulnerability is an important determinant of intention and behavior change, whereas attitudes are not affected by vulnerability. Also, even though both severity and vulnerability had significant effects on intentions and behavior, their interaction was nonsignificant. The effects we found on measures of persuasion for severity and vulnerability were also largely in line with previous meta-analyses of fear appeal studies (e.g., Boster & Mongeau, 1984; Witte & Allen, 2000), as well as recent meta-analyses of protection motivation theory (Floyd et al., 2000; Milne et al., 2000) that found small effects on measures of persuasion of both vulnerability and severity. The one discrepancy between our meta-analysis and previous findings is the weak effect of vulnerability on attitudes that was found in the present meta-analysis. In Witte and Allens meta-analysis (2000), as well as in the two protection motivation meta-analyses (Floyd et al., 2000; Milne et al., 2000), the effect of vulnerability on attitudes was much stronger. This discrepancy was expected and is due to a difference in the studies

A meta-analysis

137

included in the present versus previous meta-analyses. Previous meta-analyses included many fear appeal studies that confounded severity and vulnerability in the same manipulation, whereas the present meta-analysis excluded all these studies. In addition, the protection motivation meta-analyses included many studies that were not fear appeal experiments at all. The fact that in all confounded studies higher vulnerability was accompanied by high severity information explains these different effects on attitudes. Vulnerability and severity had similar positive effects on both perceived fear and more general measures of negative affect. This confirms our expectations that the severity of negative consequences and feelings of vulnerability arouse fear and other negative emotions. Effects on cognitive responses elicited by a fear-arousing communication only partly confirmed our expectations. Vulnerability had medium strong effects on positive thoughts about the recommendation, whereas the effect of severity was almost nonexistent. This confirms our positive processing bias assumption, which entails that vulnerable respondents exposed to a message depicting a severe risk will process the protective action information with a positive bias, resulting in more positive thoughts about a recommendation (De Hoog et al., 2005a). In contrast, both vulnerability and severity had similar positive effects on minimizing thoughts about the fear appeal, even though we had expected the effect of vulnerability to be much stronger than the severity effect. However, minimizing thoughts induced by feelings of vulnerability have been found to be mainly elicited when the severity of the risk was high (De Hoog, Stroebe & de Wit, 2004) and, therefore, it is not surprising that severity also had an effect on minimizing thoughts. This implies an interaction between vulnerability and severity with most minimizing thoughts being elicited when both vulnerability and severity are at a high level. Effects for perceived threat were in line with our expectations. Severity had a positive effect on perceived threat, whereas the effect of vulnerability was non-significant. However, additional analyses revealed that the vulnerability effect on perceptions of threat was positive for studies that included a specific recommendation and negative for studies that did not, although both effects were non-significant. Nevertheless, these results correspond with findings from studies we conducted without providing an action recommendation, in which perceptions of threat were lower for vulnerable respondents than nonvulnerable respondents (De Hoog et al., 2004). It has to be pointed out, however, that only few studies were included in this analysis and, therefore, only tentative conclusions can be drawn from these effects. Tests of a possible mediation of the effects of vulnerability on intentions and behavior by the effects of vulnerability on affect and cognitions revealed that effects of vulnerability on intention were mediated by negative emotions and minimizing thoughts elicited by the fear appeal. The stronger the effect of

138

Chapter 6

vulnerability on negative affect and minimizing thoughts about the fear appeal respectively, the stronger the intention to engage in a protective action. The effects of vulnerability on behavior were mediated by intentions. The stronger the effect of vulnerability on intentions, the stronger the effect on behavior. This confirms our predictions that feelings of vulnerability elicit negative emotions and minimizing thoughts about the fear appeal, which in turn result in higher intentions to engage in a recommended protective action, which will lead to more actual behavior. Analysis of the effects of different types of vulnerability revealed that effects for manipulated and perceived vulnerability were, as expected, very similar, whereas these effects were different to the effects of existing vulnerable groups. Positive significant effects were found for manipulated and perceived vulnerability on intentions and behavior, whereas for existing groups the effect sizes were virtually absent and non-significant. This clearly indicates that assuming that individuals are vulnerable because they belong to a certain group, and targeting these specific risk groups with fear appeals will not be very effective in influencing individuals intentions and behavior, unless these individuals actually feel vulnerable. No major differences were found in any of the analyses comparing the effects of different levels and types of severity information. It seems to make no difference whether a health threat is depicted as extremely severe or only moderately severe, or whether scary images are used. We had assumed that providing individuals with severity information would be more important than the type or level of the severity information, and we had, therefore, predicted effects to differ across severity levels and types. However, it seems that as long as a health threat is at least somewhat severe, increases in severity do not have any impact. Apart from a stronger effect on attitudes when severity information is compared to a control group than when it is compared to lower levels of severity, no other significant differences were found. The effects of all severity levels and types were positive and significant. This indicates that severity, regardless of the magnitude or type of severity information used, has a positively effect on changing attitudes, intention and behavior change. The major other variables that have been manipulated in fear appeal studies (e.g., argument quality, response efficacy, and source manipulations) did not have many effects on the dependent variables in the meta-analysis. Argument quality only had a significant positive effect on attitudes, as predicted. In addition, response efficacy had a significant positive effect on intentions, even though we had expected response efficacy to only affect attitudes (which could not be tested in this meta-analysis). This effect does correspond with findings in meta-analyses on protection motivation theory that have found positive effects of response efficacy on intention (Floyd et al., 2000; Milne et al., 2000).

A meta-analysis

139

Comparing results to predictions of the stage model and other fear appeal theories The main contribution of the stage model of the processing of fear-arousing communications (Das et al., 2003; De Hoog et al., 2005a) relative to other fear appeal theories is that the stage model specifies the cognitive processes underlying persuasion through fear-arousing communications. In addition, the stage model uniquely proposes a positive relationship between defense motivation and persuasion, and it specifies how defense motivation results in increased persuasion, through biased processing of a fear-arousing communication. Importantly, the stage model assumes that defense motivation induced by high threat perceptions does not interfere with the effectiveness of a feararousing communication, but actually contributes to behavior change. This assumption differs from those of most other fear appeal theories that have generally proposed that defense motivation will result in avoidance reactions or in other ways will undermine persuasion (e.g., Hovland et al., 1953; Witte, 1992). However, the mediation analysis in this meta-analysis refutes these assumptions of previous fear appeal theories that defense motivation undermines persuasion. It was found that stronger effects of vulnerability on minimizing thoughts about the fear appeal (indicating defense-motivated biased processing) resulted in higher intentions. This supports the stage models assumption that defense motivation does not undermine persuasion but actually enhances it. The stage models specific assumptions regarding the processing of feararousing communications also result in different predictions of the effects of components of fear appeals, compared to most other fear appeal theories. All major fear appeal theories assume a threat efficacy interaction on measures of persuasion, with respondents presumably only engaging in a recommended action when perceptions of threat are high and the recommendation is portrayed as effective (e.g., Hovland et al., 1953, Rogers, 1983; Witte, 1992). The stage model, on the other hand, assumes that defense-motivated individuals will be motivated to perceive any recommendation as effective, as long as it is at least somewhat plausible, because engaging in a recommendation will make these individuals feel save (Das et al., 2003; De Hoog et al., 2005a). More specifically, protection motivation theory (Rogers, 1983) and the extended parallel process model (Witte, 1992) assume that perceptions of threat (i.e. severity and vulnerability) and perceptions of efficacy (response efficacy and self-efficacy) affect measures of persuasion. All main factors specified in these theories (except self-efficacy, which could not be analyzed in this meta-analysis) were found to have positive significant effects on one or more measures of persuasion. This is consistent with findings of previous

140

Chapter 6

meta-analysis that higher levels of severity, vulnerability and/or response efficacy resulted in more attitude, intention or behavior change (Floyd et al., 2000; Milne et al., 2000; Witte & Allen, 2000). However, the interaction effects between response efficacy and vulnerability and response efficacy and severity were not significant for intentions and behavior. This is consistent with findings in the meta-analysis by Witte and Allen (2000), in which they were also unable to find a significant interaction between threat and efficacy. More importantly, this finding has important implications for the theoretical reasoning of the effectiveness of fear appeals. Whereas all previous fear appeal theories assume that fear-arousing communications will only result in intention and behavior change when both threat and efficacy are high, this metaanalysis clearly indicates this is not the case. These results do correspond with assumptions from the stage model that proposes that high perceptions of threat will result in increased persuasion regardless of the efficacy of the recommendation (Das et al., 2003; De Hoog et al., 2005a). In addition, the stage model assumes different effects for outcome measures that asses attitudes and those that assess behavioral intentions and behavior. This is different from most other theories, which have generally proposed the same effects for attitudes, intention and behavior. In this metaanalysis we have shown that, in concurrence with the predictions of our stage model, that attitudes toward a protective action were based mainly on evaluation of the information provided. On attitudes, a positive effect of both severity and argument quality was found, whereas the effect of vulnerability was non-significant. Yet, when it came to measures of persuasion that had behavioral implications (i.e., intentions and behavior), feeling vulnerable had a positive effect on both intentions and behavior, confirming the stage models assumption that individuals will be motivated to engage in the protective action if they feel vulnerable. This clearly shows that vulnerability plays an important role in the acceptance of fear-arousing communications. Unfortunately, as the studies in this meta-analysis show, many studies on fear appeals have ignored the concept of vulnerability altogether and have only focused on the effects of severity. Furthermore, many studies on fear appeals, even very recent ones, have not differentiated between vulnerability and severity in their manipulations (e.g., Chu, 1966; Janis & Feshbach, 1953; McMahan, Witte & Meyer, 1998; Morman, 2000; Witte, 1994). The findings in this meta-analysis have shown that vulnerability and severity have positive separate effects on measures of persuasion, and therefore, their effects should be studied independently.

A meta-analysis

141

Conclusions The present meta-analysis expands on previous meta-analyses by showing the separate impact of vulnerability to and severity of a risk, as well as the impact of levels and types of severity and vulnerability on the processing and acceptance of fear-arousing communications. We found that both severity and vulnerability positively affected measures of persuasion. In line with our predictions, severity had a stronger effect on attitudes than vulnerability, while the effects of vulnerability were stronger on intention and behavior. In addition, the meta-analysis has shown that manipulating vulnerability or using self-measures to assess perceived vulnerability has positive effects on intention and behaviors, whereas belonging to a risk group had almost no effect on intention or behavior. Thus, feeling vulnerable instead of belonging to a vulnerable group, motivates intention and behavior change. In addition, it was found that extremely fear-arousing messages are no more effective than messages that simply state the negative consequences of a certain behavior. The results of this meta-analysis have important implications for the effective use of fear-arousing communications. For fear-arousing communications to be effective it is not necessary to make the communication as gruesome as possible, because plain written information about negative consequences will probably work just as well. In addition, it is important for fear-arousing communications to not just target risk groups, but to make sure people will actually feel vulnerable to the portrayed risk. In stressing personal vulnerability to negative consequences of certain behaviors fear-arousing communications can be an effective way of changing individuals health impairing behaviors.

