Sunteți pe pagina 1din 38

Prediction and Intervention in Health-Related Behavior: A Meta-Analytic Review of Protection Motivation Theory

SARAHMILNE~
Lniversi/y of Barli Bath, United Kit~gdum

P A S C H A L SHEERAN A N D S H E l N A O R B E L L LniverJity of She/fkld Sheffield Lnrted hingdom

Protection motivation theory (PMT) was introduced by Rogers i n 1975 and has since been widely adopted as a framework for the prediction of and intervention in health-related behavior. However. PMT remains the only major cognitive model of behavior not to ha\ K been the subject o f a meta-analytic review. A quantitative revieh o f P M T i s important to assess its overall utility as a predictive model and to establish which o f its variables would be most useful to address health-education interventions. The present paper provides a comprehensive introduction to PMT and its application to health-related behavior, together with a quantitative review o f the applications of P M T to health-related intentions and behavior. The associations beween threat- and coping-appraisal variables and intentions, and all components o f the model and behavior were assessed both by meta-analysis and by vote-count procedures. Threat- and coping-appraisal components of PMT were found to be useful in the prediction of health-related intentions. The model was found to be useful in predicting concurrent behavior, but o f less utility in predicting future beha\ior. The coping-appraisal component of the model was found to have greater predicti\ e validity than was the threat-appraisal component. The main findings are discussed in relation to theory and research on social cognition models. The importance ot'the main findings to health education is also discussed, and future research directions are suggested.

Protection motivation theory (PMT; Rogers, 1975, 1983) provides an important social cognitive account of protective behavior. Rogers first introduced the theory in 1975. Since then. there have been many studies applying the model to predict and understand protective behavior, particularly health-related behavior. Other social cognition models have also been applied to the understanding and prediction of health behavior, including the health belief model (Becker, 1974; Rosenstock, 1966), the theory of reasoned action (Ajzen & Fishbein, 1980; Fishbein & Ajzen, 1975), and the theory of planned behavior (Ajzen, 1988, 1991).
'Correspondence concerning this article should be addressed to Sarah Milne, Department o t Psychology. University o f Rath, Claverton Down, Bath B A 2 7AY, United Kingdom. c-mail: s.e.miIne@!bath.ac.uk.

106
Journal ofApplied Social Psychology, 2000,30, 1 pp. 106-143. Copyright 0 2000 by V. H. Winston & Son, Inc. All rights reserved.
I

PROTECTION MOTIVATION THEORY

107

It is important to assess the utility of these models in the prediction and understanding of health-related behavior in order to establish which components to use as targets for health-education interventions. One way of doing this is to conduct a meta-analysis, in which the results of all the studies testing a model are combined in an overall assessment of the models performance. Although PMT has been the subject of recent literature reviews (Boer & Seydel, 1996; Rogers & Prentice-Dunn, 1997), it remains the only model not to have been the subject of a meta-analytic review. Meta-analytic reviews have been conducted for all of the other social cognition models commonly used in the prediction of health-related behavior, including the health belief model (Harrison, Mullen, & Green, 1992). the theory of reasoned action (Sheppard, Hartwick, & Warshaw, 1988), and the theory of planned behavior (Armitage & Conner, 1996; Godin & Kok, 1996). The aim of the present paper is to combine the results of studies testing PMT in a meta-analysis to provide a quantitative review of applications of the model to health-related behavior. Protection Motivation Theory The original version of PMT (Rogers, 1975) grew out of research on fear appeals. Afear appeal is an informative communication about a threat to an individuals well-being. Along with details of the threat itself, the communication suggests measures that can be taken to avoid it or to reduce its impact. For example, a fear appeal could be a health-education pamphlet outlining the threat of breast cancer with a recommendation to perform breast self-examination as a means to detect the cancer early, thereby reducing its potential impact. A central issue in fear-appeals research is establishing the way in which a feararousing communication can change attitudes and, subsequently, change behavior. A major problem in this area is that, although it was widely accepted that fear appeals were multifaceted stimuli, there had been little progress in identifying the variables involved, as well as their cognitive mediational effects (Rogers, 1975). Rogers introduced PMT in order to address this difficulty. It was originally developed in an attempt to provide conceptual clarity in the area of fear appeals and to bridge the gap between research on fear appeals and research on attitude change. PMT was designed to specify and operationalize the components of a fear appeal in order to determine the common variables that produced attitude change. It was assumed that each component of a fear appeal would initiate a corresponding cognitive mediating process. These processes would, in turn, influence protection motivation, in the form of an intention to adopt the recommended behavior contained within the fear appeal. Protection motivation was said to be an intervening variable that arouses, sustains, and directs activity (Rogers, 1975, p. 94). In 1983, Rogers revised his theory into a more general theory of cognitive change. The revised PMT included a broader spectrum of information sources