Chapter 7: General discussion

The aim of the present thesis was to provide more insight into the conditions under which fear-arousing communications are (in)effective, and to specify how and why they are (in)effective, by examining the processes underlying persuasion when individuals are exposed to fear-arousing communications. Experiments were conducted in which several aspects of the fear appeal and the recommended action were manipulated/varied, as well as aspects outside of the communication, using the stage model of the processing of fear-arousing communications as the theoretical framework (Das et al., 2003; De Hoog et al., 2005a; Stroebe, 2000). The main emphasis was on the impact of severity of and especially vulnerability to a health risk on the processing and acceptance of fear-arousing communications. In Chapters 2 through 4, the processing of the action recommendation was examined, by manipulating respondents vulnerability to a health risk high in severity, as well as by offering an action recommendation, which was supported by strong or weak arguments. In addition, the experiments in these chapters examined the role of different aspects of the action recommendation and aspects outside of the communication on the processing and acceptance of fear-arousing communications. The experiment in Chapter 2 examined the role of expertise of the source to which the recommendation was attributed, Chapter 3 focused on the impact of costs of engaging in the offered protective action, and Chapter 4 explored the role of two personality factors, dispositional optimism and neuroticism. Chapter 5 examined the processing of both the fear appeal and the action recommendation. Experiments 5.1 and 5.2 focused on the processing of the fear appeal, while Experiment 5.3 focused on both the processing of the fear appeal and the subsequent action recommendation. In these studies, the severity of the consequences was also manipulated, instead of keeping the severity constant and high, as was the case in Chapters 2 through 4. In Experiment 5.1 and 5.2 no specific recommendation was offered, while in Experiment 5.3 a recommendation was offered, which was supported by strong or weak arguments. In addition, Experiment 5.2 looked at the impact of credibility of the source to which the severity message was attributed. Finally, Chapter 6 looked at the processing of fear-arousing communications across fear appeal experiments reported in the literature in an extensive metaanalysis. The pattern of findings observed in the experiments conducted in this thesis is consistent with the main prediction derived from the stage model: Stressing vulnerability to a severe health risk induced defense motivation. Vulnerable defense-motivated respondents experienced more negative emotions, reported more minimizing thoughts about the fear appeal, and reported more positive

144

Chapter 7

and less negative thoughts about the recommended protective action than nonvulnerable respondents did. These findings were confirmed by the metaanalysis in which positive effects of vulnerability were found on negative affect, minimizing thoughts about the fear appeal and positive thoughts about the action recommendation. With regard to persuasion it was found that attitudes were mainly based on evaluation of the provided information and were unaffected by vulnerability. Respondents had more positive attitudes when strong arguments were used rather than weak ones, and some times when the consequences were depicted as severe rather than minor. The meta-analysis also showed positive effects of argument quality and severity on attitudes, whereas the effect of vulnerability was non-significant. However, in all experiments in this thesis, measures of persuasion that had behavioral implications (i.e., intentions and behavior) were solely determined by feelings of vulnerability. Vulnerable respondents had higher intentions, requested more information and subscribed to the recommendation more often than non-vulnerable respondents, regardless of argument quality, source, response costs, personality factors or severity. This was confirmed by the meta-analysis in which positive effects of vulnerability on intention and behavior were found. Both the experiments and the meta-analysis showed that the effect of vulnerability on intentions was mediated by affect and cognitions, whereas the effect of vulnerability on behavior was mediated by intentions. In the remainder of this chapter I will first describe the findings in this thesis in more detail divided by outcome measures, information processing, defense motivation, aspects of the action recommendation, and personality factors. This will be followed by a comparison of the stage model to previous fear appeal theories, and a description of the theoretical as well as practical implications. Finally, in the concluding remarks an answer is given to the questions voiced in the introduction. Affect and cognitions Exposing respondents to a fear-arousing communication aroused fear and more general negative affect. Fear and negative affect became stronger when the negative consequences of a health impairing behavior were depicted as more severe, and especially when feelings of vulnerability became higher. As expected, in all experiments it was found that vulnerable respondents presented with a message depicting severe health consequences experienced more fear and more negative affect than respondents in all other conditions. These results were supported by findings in the meta-analysis in which both vulnerability and severity had positive effects on negative affect and fear. In addition, results showed that vulnerable respondents perceived the depicted consequences of health impairing behaviors as more threatening than

General discussion

145

non-vulnerable respondents did, but only when a protective action was offered. In Experiment 5.1 and 5.2, in which no specific recommendation was offered, perceptions of threat of vulnerable respondents exposed to severe consequences were found to be lower. Similar effects were found in the metaanalysis in which vulnerability had opposite effects on perceived threat, depending on whether or not a specific recommendation was offered. However, the meta-analysis also showed a positive effect of severity on perceived threat. Higher severity resulted in higher perceptions of threat. This was not found in the studies reported in this thesis, but this effect of severity is not surprising because perceptions of threat depend on both feelings of vulnerability and the severity on the consequences of the depicted health risk. When asking respondents what they were thinking while reading the fear appeal, in Experiment 5.1 and 5.2, several different types of thoughts were reported of which the most predominant and most relevant were minimizing thoughts. In concurrence with our assumptions, vulnerable respondents in the high severity condition reported minimizing thoughts about the fear appeal much more often than respondents in all other conditions. Meta-analytic findings of positive effects of both vulnerability and severity on minimizing thoughts support these results. These results are also in concurrence with other studies that have shown that individuals are more critical of evidence that is highly health threatening than of less threatening evidence (Ditto & Lopez, 1992; Janis & Terwilliger, 1962; Liberman & Chaiken, 1992; Sherman, Nelson, & Steele, 2000). When asking respondents about their thoughts about the recommendation, in all experiments besides Experiments 5.1 and 5.2 (in which no recommendation was offered), respondents reported positive (i.e., supportive) or negative (i.e., rejecting) thoughts about the recommendation. Whereas negative thoughts were mainly reported when the recommendation was supported by weak arguments, positive thoughts were mainly elicited by feelings of vulnerability. As expected, vulnerable respondents reported more positive thoughts and less negative thoughts about the action recommendation than non-vulnerable respondents, regardless of argument quality, message source, response costs, optimism, or neuroticism. In the meta-analysis a positive effect of vulnerability on positive thoughts about the action recommendation was also found. Attitudes As predicted, all respondents held more favorable attitudes towards the recommendation when it was supported by strong rather than weak arguments. In the experiment reported in Chapter 3, we not only looked at attitude towards the recommendation, but also at attitudes towards participating in the recommendation. For attitudes towards participating, a

146

Chapter 7

main effect of argument quality was also found. Respondents held more favorable attitudes toward participating when the recommendation was supported by strong rather than weak arguments. These results correspond with studies on dual process theories that have shown that argument quality is an important determinant of attitudes (Baron et al., 1994; Bohner & Weinerth, 2001; Gleicher & Petty, 1992). In addition, the experiments in Chapter 5 showed slightly different effects on attitudes. Experiments 5.1 and 5.2, in which no detailed recommendation was offered, and consequently argument quality was not varied, showed no effects at all on attitude, although respondents generally had very positive attitudes. In Experiment 5.3, respondents not only held more positive attitudes when the recommendation was supported by strong rather than weak arguments, but also when the consequences were depicted as severe rather than minor. Even though attitudes were affected by argument quality, and in Experiment 5.3 also by severity, attitudes were unaffected by feelings of vulnerability. In none of the experiments in this thesis did I find an effect of vulnerability on attitudes. This suggests that vulnerable respondents were unbiased in their evaluation of the recommendation, at least as far as their attitudes were concerned. In the meta-analysis similar results were found. Positive effects of both argument quality and severity were found on attitudes, whereas the effect of vulnerability on attitudes was very small and nonsignificant. These findings confirm our expectations that attitudes are mainly based on an objective evaluation of the provided information. Intention and behavior When it came to persuasion outcomes that had behavioral implications, vulnerability was the main and usually only determinant. As predicted, vulnerable respondents had higher intentions to participate in the recommended action than non-vulnerable respondents did and requested more information, and subscribed to the recommendation more often. More specifically, vulnerable respondents who were exposed to a message depicting severe health consequences had the most positive thoughts about the recommended action, and in combination with negative affect aroused by the fear appeal, this resulted in a greater intention to engage in the offered protective action, as well as in more actual behavior. Positive effects of vulnerability on intentions and behavior were also found in the meta-analysis. In addition, in the meta-analysis there was also a positive effect of severity on intentions and behavior, although the effect of vulnerability appeared much stronger. This supports the assumption that when it comes to behavioral intention and behavior, the most important perquisite is to feel vulnerable.

General discussion

147

Information processing According to the assumptions of the stage model respondents should process the information in fear-arousing communications systematically, unless both severity and vulnerability are low. In this situation, respondents should not be motivated to spend much effort on systematically processing the information, and instead rely on heuristics. The experiments in this thesis have shown that when the consequences of a health risk are depicted as severe, indeed both vulnerable and non-vulnerable respondents process the information systematically. Support for systematic processing comes mainly from the effects of argument quality on both attitudes and cognitive responses. If respondents had not processed the information systematically but relied on heuristics, the quality of the arguments that were presented in the recommendation should not have affected attitudes or thoughts about the recommendation. However, both vulnerable and non-vulnerable respondents had more positive attitudes towards a recommendation when strong arguments were used rather than weak ones, and more negative thoughts about the recommendation were reported when weak arguments were used. Even though this thesis has shown that when severity is high both vulnerable and non-vulnerable respondents process information systematically, not much support has been found for heuristic processing when both vulnerability and severity are low. If non-vulnerable respondents in the low severity condition would have processed the information heuristically, there should have been no effect among these respondents of argument quality. However, no differentiations for the effect of argument quality on attitudes were found between conditions. Moreover, in Experiment 5.2 no effects were found for source credibility on any of the dependent variables, apart from the manipulation check. Since source credibility operates as a heuristic cue, it should have affected non-vulnerable respondents attitudes in the low severity conditions. However, given the fact that even in the low severity condition the perceived severity of the health risk was above the midpoint on the scale, these findings are not inconsistent with our theory: We appear to have failed in creating health risks that are really perceived as trivial. Mediation Mediation analyses provided more insight into how information processing affected persuasion. In Experiments 2.1, 3.1, and 5.3, attitudes were mainly determined by the quality of the arguments used in the recommendation, and this effect was mediated by the negative thoughts respondents had about the recommendation. When the recommendation was supported by weak arguments this elicited negative thoughts about the recommendation, which in turn resulted in less positive attitudes. Whereas attitudes were determined by argument quality, intentions and behavior were mainly determined by a sense

148

Chapter 7

of vulnerability. In processing the fear appeal, which was examined in Chapter 5, vulnerable respondents experienced more negative affect and their negative processing bias showed itself in more minimizing thoughts about the fear appeal. Biased processing of the action recommendation resulted in more positive thoughts about the recommendation, regardless of argument quality, and these positive thoughts about the recommendation together with the negative emotions elicited by the fear appeal, mediated the effect of vulnerability on intention. Even though minimizing thoughts about the fear appeal did not mediate the effect of vulnerability on intentions, the significant correlation between minimizing thoughts about the fear appeal and positive thoughts about the recommendation as well as intention, indicates that minimizing thoughts also contributed to increased persuasion. The negative bias in processing the fear appeal, which results in minimizing thoughts, was positively related to positive thoughts aroused by processing of the subsequent recommended action. In other words, it is not the case that vulnerability to a severe health risk, which induces defense motivation, results in either minimization of the threat or only in maximization of the recommended action, but rather the negative processing bias in threat appraisal contributes to the positive processing bias in coping appraisal. These results were supported by the meta-analysis, in which stronger effects of vulnerability on minimizing thoughts were found to result in stronger intentions. Defense motivation The experiments in this thesis demonstrate that exposing vulnerable respondents to a severe health risk arouses defense motivation, which manifests itself not in avoidance reactions but rather in biased systematic processing. The expected negative processing bias in the appraisal of the fear appeal was found and a positive processing bias in the appraisal of the action recommendation was equally obtained. In the experiments reported in Chapter 5, vulnerable respondents experienced most negative affect after exposure to the high fear message and this aroused minimizing thoughts about the severity of the consequences as well as their own vulnerability. Positive effects of vulnerability on minimizing thoughts were also obtained in the meta-analysis. Yet, when it came to the processing of a possible solution to the threat, in Experiments 2.1, 3.1, 4.1 and 5.3, vulnerable respondents had more positive thoughts about the recommended protective action. These effects were confirmed in the meta-analysis in which a positive effect of vulnerability on positive thoughts was found. In addition, vulnerable respondents had a stronger intention to change their behavior, regardless of argument quality. The fact that the impact of vulnerability on intentions was not tempered by argument quality is also suggestive of defense motivation, because had

General discussion

149

intentions been formed on a solely informational basis, respondents should have weighed the threat of the health risk against the effectiveness of the recommendation. However, even though negative emotions aroused by the fear appeal and positive thoughts elicited by the recommendation mediated the effect of vulnerability on intention, perceived effectiveness, as manipulated by argument quality, did not. Additionally, it was found that vulnerable individuals do not only disregard response efficacy (i.e., argument quality), but also the source of the message (Chapter 2 and 5) and the response costs related to the recommendation (Chapter 3) in deciding to engage in actions that protect them against the deleterious effects of a serious health risk. Of course there could be limitations to the willingness of defense-motivated individuals to engage in an offered protective action. These individuals should at least have some conviction that the action they intent to engage in offers some protection against a health risk. In a similar way, even though it was found that vulnerable respondents are willing to put more effort into engaging in a recommendation, I am sure they would not be prepared to follow a certain action at all costs, just as they are unlikely to participate in an action that is obviously ineffective or pointless. Aspects of the action recommendation In this thesis, several characteristics of the action recommendation were manipulated, including quality of the arguments supporting the recommendation, the source to which the recommendation was attributed and the response costs of participating in the recommended action. Respondents reported more positive attitudes when strong arguments were used rather than weak ones. In addition, argument quality had an effect on cognitive responses, with weak arguments eliciting more negative thoughts than strong arguments. In the meta-analysis it was also found that argument quality had a positive effect on attitudes. These effects are similar to those found in studies assessing predictions from dual process theories in which argument quality affected respondents attitudes toward a host of different subjects (e.g., Baron et al., 1994; Bohner & Weinerth, 2001; Gleicher & Petty, 1992). The concept of argument quality is closely related to that of responseefficacy, which is a major determinant of persuasion in many fear appeal theories (e.g., Rogers, 1983; Witte, 1992). Higher levels of response efficacy have also been found to be related to more positive attitudes, just as argument quality (see Floyd et al., 2000; Milne et al., 2000). The message source was manipulated in two slightly different ways, as source expertise in Experiment 2.1 and as source credibility in Experiment 5.2. In both experiments manipulations of source failed to affect attitudes, even though the manipulation checks clearly indicated that the source manipulations had been successful. This was consistent with the assumption in Experiment

150

Chapter 7

2.1 that respondents would process information systematically, and therefore would not rely on heuristic cues like the source of the message. However, in Experiment 5.2, in which the severity of the consequences was varied, it was expected that source would affect attitudes when non-vulnerable respondents were exposed to consequences low in severity. This failure of manipulations of source to influence attitudes when both vulnerability and severity were low could indicate that respondents in this condition did not process the information heuristically because the low severity condition was still not trivial enough to induce heuristic processing. The metaanalyses also failed to find a significant effect of message source on attitudes. However, research on the effect of message source on attitudes has generally resulted in conflicting findings (Dembroski, Lasater & Ramirez, 1978; Hewgill & Miller, 1965; Powel & Miller, 1967; Williams, Ward & Gray, 1985). In Chapter 3 it was found, as expected, that for respondents who felt vulnerable to the depicted health consequences, the effort required to engage in the protective action had little impact on intention and behavior. Whereas non-vulnerable respondents intentions became lower when a recommendation required high effort, for vulnerable respondents the effort required in following the recommendation had no effect on their intention to engage in the recommendation or on their actual behavior. This confirms our assumption that vulnerable respondents form a strong intention to perform a recommended protective action, regardless of response costs. In sum, although argument quality was found to be an important determinant of attitudes, source and response cost manipulations had little if any effects on persuasion. These results were, in part, confirmed by the metaanalysis in which positive effects of argument quality on attitudes were found, whereas the effect of source on attitudes was very small and nonsignificant. Personality factors In Chapter 4, the possible influence of personality factors on processing and acceptance of fear-arousing communications was explored, using two personality factors, namely dispositional optimism and neuroticism. As expected, respondents high in dispositional optimism experienced less negative affect and had more positive and less negative thoughts about the recommendation than respondents low in dispositional optimism. In contrast, respondents high in neuroticism experienced more negative affect than respondents low in neuroticism did. Furthermore, only dispositional optimism and not neuroticism had an effect on measures of persuasion. Respondents high in dispositional optimism had higher intentions and requested information more often than respondents low in dispositional optimism. These results correspond with findings that optimistic individuals generally have more active and problem-focused ways of coping, whereas neurotic individuals tend to

General discussion

151

react to stressors with more negative emotions, but they usually make no attempt to actively cope with the problem (Costa & McCrae, 1987; Scheier et al., 1986; Watson & Clark, 1984). As expected, none of the effects of optimism or neuroticism interacted with or overruled any of the effects of vulnerability and argument quality on any of the dependent variables. This is not that surprising, because most of the effects of optimism, and the one effect of neuroticism, were only marginally significant. The effects of other personality factors on acceptance of feararousing communications have also been assessed, but studies have produced mixed findings (Dabbs & Leventhal, 1966; Krisher, Darley & Darley, 1973; Sherman et al., 2000). This seems to imply that, although personality factors may have some effect on how respondents respond to fear-arousing communications, the main determinant in getting individuals to intent or to actually change their health impairing behavior is through feelings of vulnerability. Comparing the stage model to other fear appeal theories The main contribution of the stage model of the processing of fear-arousing communications relative to other fear appeal theories is that the stage model specifies the cognitive processes underlying persuasion through fear-arousing communications. In addition, the stage model uniquely proposes a positive relationship between defense motivation and persuasion, and it specifies how defense motivation results in increased persuasion, through biased processing of a fear-arousing communication. Even though all major fear appeal theories assume that fear arousal and/or cognitive processes mediate persuasion (e.g. Hovland et al., 1953; Rogers, 1983; Witte, 1992), none of these theories have made specific predictions about the cognitive processes underlying persuasion. The stage model does, and specifies these cognitive processes underlying persuasion. The stage model proposes that underlying cognitive processes are affected by defense motivation. Defense motivation induces biased processing of a fear-arousing communication, with a negative bias in processing the fear appeal and a positive bias in processing the action recommendation, which in combination enhances persuasion. Whereas research based on dual process theories of attitude change has already shown that a negative bias can occur in the processing threatening health information (Ditto & Lopez, 1992; Kunda, 1987; Liberman & Chaiken, 1992; Reed & Aspinwall, 1998; Sherman et al., 2000), the assumption of a positive bias in the processing of a subsequent provided action recommendation is new. Importantly, the stage model assumes that defense motivation induced by high threat perceptions does not interfere with the effectiveness of a feararousing communication, but actually contributes to behavior change. This

152

Chapter 7

assumption differs from those of most other fear appeal theories that have generally proposed that defense motivation will result in avoidance reactions or in other ways will undermine persuasion (e.g., Hovland et al., 1953; Witte, 1992). These specific assumptions regarding the processing of fear-arousing communications also result in different predictions of the effects of components of fear appeals, compared to most other fear appeal theories. All major fear appeal theories assume a threat efficacy interaction on measures of persuasion, with respondents presumably only engaging in a recommended action when perceptions of threat are high and the recommendation is portrayed as effective (e.g., Hovland et al, 1953, Rogers, 1983; Witte, 1992). The stage model, on the other hand, assumes that defense-motivated individuals will be motivated to perceive any recommendation as effective, as long as it is at least somewhat plausible, because engaging in a recommendation will make these individuals feel save. In addition, the stage model assumes different effects for outcome measures that asses attitudes and those that assess behavioral intentions and behavior. Attitudes toward a protective action have been found to be based mainly on evaluation of the provided information. Yet, when it comes to measures of persuasion that have behavioral implications (i.e., intentions and behavior), the stage model assumes that individuals will only be motivated to engage in the protective action if they actually feel vulnerable. This is different from most other theories, which have proposed the same effects for attitudes, intentions and behaviors. Theoretical implications The studies in this thesis have clearly shown that vulnerability plays an important role in both the processing and acceptance of fear-arousing communications. Whereas the severity of the consequences only affects attitudes, vulnerability was in all experiments the main determinant of intentions and behavior. This implies that fear-arousing communications will only work when individuals feel vulnerable to a health risk. A health risk, no matter how severe or no matter how vivid and scary the negative consequences of a health risk are depicted is, hence, unlikely to change intentions and behavior unless individuals feel vulnerable. These findings indicate the importance of vulnerability in the effectiveness of fear-arousing communications. Unfortunately, many studies on fear appeals have ignored the concept of vulnerability altogether and have only focused on the effects of severity (see Boster & Mongeau, 1984; Sutton, 1982; Witte, 1992). The findings in this thesis clearly show that severity of the consequences only plays a limited role in the effectiveness of fear appeals in changing attitudes, intention and behavior.