108 MILNE ET AL.


that could initiate a coping process. Fear appeals remained one such source of information, but observational learning, personality, and prior experience were also included as information sources capable of initiating cognitive activity leading to protection motivation. Additional cognitive mediating processes were added, including an account of the appraisal processes leading to maladaptive coping responses, such as continuing or adopting smoking cigarettes. The coping-appraisal section of the model was also expanded. The revised theory acknowledged the importance of social learning theory (Bandura, 1977. 1986, 1991) by incorporating Banduras ( 1977, 1986) construct, perceived selfefficacy, into the model. The PMT Models Structure and Variables The structure of PMT was influenced by expectancy-value theory (Edwards, 1954). This is central to the major social cognition behavioral models, as well as to the models that were influential in the formulation of PMT (e.g., the parallel response model: Leventhal, 1970; and the drive-reduction model: Janis, 1967). In expectancy-value theory, the tendency to adopt a given behavior is said to be a function of expectancies regarding the consequences of the behavior and the value of those consequences. Hovland, Janis, and Kelleys ( 1 953) expectancyvalue theory suggested that there are three main stimulus variables in a fear appeal: (a) the magnitude of noxiousness of a given event, (b) the probability that the given event will occur if no protective behavior is adopted or existing behavior modified, and (c) the availability and effectiveness of a coping response to reduce or eliminate the noxious stimulus (Rogers, 1975). Rogers adopted these three components as the basis for the original formulation of PMT. He proposed that each of these constitutes a cognitive mediational process: The magnitude of noxiousness initiates perceived severity; the probability of occurrence initiates perceived vulnerability; and the efficacy of the recommended response initiates perceived response efficacy. These cognitive mediational processes could be characterized as having two form-threat appraisal and coping appraisal. Threat appraisal concerns the process of evaluating the components of a fear appeal that are relevant to an individuals perception of how threatened he or she feels. The PMT variables that capture threat appraisal are perceived vulnerability, perceived severity, and fear arousal. Perceived vulnerability assesses ho\v personally susceptible an individual feels to the communicated threat. It is typically measured by items such as Considering all of the different factors that may contribute to AIDS, including your own past and present behavior, what would you say are your chances of getting AIDS? (answered on a Likert scale with endpoints J am almost certain J will to J am almost certain I will not; Aspinwall, Kemeny, Taylor, Schneider, 8~Dudley, 1991). Perceived severity assesses how serious the individual believes that the threat would be to his or her own life. This

PROTECTION MOTIVATION THEORY

109

is measured by items such as Osteoporosis is a very serious disease (strongly agree to strongly disagree; Wurtele, 1988). Fear arousal assesses how much fear the threat evokes for the individual and is measured by items such as The thought of breast cancer makes me feel (very anxious to not at all anxious; Hodgkins & Orbell, 1998). Fear is seen as an intervening variable; the more vulnerable an individual feels to a threat and the more serious he or she believes it to be, the more fear will be aroused and the greater the appraised threat will be. The greater the perceived threat, the more likely the individual is to be motivated to protect himself or herself; that is, the more likely a behavioral intention to adopt a protective behavior will be formed. Rogers (1983) revision of PMT includes a component appraising the rewards of not adopting the recommended coping response as part of the threat-appraisal process. The higher the rewards of not adopting the coping response, the less likely the individual is to adopt it. To our knowledge, only one PMT study has attempted to include rewards (Abraham, Sheeran, Abrams, & Spears, 1994). An example of how this was done is Sex would be more exciting without a condom (strongly agree to strongly disagree). Because of a lack of data, the rewards component of the model will not be assessed in the present review. Coping appraisal evaluates the components of a fear appeal that are relevant to an individuals assessment of the recommended coping response to the appraised threat. In the original formulation of the theory, Rogers ( 1983) identified response efficacy as the main determinant of coping appraisal. Response ejicacy concerns beliefs about whether the recommended coping response will be effective in reducing threat to the individual and is measured by such items as If I quit smoking I will greatly increase my chances of living a longer life (strongly agree to strongly disagree; Maddux & Rogers, 1983). The revised PMT includes self-efficacy and response costs in the coping-appraisal component of the model. Self-efficacy concerns an individuals beliefs about whether he or she is able to perform the recommended coping response. A typical measure of self-efficacy is Sticking with a regular program of exercise would be very difficult for me to do (strongly agree to strongly disagree; Wurtele & Maddux, 1987). Response costs concern beliefs about how costly performing the recommended response will be to the individual; for example, 1 would feel awkward examining my breasts (likely to unlikely; Hodgkins & Orbell, 1998). Protection motivation is a key mediator of the relationship between behavior and threat and coping appraisal. Protection motivation is synonymous with the intention to perform a behavior (e.g., I intend to cany out a breast self-examination in the next month; strongly agree to strongly disagree; Hodgkins & Orbell, 1998) and is a positive linear function of the beliefs that (a) the threat is severe, (b) the individual is personally vulnerable to the threat, (c) the recommended response is effective, (d) the individual is able to perform the recommended response, and is a negative linear function of the belief that (e) the perceived