General discussion

153

In addition, the findings in this thesis show the importance of differentiation between vulnerability and severity in assessing the effectiveness of feararousing communications. However, many studies on fear appeals, even very recent ones, have not differentiated between vulnerability and severity in their manipulations (Chu, 1966; Janis & Feshbach, 1953; McMahan, Witte & Meyer, 1998; Morman, 2000; Witte, 1994). They typically compare high levels of severity and vulnerability with low levels of severity and vulnerability, making it impossible to determine what the separate effects of both variables are. The findings in this thesis about the processing of fear-arousing communications, especially the positive bias in the processing of action recommendations, have important theoretical implications for those researchers examining the processing of fear appeals. Research has shown that biased processing of the fear appeal can result in more criticizing of the information and, therefore, sometimes lead to decreased persuasion (Ditto & Lopez, 1992; Kunda, 1987; Liberman & Chaiken, 1992; Reed & Aspinwall, 1998; Sherman et al., 2000). However, the studies in this thesis have shown that when providing defense-motivated individuals with an action recommendation they will actually be motivated to engage in this recommendation, and they will process this information with a positive bias, and thus increased persuasion will occur. Practical implications The one clear finding that has emerged from the experiments in this thesis is the importance of vulnerability in intention and behavior change. In addition, the meta-analysis has shown that manipulating vulnerability or using selfmeasures to assess perceived vulnerability resulted in positive effects on intention and behaviors, whereas belonging to a risk group had almost no effect on intention or behavior. Thus, feeling vulnerable instead of belonging to a vulnerable group, motivates intention and behavior change. In addition, it was found that extremely fear-arousing messages are no more effective than messages that simply state the negative consequences of a certain behavior. These findings have important practical implications. The emphasis of health education campaigns has frequently been on the severity of negative health consequences by presenting vivid, scary materials, as well as on stressing the response efficacy of the recommended action. However, this thesis shows that, although these factors affect attitudes, they fail to have much of an impact on intention and behavior. Furthermore, vivid scary images are in no way more effective than just presenting negative consequences in a sober way. This suggests that however severe a health risk, and however effective the protection offered by the recommendation, unless one persuades individuals that they are vulnerable to the health risk, they are unlikely to take protective action. Health education campaigns should, thus,

154

Chapter 7

stress an individuals vulnerability to a health risk and not merely vividly depict the severity of the risk. In stressing personal vulnerability to negative consequences of certain behaviors fear-arousing communications can be an effective way of changing individuals health impairing behaviors. Concluding remarks I started this thesis with the example of the seemingly ineffectiveness of warnings on cigarette packs. These warnings are supposed to make smokers concerned about the hazardous consequences of smoking in such a way that they will stop smoking, yet the majority of smokers continue to smoke (Defacto, 2002). The questions that arose from this have been addressed in this thesis and can now be answered. I asked two types of questions: Are the warnings on cigarette packs ineffective? and What should be put on a cigarette pack to convince smokers to quit smoking? The first question can be answered affirmatively. Although to date little research has been done to assess the effectiveness of cigarette warning labels on quitting smoking, the limited research that is available shows that these warnings mainly affect smokers who already intended to quit within the next year (Defacto, 2002; Joossens, 2004). More importantly, the present thesis has clearly shown that only stressing the negative consequences of a certain behavior is not enough in changing individuals intentions and behaviors. That leaves us with the question what should be put on a cigarette pack to convince smokers to quit smoking. The plan in the Netherlands in 2006 is to put explicit scary pictures of individuals suffering from the consequences of smoking on cigarette packs. These pictures will show individuals with bad tooth decay or extensive tumors. However, I propose that this is unlikely to make smokers quit in any substantial way. Mostly, because the experiments in this thesis have shown that only stressing the negative consequences of a certain behavior is not enough to make individuals change their intentions and behavior. Moreover, the meta-analysis in this thesis has shown that the use of scary images is not more effective than only stressing the negative consequences of a certain behavior. Therefore, if the warnings are having a limited effect on smoking cessation now, adding scary pictures will not make much of a difference. As has been clearly demonstrated in this thesis, instead of presenting the consequences as more severe, it would be much more effective if the emphasis would be on stressing individual vulnerability to the negative consequences of smoking.

References

References marked with an asterisk indicate studies included in the metaanalysis.

3VO (2000). Alcohol in het verkeer. Ruim dertig jaar VVN-campagnes alcohol en verkeer. Huizen: intern rapport. *Allen, B. (1993). Frightening information and extraneous arousal: Changing cognitions and behavior regarding nuclear war. Journal of Social Psychology, 133, 459-467. Aspinwall, L.G., & Burnhart, S.M. (1996). Distinguishing optimism from denial: optimistic beliefs predict attention to health threats. Personality and Social Psychology Bulletin, 22, 993-1003. Aspinwall, L.G., & Taylor, S.E. (1992). Modeling cognitive adaptation: A longitudinal investigation of the impact of individual differences and coping on college adjustment and performance. Journal of Personality and Social Psychology, 63, 989-1003. Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84(2), 191-215. Baron, R.M., & Kenny, D.A. (1986). The moderator-mediator variable distinction in social psychology research: Conceptual, strategic, and statistical considerations. Journal of Personality and Social Psychology, 51, 1173-1182. *Baron, R., Logan, H., Lilly, J., Inman, M., & Brennan, M. (1994). Negative emotion and message processing. Journal of Experimental Social Psychology, 30, 181-201. *Beck, K.H. (1984). The effects of risk probability, outcome severity, efficacy of protection and access to protection on decision making: A further test of protection motivation theory. Social Behavior and Personality, 12, 121-125. *Beck, K.H., & Davis, C.M. (1978). Effects of fear-arousing communications and topic importance on attitude change. Journal of Social Psychology, 104, 81-95. *Beck, K.H., & Lund, A.K. (1981). The effects of health threat seriousness and efficacy upon intention and behavior. Journal of Applied Social Psychology, 1, 401-415. Becker, M. H., Maiman, L.A., Kirscht, J.P., Haefner, D.P., & Drachman, R.H. (1977). The health belief model and prediction of dietary compliance: a field experiment. Journal of Health and Social Behavior, 18, 348-366. *Ben-Ari, O.T., Florian, V., & Mikulincer, M. (2000). Does a threat appeal moderate reckless driving? A terror management theory perspective. Accident Analysis and Prevention, 32, 1-10. *Berkowitz, L., & Cottingham, D.R. (1962). The interest value and relevance of fear -arousing communications. Journal of Abnormal and Social Psychology, 60(1), 37-43. personal

158

References

*Block, L.G., & Williams, P. (2002). Undoing the effects of seizing and freezing: decreasing defensive processing of personally relevant messages. Journal of Applied Social Psychology, 32(4), 803-833. *Bohner, G., & Weinerth, T. (2001). Negative affect can increase or decrease message scrutiny: the affect interpretation hypothesis. Personality and Social Psychology Bulletin, 27, 1417-1428. Bolger, N. (1990). Coping as a personality process: A prospective study. Journal of Personality and Social Psychology, 59, 525-537. Boster, F. J., & Mongeau, P. (1984). Fear-arousing persuasive messages. In R.N. Bostrom (Ed.), Communication yearbook (Vol. 8, pp. 330-375). Beverly Hills: Sage. *Broemer, P. (2004). Ease of imagination moderates reactions to differently framed health messages. European Journal of Social Psychology, 34, 103-119. *Brown, R.A. (1979). Fear-induced attitude change as a function of conformity and drinking pattern in alcoholics. Journal of Clinical Psychology, 35, 454-456. *Burnett, J.J. (1981). Internal-external locus of control as a moderator of fear appeals. Journal of Applied Psychology, 66, 390-393. *Cameron, L.D., & Leventhal, H. (1996). Vulnerability beliefs, symptom experiences, and the processing of health threat information: A self-regulatory perspective. Journal of Applied Social Psychology, 25(21), 1859-1883. Chaiken, S. (1980). Heuristic versus systematic information processing and the use of source versus message cues in persuasion. Journal of Personality and Social Psychology, 39, 752-766. Chaiken, S., Liberman, A., & Eagly, A.H. (1989). Heuristic and systematic information processing within and beyond the persuasion context. In J.S. Uleman, & J.A. Bargh (Eds.), Unintended thought (pp. 212-252). New York: Guilford Press. *Chebat, J. (1995). Affect and memory in advertising: An empirical study of the compensatory processes. Journal of Social Psychology, 135(4), 425-437. Chen, S., & Chaiken, S. (1999). The heuristic-systematic model in its broader context. In S. Chaiken & Y. Trope (Eds.), Dual process theories in social psychology (pp. 73 -96). New York: Guilford Press. Chu, G. C. (1966). Fear arousal, efficacy, and imminency. Journal of Personality and Social Psychology, 4(5), 517-524. Clark, L.A., Watson, D., & Mineka, S. (1994). Temperament, personality, and the mood and anxiety disorders. Journal of Abnormal Psychology, 103(1), 103-116. Cohen, J. (1977). Statistical power analysis for the behavioral sciences. New York: Academic Press. Cortina, J.M., & Nouri, H. (2000). Effect size for ANOVA designs. Thousand Oaks, CA: Sage. Costa, P.T., & McCrae, R.R. (1987). Somatic complaints in males as a function of age and neuroticism: A longitudinal analysis. Journal of Behavioral Medicine, 3, 245-257.

References

159

*Dabbs, J. M., & Leventhal, H. (1966). Effects of varying the recommendations in a fear -arousing communication. Journal of Personality and Social Psychology, 4(5), 525-531. *Das, E.H.H.J., de Wit, J.B.F., & Stroebe, W. (2003). Fear appeals motivate acceptance of action recommendations: Evidence for a positive bias in the processing of persuasive messages. Personality and Social Psychology Bulletin, 29(5), 650-664. Defacto (2002). De veranderende kijk op tabaksproducten. Effecten van de nieuwe gezondheidswaarschuwingen op rokers. Den Haag: intern rapport. *De Hoog, N., Stroebe, W., & de Wit, J.B.F. (2004). The processing of fear-arousing communications: How biased processing lead to persuasion. Manuscript submitted for publication. *De Hoog, N., Stroebe, W., & de Wit, J. B.F. (2005a). The impact of fear appeals on the processing and acceptance of action recommendations. Personality and Social Psychology Bulletin, 31(1), 24-33. *De Hoog, N., Stroebe, W., & de Wit, J.B.F. (2005b). Fear appeals, processing of action recommendations and persuasion: the role of response costs. Manuscript submitted for publication. *De Hoog, N., Stroebe, W., & de Wit, J.B.F. (2005c). The impact of fear appeals on processing and acceptance of action recommendations: the role of dispositional optimism and neuroticism. Manuscript submitted for publication. De Hoog, N., Stroebe, W., & de Wit, J.B.F. (2005d). The impact of vulnerability and severity on the processing and acceptance of fear-arousing communications: A meta-analysis. Manuscript submitted for publication. *Dembroski, T.M., Lasater, T.M., & Ramirez, A. (1978). Communicator similarity, fear -arousing communications, and compliance with the health care recommendations. Journal of Applied Social Psychology, 8, 254-269. *Ditto, P. H., Jemmott, J.B., & Darley, J.M. (1988). Appraising the threat of illness: A mental representational approach. Health Psychology, 7(2), 183-201. Ditto, P.H., & Lopez, D.F. (1992). Motivated skepticism: Use of differential decision criteria for preferred and nonpreferred conclusions. Journal of Personality and Social Psychology, 63, 568-584. *Duval, T.S. (1999). A person-relative-to-event (PrE) approach to negative threat appeals and earthquake preparedness. Journal of Applied Psychology, 29(3), 495-516. *Dziokonski, W., & Weber, S.J. (1977). Repression-sensitization, perceived vulnerability and the fear appeal communication. Journal of Social Psychology, 102, 105-112. Eagly, A.H., & Chaiken, S. (1993). The psychology of attitudes. Fort Worth: Harcourt, Brace, Jovanovich. Eppright, D.R., Tanner, J.F., & Hunt, J.B. (1994). Knowledge and the ordered protection motivation model: tools for preventing AIDS. Journal of Business Research, 30, 13-24.

160

References

*Evans, R.I., Rozelle, R.M., Lasater, T.M., Dembroski, T.M., & Allen, B.P. (1968). New measure of effects of persuasive communications: A chemical indicator of tooth brushing behavior. Psychological Reports, 23, 731-736. *Evans, R.I., Rozelle, R.M., Lasater, T.M., Dembroski, T.M., & Allen, B.P. (1970). Fear arousal, persuasion, and actual versus implied behavioral change. New perspective utilizing a real-life dental hygiene program. Journal of Personality and Social Psychology, 16, 220-227. *Evans, R.I., Rozelle, R.M., Noblitt, R., & Williams, D.L. (1975). Explicit and implicit persuasive communications to initiate and maintain behavior change: new perspective utilizing a real-life dental hygiene situation. Journal of Applied Social Psychology, 5, 150-156. Eysenck, H. J., & Eysenck, S.B.G. (1975). Manual of the Eysenck personality questionnaire. London: Hodder and Stoughton. Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction to theory and research. Reading, MA: Addison-Wesley. Floyd, D. L., Prentice-Dunn, S., & Rogers, R.W. (2000). A meta-analysis of research on protection motivation theory. Journal of Applied Social Psychology, 30(2), 407-429. *Frandsen, K. (1963). Effects of threat appeals and media of transmission. Speech Monographs, 30, 101-104. *Gleicher, F., & Petty, R.E. (1992). Expectations of reassurance influence the nature of fear-stimulated attitude change. Journal of Experimental Social Psychology, 28, 86-100. Goldstein, M.J. (1959). The relationship between coping and avoiding behavior and response to fear-arousing propaganda. Journal of Abnormal and Social Psychology, 58, 247-252. *Griffeth, R. W., & Rogers, R.W. (1976). Effects of fear-arousing components of driver education on students' safety attitudes and simulator performance. Journal of Educational Psychology, 68(4), 501-506. *Haefner, D.P. (1965). Arousing fear in dental health education. Journal of Public Health Dentistry, 25, 140-146. Hallet, R., & Sutton, S. (1986). Factors influencing the decision to attempt to stop smoking in a media-based smoking intervention programme. Health Education Research, 1, 163-173. Harrison, J.A., Mullen, P.D., & Green, L.W. (1992). A meta-analysis of studies of the health belief model with adults. Health Education Research, 7, 107-116. *Hass, J.W., Bagley, G.S., & Rogers, R.W. (1975). Coping with the energy crisis: effects of fear appeals upon attitudes toward energy consumption. Journal of Applied Psychology, 60, 754-756. Hedges, L.V., & Olkin, I. (1985). Statistical methods for meta-analysis. Orlando, FL: Academic Press

References

161

*Hewgill, M.A., & Miller, G.R. (1965). Source credibility and response to fear-arousing communications. Speech Monographs, 32, 95-101. Higbee, K. L. (1969). Fifteen years of fear arousal: Research on threat appeals: 1953 -1968. Psychological Bulletin, 72(6), 426-444. *Ho, R. (1992). Cigarette health warnings: The effects of perceived severity, expectancy of occurrence, and self-efficacy on intention to give up smoking. Australian Psychologist, 27, 109-113. *Horowitz, I.A. (1969). Effects of volunteering, fear arousal, and number of communications on attitude change. Journal of Personality and Social Psychology, 11, 34-37. *Horowitz, I.A., & Gumenik, W.E. (1970). Effects of the volunteer subject, choice, and fear arousal on attitude change. Journal of Experimental Social Psychology, 6, 293-303. Hovland, C.I., Janis, I.L., & Kelley, H.H. (1953). Communication and persuasion: Psychological studies of opinion change. New Haven: Yale University Press. Janis, I. L. (1967). Effects of fear arousal on attitude change: Recent developments in theory and experimental research. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 3, pp. 166-224). San Diego: Academic Press. Janis, I.L., & Feshbach, S. (1953). Effects of fear-arousing communications. Journal of Abnormal and Social Psychology, 48, 78-92. *Janis, I.L., & Milholland, H.C. (1954). The influence of threat-appeals on selective learning of the content of a persuasive message. Journal of Psychology, 37, 75-80. *Janis, I.L., & Terwilliger, R.F. (1962). An experimental study of psychological resistance to fear-arousing communications. Journal of Abnormal and Social Psychology, 6, 403-410. Janz, N., & Becker, M.H. (1984). The health belief model: a decade later. Health Education Quarterly, 11, 1-47. *Jemmott, J.B., Ditto, P.H., & Croyle, R.T. (1986). Judging health status effects of perceived prevalence and personal relevance. Journal of Personality and Social Psychology, 50(5), 899-905. *Jepson, C., & Chaiken, S. (1990). Chronic issue-specific fear inhibits systematic processing of persuasive communications. In M. Booth-Butterfield (Ed.), Communication, cognition, and anxiety. (Special Issue). Journal of Social Behavior and Personality, 5, 61-84. Johnson, B. T., Eagly, A.H. (1989). Effects of involvement on persuasion: A meta -analysis. Psychological Bulletin, 106(2), 290-314. Joossens, L. (2004). Onderzoek naar het effect van gezondheidswaarschuwingen op sigarettenpakjes in Belgi. Brussel: Vlaams Instituut voor Gezondheidspromotie i.s.m. Belgische Federatie tegen Kanker.

162

References

*Kantola, S.J., Syme, G., & Nesdale, A.R. (1983). The effects of appraised severity and efficacy in promoting water conservation: An information analysis. Journal of Applied Social Psychology, 13, 164-182. *Keesling, B. (1995). Interventions to prevent skin cancer: Experimental evaluation of informational and fear appeals. Psychology and Health, 10(6), 477-490. *Keller, P.A. (1996). Increasing the persuasiveness of fear appeals: The effect of arousal and elaboration. Journal of Consumer Research, 22(4), 448-459. *Keller, P.A. (1999). Converting the unconverted: The effect of inclination and opportunity to discount health-related fear appeals. Journal of Applied Psychology, 84(3), 403-415. *Kirscht, J. P., Becker, M.H., Haefner, D.P., & Maiman, L.A. (1978). Effects of threatening communications and mothers' health beliefs on weight change in obese children. Journal of Behavioral Medicine, 1(2), 147-157. *Kirscht, J.P., & Haefner, D.P. (1973). Effects of repeated threatening health communications. International Journal of Health Education, 16, 268-277. *Kleinot, M.C., & Rogers, R.W. (1982). Identifying effective components of alcohol misuse prevention programs. Journal of Studies in Alcohol, 43(7), 802-811. *Kohn, P.M., Goodstadt, M.S., Cook, G.M., Sheppard, M., & Chan, G. (1982). Ineffectiveness of threat appeals about drinking and driving. Accident Analysis and Prevention, 14, 457-464. *Krisher, H. P., Darley, S.A., & Darley, J.M. (1973). Fear-provoking recommendations, intentions to take preventive actions, and actual preventive actions. Journal of Personality and Social Psychology, 26(2), 301-308. Kunda, Z. (1987). Motivated inference: Self-serving generation and evaluation of causal theories. Journal of Personality and Social Psychology, 53, 636-647. LaTour, M.S., & Rotfeld, J.J. (1997). There are threats and (maybe) fear-caused arousal: Theory and confusion of appeals to fear and fear arousal itself. Journal of Advertising, 26(3), 45-60. *LaTour, M.S., Snipes, R., & Bliss, S. (1996). Dont be afraid to use fear appeals: An experimental study. Journal of Advertising Research, 28, 59-67. *LaTour, M.S., & Tanner, J.F. (2003). Radon: Appealing to our fear. Psychology and Marketing, 20(5), 377-394. Lazarus, R.S., & Folkman, S. (1984). Stress, appraisal, and coping. New York: Springer. Leventhal, H. (1970). Findings and theory in the study of fear communications. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 5, pp. 119-186). San Diego: Academic Press. *Leventhal, H., Jones, S., & Trembly, G. (1966). Sex differences in attitude and behavior change under conditions of fear and specific instructions. Journal of Experimental Social Psychology, 2, 387-399. *Leventhal, H., & Niles, P. (1964). A field experiment on fear arousal with data on the validity of questionnaire measures. Journal of Personality, 32, 459-479.

References

163

*Leventhal, H., & Singer, R.P. (1966). Affect arousal and positioning of recommendations in persuasive communications. Journal of Personality and Social Psychology, 4(2), 137-146. *Leventhal, H., Singer, R.P., & Jones, S. (1965). The effects of fear and specificity of recommendation upon attitudes and behavior. Personality and Social Psychology, 2, 20-29. *Leventhal, H., & Trembly, G. (1968). Negative emotions and persuasion. Journal of Personality, 36, 154-168. *Leventhal, H., & Watts, J.C. (1966). Sources of resistance to fear-arousing communications on smoking and lung cancer. Journal of Personality, 34, 155 -175. *Leventhal, H., Watts, J.C., & Pagano, F. (1967). Effects of fear and instructions on how to cope with danger. Journal of Personality and Social Psychology, 6, 313-321. *Liberman, A., & Chaiken, S. (1992). Defensive processing of personal relevant health messages. Personality and Social Psychology Bulletin, 18, 669-679. Lin, E.H., & Peterson, C. (1990). Pessimistic explanatory style and response to illness. Behavior Research and Therapy, 28, 243-248. Lipsey, M.W., & Wilson, D.B. (2001). Practical meta-analysis. California: Sage Publications. Lord, C.G., Ross, L., & Lepper, M.R. (1979). Biased assimilation and attitude polarization: The effects of prior theories on subsequently considered evidence. Journal of Personality and Social Psychology, 37, 2098-2109. *Maddux, J. E., & Rogers, R.W. (1983). Protection motivation and self-efficacy: A revised theory of fear appeals and attitude change. Journal of Experimental Social Psychology, 19, 469-479. McGuire, W. J. (1968). Personality and attitude change: An information-processing theory. In A.G. Greenwald, Brook, T.C., & Ostrom, T.M. (Ed.), Psychological foundations of attitudes. San Diego, CA: Academic Press. McMahan, S., Witte, K., & Meyer, J. (1998). The perception of risk messages regarding electromagnetic fields: Extending the extended parallel process model to an unknown risk. Health Communication, 10(3), 247-259. Milne, S., Sheeran, P., & Orbell, S. (2000). Prediction and intervention in health-related behavior: A meta-analytic review of protection motivation theory. Journal of Applied Social Psychology, 30(1), 106-143. *Moan, C.E. (1968). Support and refutation of threat-inducing communicating in change of attitude toward cigarette smoking. Psychological Report, 22(3), 1054. Morman, M.T. (2000). The influence of fear appeals, message design, and masculinity on men's motivation to perform the testicular self-exam. Journal of Applied Communication Research, 28 (2), 91-116. *Mulilis, J., & Lippa, R. (1990). Behavioral change in earthquake preparedness due to negative threat appeals: A test of protection motivation theory. Journal of Applied Social Psychology, 20(8), 619-638.

164

References

*Neuwirth, K., Dunwoody, S., & Griffin, R.J. (2000). Protection motivation theory and risk communication. Risk Analysis, 20(5), 721-734. NIGZ (2000). De publiekscampagne Drank maakt meer kapot dan je lief is. Woerden: intern rapport. Petty, R.E., & Cacioppo, J.T. (1986). The elaboration likelihood of persuasion. Advances in Experimental Social Psychology, 19, 193-205. *Powell, F.A. (1965). The effect of anxiety-arousing messages when related to personal, familial, and impersonal referents. Speech Monographs, 32, 102-106. *Powell, F.A., & Miller, G.R. (1967). Social approval and disapproval cues in anxiety -arousing communications. Speech Monographs, 34, 152-159. Pyszczynski, T., & Greenberg, J. (1987). Toward an integration of cognitive and motivational perspectives on social inference: A biased hypothesis-testing model. In L. Berkowitz (Ed.), Advances in Experimental Social Psychology (Vol. 20, pp. 297 -340). New York: Academic Press. *Ramirez, A., & Lasater, T.L. (1976). Attitudinal and behavioral reactions to fear -arousing communications. Psychological Reports, 38, 811-818. *Ramirez, A., & Lasater, T.L. (1977). Ethnicity of communicator, self-esteem, and reactions to fear-arousing communicators. Journal of Social Psychology, 102, 79 -91. *Reed, M.B., & Aspinwall, L.G. (1998). Self-affirmation reduces biased processing of health-risk information. Motivation and Emotion, 22, 99-132. Rimal, R. N. (2001). Perceived risk and self-efficacy as motivators: understanding individuals' long-term use of health information. Journal of Communication, 633 -654. Rogers, R.W. (1975). A protection motivation theory of fear appeals and attitude change. Journal of Psychology, 91, 93-114. Rogers, R.W. (1983). Cognitive and physiological processes in fear appeals and attitude change: a revised theory of protection motivation. In J.T. Cacioppo & R.E. Petty (Eds.). Social Psychophysiology: A sourcebook (pp. 153-176). New York: Press. *Rogers, R.W., & Deckner, C.W. (1975). Effects of fear appeals and physiological arousal upon emotion, attitudes and cigarette smoking. Journal of Personality and Social Psychology, 32, 222-230. *Rogers, R.W., Deckner, C.W., & Mewborn, C.R. (1978). An expectancy-value theory approach to the long-term modification of smoking behavior. Journal of Clinical Psychology, 34, 562-566. *Rogers, R. W., & Mewborn, C.R. (1976). Fear appeals and attitude change: Effects of a threat's noxiousness, probability of occurrence, and the efficacy of coping responses. Journal of Personality and Social Psychology, 34(1), 54-61. Rogers, R. W., & Prentice-Dunn, S. (1997). Protection motivation theory. In D. S. Gochman (Ed.), Handbook of health behavior research I: Personal and social determinants (pp. 113-132). New York: Plenum press. Guilford

References

165

*Rogers, R. W., & Thistlethwaite, D.L. (1970). Effects of fear arousal and reassurance on attitude change. Journal of Personality and Social Psychology, 15(3), 227-233. Ronis, D. L. (1992). Conditional health threats: Health beliefs, decisions, and behaviors among adults. Health Psychology, 11(2), 127-134. *Rosen, T. J., Terry, N.S., & Leventhal, H. (1982). The role of esteem and coping in response to a threat communication. Journal of Research in Personality, 16, 90 -107. Rosenstock, I.M. (1974). Historical origins of the health belief model. Health Education Monographs, 2, 1-8. Rosenthal, R. (1991) Meta-analytic procedures for social research. California: Sage. *Rozelle, R. M., Evans, R.I., Lasater, T.M., Dembroski, T.M., & Allen, B.P. (1973). Need for approval as related to the effects of persuasive communications on actual, reported and intended behavior change-a viable predictor. Psychological Reports, 33, 719-725. *Ruiter, R.A.C., Kok, G., Verplanken, B., & Brug, J. (2001). Evoked fear and effects of appeals on attitudes to performing breast self-examination: An information -processing perspective. Health Education Research, 16(3), 307-320. Ruiter, R.A.C., Abraham, C., & Kok, G. (2001). Scary warnings and rational precautions: A review of the psychology of fear appeals. Psychology and Health, 16, 613-630. Scheier, M.F., & Carver, C.S. (1985). Optimism, coping, and health: Assessment and implications of generalized outcome expectancies. Health Psychology, 4, 219 well-being: theoretical overview and empirical update. Cognitive Therapy and Research, 16, 201-228. Scheier, M.F., Carver, C.S., & Bridges, M.W. (1994). Distinguishing optimism from neuroticism (and trait anxiety, self-mastery, and self-esteem): A reevaluation of the life orientation test. Journal of Personality and Social Psychology, 67(6), 1063 -1078. Schwarzer, R. (1989). Manual for meta software, 5.3. Free University of Berlin. *Shelton, M.L., & Rogers, R.W. (1981). Fear-arousing and empathy arousing appeals to help: The pathos of persuasion. Journal of Applied Social Psychology, 11, 366-378. *Sherman, D.A.K., Nelson, L.D., & Steele, C.M. (2000). Do messages about health risks threaten the self? Increasing the acceptance of threatening health messages via self -affirmation. Personality and Social Psychology Bulletin, 26, 1046-1058. *Siero, S., Kok, G., & Pruyn, J. (1984). Effects of public education about breast cancer and BSE. Social Science and Medicine, 18, 881-888. SIRE. Archief Stichting ideele reclame. Amstelveen: intern rapport. *Skilbeck, C., Tulips, J., & Ley, P. (1977). The effects of fear arousal, fear position, fear exposure, and sidedness on compliance with dietary instructions. European Journal of Social Psychology, 7, 221-239. -247. Scheier, M.F., & Carver, C.S. (1992). Effects of optimism on psychological and physical

166

References

*Slater, M.D. (2002). Effects of threatening visuals and announcer differences on responses to televised alcohol warnings. Journal of Applied Communication Research, 30(1), 27-49. *Smart, R.G., & Fejer, D. (1974). The effects of high and low fear messages about drugs. Journal of Drug Education, 4, 225-235. *Smith-Klohn, L., & Rogers, R.W. (1991). Dimensions of severity of health threats: The persuasive effects of visibility, time of onset, and rate of onset on young womens intentions to prevent osteoporosis. Health Psychology, 10, 323-329. *Southwick-Bensley, L., & Wu, R. (1991). The role of psychological reactance in drinking following alcohol prevention messages. Journal of Applied Social Psychology, 21(13), 1111-1124. *Steffen, V.J., Sternberg, L., Teegarden, L.A., & Shepherd, K. (1994). Practice and persuasive frame: Effects on beliefs, intention and performance of a cancer self -examination. Journal of Applied Social Psychology, 24(10), 897-925. Stroebe, W. (2000). Social psychology and health (second edition). Buckingham: Open University Press. *Struckman-Johnson, C. (1994). Effect of persuasive appeals in AIDS PSAs and condom commercials on intentions to use condoms. Journal of Applied Social Psychology, 24(24), 2223-2244. Sutton, S. R. (1982). Fear-arousing communications: a critical examination of theory and research. In J. R. Eiser (Ed.), Social psychology and behavioral medicine (pp. 303 -337). Chichester: Wiley. Sutton, S., & Eiser, J.R. (1984). The effects of fear-arousing communications on cigarette smoking: An expectancy value approach. Journal of Behavioral Medicine, 7, 13-33. *Sutton, S., & Hallet, R. (1988). Understanding the effects of fear-arousing communications: The role of cognitive factors and amount of fear aroused. Journal of Behavioral Medicine, 11(4), 353-359. *Sutton, S., & Hallet, R. (1989). Understanding seat-belt intentions and behavior: A decision-making approach. Journal of Applied Social Psychology, 19(15), 1310-1325. *Tanner, J.F., Day, E., & Crask, M.R. (1989). Protection motivation theory: An extension of fear appeals theory in communication. Journal of Business Research, 19, 267 -276. Tanner, J.F., Hunt, J.B., & Eppright, D.R. (1991). Protection motivation model: a normative model of fear appeals. Journal of Marketing, 55, 36-45. *Taylor, K.M., & Shepperd, J.A. (1998). Bracing for the worst: Severity, testing and feedback timing as moderators of the optimistic bias. Personality and Socia Psychology Bulletin, 24(9), 915-926. Van Kampen, D. (1993). The 3DPT dimensions S, E, and N: a critical evaluation of Eysencks psychoticism model. European Journal of Personality, 7, 65-105. Vinck, J., Wels, G., Arickx, M., & Vinck, S. (1998). Optimisme gemeten: Validatie van de Nederlandstalige Levensorientatietest bij jongeren. Gedrag en Gezondheid, 26(2), 79-90.

References

167

Watson, D., & Clark, L.A. (1984). Negative affectivity: The disposition to experience aversive emotional states. Psychological Bulletin, 96, 465-490. *Watson, M., Pettingale, K.W., & Goldstein, D. (1983). Effects of fear appeals on arousal, self-reported anxiety, and attitude towards smoking. Psychological Reports, 52, 139-146. Weinstein, N.D. (1982). Unrealistic optimism about susceptibility to health problems. Journal of Behavioral Medicine, 5, 441-460. *Wheatley, J.J., & Oshikawa, S. (1970). The relationship between anxiety and positive and negative advertising appeals. Journal of Marketing Research, 7, 85-89. Williams, P.G., OBrien, C.D., & Colder, C.R. (2004). The effects of neuroticism and extraversion on self-assessed health and health-relevant cognition. Personality and Individual Differences, 37, 83-94. *Williams, R.E., Ward, D.A., & Gray, L.N. (1986). The persistence of experimentally induced cognitive change: A neglected dimension in the assessment of drug prevention programs. Journal of Drug Education, 15(1), 33-42. Witte, K. (1992). Putting the fear back into fear appeals: The extended parallel process model. Communication Monographs, 59, 329-349. Witte, K. (1994). Fear Control and Danger Control: a test of the extended parallel process model. Communication Monographs, 61, 113-134. Witte, K., & Allen, M. (2000). A meta-analysis of fear appeals: Implications for effective public health campaigns. Health Education and Behavior, 27(5), 591-616. Witte, K., & Morrison, K. (1995). Using scare tactics to promote safer sex among juvenile detention and high school youth. Journal of Applied Communication Research, 23, 128-142. *Wolf, S., Gregory, W.L., & Stephan, W.G. (1986). Protection motivation theory: Prediction of intention to engage in anti-nuclear war behaviors. Journal of Applied Social Psychology, 16, 310-321. *Wurtele, S. (1988). Increasing womens calcium intake: The role of health beliefs, intentions, and health value. Journal of Applied Social Psychology, 18, 627-639. *Wurtele, S. K., & Maddux, J.E. (1987). Relative contributions of PMT components in predicting exercise intentions and behavior. Health Psychology, 6(5), 453-466.

Summary

The present thesis deals with the processing and acceptance of fear-arousing communications. Fear-arousing communications have been used in health education campaigns for over fifty years. The basic assumption underlying fear-arousing communications is that the more one succeeds in making individuals concerned about the negative consequences of a certain behavior, the greater the probability that they will change their health impairing behavior. Over fifty years of empirical research on fear-arousing communications has shown that high fear messages are generally more effective than low fear messages in changing individuals attitudes, intentions and behavior (Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000). However, these empirical studies have not succeeded in fully explaining the processes underlying the effects of fear-arousing communications on persuasion. The present thesis examines these processes underlying persuasion through fear-arousing communications, providing more insight into the conditions under which fear-arousing communications are (in)effective, and more importantly, to specify how and why they are (in)effective. The main emphasis is on the impact of severity of and especially vulnerability to a health risk on the processing and acceptance of fear-arousing communications. A stage model of the processing of fear-arousing communications is used as the theoretical framework (Das, de Wit & Stroebe, 2003; Stroebe, 2000). This stage model of the processing of fear-arousing communications integrates ideas derived from dual-process theories (e.g., Chaiken, 1980), with those of earlier theories of fear-arousing communications (e.g., Leventhal, 1970; Rogers, 1983; Witte, 1992). In line with most fear appeal theories (Rogers, 1983; Witte, 1992), the stage model assumes that individuals exposed to a fear-arousing communication engage in two types of appraisal, namely appraisal of the health risk, and appraisal of coping strategies available for reducing or eliminating the health risk. Appraisal of the magnitude of a threat, based on the information in a fear appeal about the severity of and personal vulnerability to a risk, is assumed to determine both individuals processing mode (i.e., depth of processing) and their processing goal (i.e., accuracy or defense motivation). According to the stage model, if a health risk is trivial and individuals do not feel vulnerable to the health risk, they are unlikely to be motivated to invest much effort in thinking about the contents of the communication and they might rely on heuristic processing modes. In contrast, if individuals feel vulnerable to a minor health risk, the feelings of vulnerability should be

170

Summary

sufficient motivation to invest effort into systematically processing the contents of the communication. Similarly, if individuals do not feel vulnerable but a health risk is depicted as severe, individuals are likely to invest effort in processing the contents of a communication describing a severe risk, because it is useful to be well informed about a serious health risk, even if the danger is not immanent. According to the stage model, if individuals are exposed to information about a severe health risk and if they feel vulnerable to this risk this will threaten their self-definitional belief of being healthy, which will arouse defense motivation as well as the motivation to carefully scrutinize the information given. According to the stage model, defense motivation manifests itself in biased systematic processing. The direction of the bias aroused by defense motivation is said to vary with the type of appraisal; the stage model proposes a negative processing bias in appraisal of the threat and a positive processing bias in the appraisal of coping strategies. In appraising the health risk, defense-motivated individuals will attempt to minimize the threat by critically looking at the content of a fear appeal. They will engage in a biased search for inconsistencies and logical errors, and their evaluation of evidence will consequently be biased in the direction of the preferred conclusion (Ditto & Lopez, 1992; Liberman & Chaiken, 1992; Sherman, Nelson & Steele, 2000). If this strategy proves unsuccessful, because even biased processing is constrained by evidence and rules of inference (Kunda, 1987), individuals will have to accept that they are personally at risk. In this case, the subsequent processing of a recommendation will be biased as well, but in a positive direction. At this stage, the processing goal of defense-motivated individuals will be to find the protective action effective, because then these individuals can feel save. The processing will involve attempts to make the recommendation appear highly effective by means of a biased search for arguments supporting the effectiveness of the protective action, and through biased evaluation of these arguments. In other words, defense motivation will lead to a positive bias in the processing of the action recommendation, and consequently heighten the motivation to accept a solution to a particular threat, regardless of the quality of the arguments supporting this recommendation. In this thesis experiments are reported that examine the processing of both the fear appeal and the recommended action, and the consequent effects on persuasion. In these experiments, severity of and vulnerability to different health risks are manipulated/varied, as well as aspects of the recommended action (argument quality, message source, and response costs), and factors outside the communication (personality factors).

Summary

171

In Chapters 2 through 4, the processing of the action recommendation is examined, by manipulating respondents vulnerability to a health risk high in severity, as well as by offering an action recommendation, which is supported by strong or weak arguments. In addition, the experiments in these chapters examine the role of different aspects of the action recommendation and factors outside of the communication on the processing and acceptance of feararousing communications. The experiment in Chapter 2 examines the role of the expertise of the source to which the recommendation is attributed; Chapter 3 focuses on the impact of costs of engaging in the offered protective action, while Chapter 4 explores the role of two personality factors, dispositional optimism and neuroticism. Chapter 5 also examines the processing of the fear appeal. Experiment 5.1 and 5.2 focus exclusively on the processing of the fear appeal, while Experiment 5.3 focuses on both the processing of the fear appeal and the action recommendation. In these studies, the severity of the consequences is also manipulated, instead of only focusing on the impact of high severity as was the case in Chapters 2 through 4. In Experiment 5.1 and 5.2 no specific recommendation is offered, while in Experiment 5.3 one is, which is supported by strong or weak arguments. In addition, Experiment 5.2 looks at the impact of credibility of the source to which the severity message is attributed. Chapter 6 looks at the processing of fear-arousing communications across fear appeal experiments in the literature in an extensive meta-analysis. Finally, Chapter 7 summarizes and discusses the findings from the previous chapters, as well their theoretical and practical implications. The pattern of findings observed in the experiments conducted in this thesis is consistent with the main prediction derived from the stage model: Stressing vulnerability to a severe health risk induced defense motivation. Vulnerable defense-motivated respondents experienced more negative emotions, reported more minimizing thoughts about the fear appeal, and more positive and less negative thoughts about the recommended action than non-vulnerable respondents. These findings were confirmed by the meta-analysis that showed positive effects of vulnerability on negative affect, minimizing thoughts about the fear appeal and positive thoughts about the recommendation. With regard to persuasion it was found that attitudes were mainly based on evaluation of the provided information and were unaffected by vulnerability. Respondents had more positive attitudes when strong arguments were used rather than weak ones, and some times when the consequences were depicted as severe rather than minor. The meta-analysis also showed positive effects of argument quality and severity on attitudes, whereas the effect of vulnerability was non-significant. However, in all experiments in this thesis, measures of persuasion that had

172

Summary

behavioral implications (i.e., intentions and behavior) were solely determined by feelings of vulnerability. Vulnerable respondents had higher intentions, requested more information and subscribed to the recommendation more often than non-vulnerable respondents, regardless of argument quality, message source, response costs, or personality factors. This was confirmed by the metaanalysis in which positive effects of vulnerability on intention and behavior were found. Both the experiments and the meta-analysis showed that the effect of vulnerability on intentions was mediated by affect and cognitions, whereas the effect of vulnerability on behavior was mediated by intentions. The present thesis gives a detailed description of how the processing of fear-arousing communications works and why fear-arousing communications can be effective in influencing attitudes, intention and behavior. Fear-arousing communications are most effective when perceptions of vulnerability to a severe health risk are high. In this situation defense motivation is induced, which shows itself in a negative bias in processing the fear appeal and in a positive bias in processing the recommendation, which combined results in intention and behavior change. These findings have important theoretical and practical implications. The studies in this thesis have shown the importance of vulnerability, and the importance of differentiation between vulnerability and severity in assessing the effectiveness of fear-arousing communications. However, many studies on fear appeals have ignored the concept of vulnerability altogether and have only focused on the effects of severity. In addition, those studies that have focused on vulnerability have often not differentiated between vulnerability and severity in their manipulations. They typically compared high levels of severity and vulnerability with low levels of severity and vulnerability, making it impossible to determine what the separate effects of both variables are. The emphasis of health education campaigns has frequently been on the severity of negative health consequences by presenting vivid, scary materials, as well as on stressing the response efficacy of the recommended action. However, this thesis shows that, although these factors affect attitudes, they fail to have much of an impact on intention and behavior. Furthermore, vivid scary images are in no way more effective than only presenting negative consequences in a sober way. This suggests that however severe a health risk, and however effective the protection offered by the recommendation, unless one persuades individuals that they are vulnerable to a health risk, they are unlikely to take protective action. Health education campaigns should, hence, stress an individuals vulnerability to a health risk and not merely vividly depict the severity of the risk. In stressing personal vulnerability to negative consequences of certain behaviors, fear-arousing communications can be an effective way of changing individuals health-impairing behaviors.

Samenvatting

Dit proefschrift handelt over de verwerking en acceptatie van angstcommunicaties. Angstcommunicaties worden al sinds de jaren vijftig gebruikt in gezondheidsvoorlichting. De achterliggende gedachte van angstcommunicaties is dat hoe meer het lukt om individuen zich zorgen te laten maken over de negatieve gevolgen van een bepaald gedrag, hoe groter de kans is dat zij hun slechte gedrag zullen veranderen. Meer dan vijftig jaar onderzoek naar angstcommunicaties heeft laten zien dat hoge angstboodschappen in het algemeen effectiever zijn dan lage angstboodschappen in het veranderen van attitudes, intenties en gedrag (Boster & Mongeau, 1984; Sutton, 1982; Witte & Allen, 2000). Echter, dit onderzoek is niet in staat geweest de onderliggende processen te benoemen die aan de effectiviteit van angstcommunicaties ten grondslag liggen. Dit proefschrift onderzoekt deze onderliggende processen met als doel meer inzicht te verschaffen in de condities waaronder angstcommunicaties (in)effectief kunnen zijn en belangrijker, hoe en waarom angstcommunicaties (in)effectief zijn. De nadruk ligt in dit proefschrift op de invloed van de ernst van en vooral kwetsbaarheid voor een gezondheidsrisico op de verwerking en acceptatie van angstcommunicaties. Als theoretisch raamwerk wordt een stadia model voor de verwerking van angstcommunicaties gebruikt (Das, de Wit & Stroebe, 2003; Stroebe, 2000). Dit stadia model integreert ideen van dualprocess theorien (Chaiken, 1980) met die van eerdere fear appeal theorien (Leventhal, 1970; Rogers, 1983; Witte, 1992). Net als de meeste fear appeal theorien (Rogers, 1983; Witte, 1992) veronderstelt het stadia model dat blootstelling aan een angstcommunicatie voor twee soorten evaluaties zorgt: een evaluatie van het gezondheidsrisico en een evaluatie van de beschikbare coping strategien om het gezondheidsrisico te verminderen. Deze twee soorten evaluaties komen overeen met de twee componenten van angstcommunicaties, namelijk (1) een angstboodschap die de ernst van en kwetsbaarheid voor een gezondheidsrisico benadrukt en (2) een aanbeveling die informatie verstrekt om de bedreiging te verminderen. De evaluatie van de gezondheidsbedreiging, gebaseerd op de informatie in de angstboodschap over de ernst van en kwetsbaarheid voor een bedreiging, bepaalt zowel de motivatie tot verwerken als het soort informatieverwerking. Het stadia model stelt dat wanneer een gezondheidsrisico triviaal is en iemand zich niet kwetsbaar voelt voor dit gezondheidsrisico, diegene niet erg gemotiveerd zal zijn (veel) tijd en moeite te steken in de verwerking van en denken over de informatie en de informatie heuristisch verwerkt zal worden. Wanneer iemand zich wel kwetsbaar voelt voor een niet ernstig

176

Samenvatting

gezondheidsrisico, dan voorspelt het stadia model dat de gevoelens van kwetsbaarheid iemand motiveren om moeite te doen om de informatie systematisch te verwerken. Wanneer iemand zich niet kwetsbaar voelt maar het betreft een ernstig gezondheidsrisico, dan zal de ernst van het risico iemand motiveren de informatie systematisch te verwerken, omdat het goed is genformeerd te zijn over een serieus gezondheidsrisico, ook al is het niet perse een directe bedreiging voor de persoon zelf. Het stadia model stelt dat wanneer iemand zicht kwetsbaar voelt voor een ernstig gezondheidsrisico iemands overtuiging gezond te zijn wordt bedreigd. In dit geval zal er sprake zijn van zowel defensieve motivatie als systematische informatieverwerking. Defensieve motivatie uit zich in een vertekening in systematische informatieverwerking. De richting van de vertekening hangt af van de soort evaluatie. Het stadia model voorspelt een negatieve vertekening in de evaluatie van de angstboodschap en een positieve vertekening in de evaluatie van de aanbeveling. Defensief gemotiveerde individuen zullen kritisch naar de inhoud van de angstboodschap kijken en proberen deze te minimaliseren. Dit kan door kritiek te leveren en de informatie te relativeren om zo de bedreiging te verminderen. De evaluatie van de angstboodschap zal dan ook vertekend worden in de richting van de geprefereerde conclusie (Ditto & Lopez, 1992; Liberman & Chaiken, 1992; Sherman, Nelson & Steele, 2000). Een ernstige angstboodschap maakt het echter moeilijk de gezondheidsbedreiging helemaal te negeren en iemand moet haast wel accepteren dat hij/zij een risico loopt. Wanneer dit gebeurt zal defensieve motivatie doorwerken in de evaluatie van de aanbeveling. Een aanbeveling biedt een mogelijke oplossing voor de bedreiging, waardoor deze informatie verwerkt zal worden met een positieve vertekening. De aanbeveling zal kritisch worden bekeken, maar als meer valide worden gevalueerd omdat acceptatie van de aanbeveling de bedreiging reduceert. Met andere woorden, defensieve motivatie leidt tot een positieve vertekening in informatieverwerking, waardoor iemand gemotiveerd raakt een aanbeveling voor een gezondheidsbedreiging te accepteren. In dit proefschrift worden experimenten beschreven waarin gekeken wordt naar de verwerking van zowel de angstboodschap als de aanbeveling en de gevolgen hiervan op overtuiging. In deze experimenten wordt de ernst van en/of kwetsbaarheid voor een gezondheidsrisico gemanipuleerd/gevarieerd, evenals aspecten van de aanbeveling (de kwaliteit van de argumenten, de bron van de boodschap en de kosten om een aanbeveling op te volgen) en aspecten buiten de communicatie (persoonlijkheidsfactoren). In Hoofdstuk 2 t/m 4 wordt de verwerking van de aanbeveling onderzocht, waarbij kwetsbaarheid voor een ernstige gezondheidsbedreiging gemanipuleerd wordt en een aanbeveling wordt aangeboden die ondersteund wordt door sterke of zwakke argumenten.

Samenvatting

177

Bovendien wordt in deze hoofdstukken verschillende aspecten van de aanbeveling en aspecten buiten de communicatie onderzocht. In Hoofdstuk 2 wordt gekeken naar de rol van expertise van de bron; Hoofdstuk 3 richt zich op de invloed van de kosten om een aanbeveling op te volgen en Hoofdstuk 4 kijkt naar de rol van twee persoonlijkheidskenmerken, optimisme en neuroticisme. In Hoofdstuk 5 wordt ook de verwerking van de angstboodschap onderzocht. Experimenten 5.1 en 5.2 richten zich enkel op de verwerking van de angstboodschap, terwijl Experiment 5.3 zich richt op zowel de verwerking van de angstboodschap als de verwerking van de aanbeveling. In deze studies wordt de ernst van de gezondheidsconsequenties ook gemanipuleerd. In Experimenten 5.1 en 5.2 wordt geen specifieke aanbeveling gedaan, maar in Experiment 5.3 wel en deze wordt ondersteund door sterke of zwakke argumenten. Daarnaast wordt in Experiment 5.2 gekeken naar de invloed van de geloofwaardigheid van de bron. In Hoofdstuk 6 wordt de verwerking en acceptatie van angstcommunicaties bekeken in eerdere fear appeal studies in een meta-analyse. Tenslotte wordt in Hoofdstuk 7 de bevindingen uit de vorige hoofdstukken en hun theoretische en praktische implicaties besproken. Het patroon van bevindingen in de experimenten in dit proefschrift kwam overeen met de belangrijkste voorspellingen van het stadia model: het benadrukken van kwetsbaarheid voor een ernstig gezondheidsrisico veroorzaakt defensieve motivatie. Kwetsbare defensief-gemotiveerde respondenten ervoeren meer negatieve emoties, hadden meer relativerende gedachten over de angstboodschap en hadden meer positieve gedachten over de aanbeveling dan niet-kwetsbare respondenten. Deze bevindingen werden bevestigd door de meta-analyse waarin positieve effecten gevonden werden van kwetsbaarheid op negatieve emoties, relativerende gedachten over de angstboodschap en positieve gedachten over de aanbeveling. Wat betreft overtuiging werd gevonden dat attitudes vooral gebaseerd zijn op een evaluatie van de gegeven informatie en niet worden benvloed door kwetsbaarheid. Respondenten hadden positievere attitudes wanneer een aanbeveling ondersteund werd door sterke argumenten dan wanneer zwakke argumenten gebruikt werden en soms ook wanneer de gevolgen gepresenteerd werden als ernstig in plaats van niet ernstig. De meta-analyse liet ook positieve effecten zien van kwaliteit van de argumenten en ernst op attitudes, terwijl het effect van kwetsbaarheid op attitudes niet significant was. Echter, in alle experimenten in dit proefschrift werden overtuigingsmaten die gedragsimplicaties hadden (intenties en gedrag) voornamelijk benvloed door gevoelens van kwetsbaarheid. Kwetsbare respondenten hadden hogere intenties, vroegen vaker informatie aan en gaven zich vaker op voor een aanbeveling dan niet-kwetsbare respondenten, ongeacht de kwaliteit van de argumenten, de bron van de boodschap, de response kosten, of

178

Samenvatting

persoonlijkheidsfactoren. Dit werd bevestigd door de meta-analyse waarin positieve effecten van kwetsbaarheid werden gevonden op intentie en gedrag. Zowel de experimenten in dit proefschrift als de meta-analyse lieten verder zien dat het effect van kwetsbaarheid op intentie gemedieerd werd door emoties en cognities, terwijl het effect van kwetsbaarheid op gedrag gemedieerd werd door intentie. Dit proefschrift geeft een gedetailleerde beschrijving van de verwerking van angstcommunicaties en waarom angstcommunicaties effectief kunnen zijn in het veranderen van attitudes, intenties en gedrag. Angstcommunicaties zijn het effectiefst wanneer percepties van kwetsbaarheid voor een ernstig gezondheidsrisico hoog zijn. In dat geval wordt defensieve motivatie veroorzaakt dat zich uit in een negatieve vertekening in de verwerking van de angstboodschap en een positieve vertekening in de verwerking van de aanbeveling, hetgeen resulteert in meer intentie en gedragsverandering. Deze bevindingen hebben belangrijke theoretische en praktische implicaties. De studies in dit proefschrift laten zien hoe essentieel kwetsbaarheid is en dientengevolge differentiatie tussen kwetsbaarheid en ernst in het bepalen van de effectiviteit van angstcommunicaties. Echter, veel fear appeal studies kijken enkel naar de effecten van ernst en negeren kwetsbaarheid. Bovendien wordt in studies die wel kijken naar zowel kwetsbaarheid als ernst vaak geen onderscheid gemaakt tussen deze twee factoren. Meestal vergelijken deze studies hoge ernst en hoge kwetsbaarheid met lage ernst en lage kwetsbaarheid waardoor het onmogelijk is om te bepalen wat de onafhankelijke effecten van beide variabelen zijn. De nadruk in gezondheidsvoorlichting ligt meestal op de ernst van de negatieve gezondheidsconsequenties van een bepaald gedrag, door middel van het presenteren van enge beelden of plaatjes, evenals het benadrukken van de effectiviteit van een aanbeveling. Echter, dit proefschrift laat zien dat, alhoewel deze factoren attitudes benvloeden, ze weinig invloed hebben op intenties en gedrag. Bovendien werd in de meta-analyse gevonden dat confronterende enge beelden niet effectiever zijn dan het enkel presenteren van negatieve consequenties op een sobere manier. Dit suggereert dat ook al is een gezondheidsrisico nog zo ernstig en is een aanbeveling nog zo effectief, zolang men individuen niet kan overtuigen dat ze kwetsbaar zijn voor een gezondheidrisico zullen ze hun gedrag niet veranderen. Voorlichtingcampagnes zouden dus iemands persoonlijke kwetsbaarheid voor een gezondheidsrisico moeten benadrukken en niet enkel de negatieve gevolgen van een bepaald gedrag. Door het benadrukken van persoonlijke kwetsbaarheid voor negatieve consequenties van bepaalde gedragingen kunnen angstcommunicaties een effectieve manier zijn om het gedrag van individuen te veranderen.

181

Dankwoord

Dit proefschrift is het resultaat van de afgelopen vier jaar dat ik met veel plezier gewerkt heb bij de vakgroep Sociale en Organisatie Psychologie van de Universiteit Utrecht. Veel mensen hebben op n of andere manier bijgedragen aan het tot stand komen van dit proefschrift. Graag zou ik een aantal van die mensen hier bij name willen noemen: Wolfgang Stroebe, mijn promotor, en John de Wit, mijn co-promotor, bedankt voor al jullie advies en steun en jullie vertrouwen in mij als onderzoeker. Het was een heel prettige samenwerking en ik had mij geen betere begeleiders kunnen wensen; Minke de Gruil, ik wil je bedanken voor je hulp bij het coderen van de gedachten in al mijn experimenten, omdat je mijn paranimf wilde zijn, en vooral voor de gezellige tijd op en buiten het werk. Zonder jou was het niet hetzelfde geweest; Wendy Roos, bedankt dat je mijn paranimf wilde zijn. Het is dan wel geen foto geworden, maar het is de intentie die telt; Bart Oudemans, ik wil je bedanken voor je hulp bij het coderen van de studies in de meta-analyse en omdat je een heel gezellige kamergenoot was. Ik ga die Duitse schlagers nog missen. Klinker kopen?; Annefloor, Chava, Despoina, Josine, Marjoliek, Meriem, Mike, Natasja, Rosalie, Saar en alle andere docenten en aios, bedankt voor de leuke tijd tijdens het werk en voor al die biertjes in de Basket; En verder iedereen binnen en buiten de universiteit die mij de afgelopen vier jaar geholpen en/of gesteund heeft. Dankjewel! Natascha de Hoog juni 2005

183

Curriculum Vitae

Natascha de Hoog werd op 11 november 1976 geboren in Rotterdam. Na het vwo met succes te hebben afgerond op de Christelijke Scholengemeenschap Angelus Merula in Spijkenisse, ging zij in 1995 psychologie studeren aan de Universiteit Leiden. In januari 2001 studeerde zij af met een doctoraal in Klinische en Gezondheidspsychologie. In april 2001 begon zij als Assistent in Opleiding bij de vakgroep Sociale en Organisatie Psychologie van de Universiteit Utrecht, waarvan dit proefschrift het resultaat is.