110

MILNE ET AL.

costs of the recommended coping response would be high.2 A schematic representation of the PMT is shown in Figure 1. Applications of the Model Although PMT was originally developed as an extension of fear-appeal research, it has since been adopted as a more general model of decision making in relation to threats (Maddux, 1993). PMT has been applied to a number ofthreats, the majority being health-related threats, where the model has been used to understand and predict protective health behavior. Only applications of PMT to health-related threats will be included in this review. Other protective behaviors to which PMT has been applied include antinuclear behaviors (Axelrod & Newton, 1991; Wolf, Gregory, & Stephan, t 986), increasing earthquake preparedness (Mulilis & Lippa, 1990), water conservation (Kantola, Syme, & Nesdale, 1983), coping with technological and environmental hazards (Weigman, Taal, Van den Bogaard, & Gutteling, 1992), and coping with burglary (Weigman et al., 1992). Aims of the Present Review The first aim of the present review is to establish the overall success of PMT as a predictive model of health-related intentions and behavior. Meta-analysis and vote-count procedures will be used to evaluate the success of each of the models components in predicting intention (threat- and coping-appraisal variables), and behavior (all components of the model) will be assessed to identify those best to
*In addition to their influence on protection motivation, threat- and coping-appraisal processes may result in maladaptive coping responses, such as denial or avoidance. Because of insufficient data, maladaptive coping responses will not be addressed in the present review. However, a brief sunimary of the main findings concerning maladaptive coping is offrred here. Maladaptive coping has been operationalized in terms ofavoidance (Fruin, Pratt, & Owen. 1991; Rippetoe & Rogers, 1987: IJmeh, in press; van der Velde & van der Pligt, 1991 ). denial (Abraham et al.. 1994). fatalism (Abraham et al. 1994; Fruin el al.. 1991; Rippetoe & Rogers. 1987). wishful thinking (Abraham et al., 1994; Rippetoe & Rogers, 1987). and hopelessness (Fruin et al.. 1991; Rippetoe & Rogers, 1987). Thrcat appraisal has been found to be positively correlated with maladaptive coping responses, indicating that high threat perception makes one likely to adopt some coping response, either adaptive or maladaptive ( e g ,Abraham et al., 1994; Rippetoe & Rogers. 1987). In general, self-efficacy and response efficacy have been found to be negatively correlated with maladaptive coping responses (e.g., Abraham et al., 1994; Eppright. Tanner. & Hunt, 1994; Rippetoe & Rogers, 1987; Tanner, Day, & Crask. 1989). In addition, maladaptive coping responses (avoidance. in particular) have been found to inhibit protection motivation (Abraham et al., 1994; Rippetoe & Rogers, 1987; van der Velde & van der Pligt, 1991). It has also been found that a tendency toward maladaptive coping affects the coping process. outlined in PMT itself. For example. avoidance has been shown to inhibit severity, self-efficacy (Rippetoe & Rogers, 1987; Umeh. in press; van der Velde & van der Pligt, 1991 ), and response efficacy (Umeh, in press), and hypervigilance has been found to lead to increased self-efficacy and response efficacy. stronger fear appraisal. and higher perceived severity (Umeh, in press).

Cognitive mediating processes Behavior

Environmental Communication Observational Learning

Intraoersonal Personality variables Prior experience

Perceived self-efficacy Perceived response-efficacy Perceived response-cost

Denial Fatalism Wishful thinking

+ve

Health &protective behavior

Figure 1. Schematic representation of protection motivation theory (adapted from Rogers, 1983). +ve = positive association; -ve = negative association.

112 MlLNE ET AL.

target in health-education interventions. These techniques complement one another since meta-analysis establishes the overall strength of association between two variables, whereas a vote count shows how often these associations are significant. Meta-analysis will be used to determine the average correlation between each PMT variable and intention and behavior, and a vote count will be used to see how often these associations are significant across studies. PMT has an advantage over the health belief model (Becker, 1974; Rosenstock, 1966), the theory ofreasoned action (Ajzen & Fishbein, 1980; Fishbein & Ajzen, 1975), and the theory of planned behavior (Ajzen, 1988, 1991) in that it has been consistently subjected to experimental tests. In these studies, researchers have presented communications designed to manipulate PMT variables and then measured the effects of the communication on PMT variables. The second aim of this paper is to review the methods by which PMT variables have been manipulated in experimental studies and to assess the success of such manipulations in producing belief change. The following four questions will be addressed by the review:
1. What is the association between threat- and coping-appraisal variables and intentions to perform health-related behaviors? 2. What is the association between PMT variables and measures of concurrent health behavior?

3. What is the association between PMT variables and health behavior in prospective studies?

4. How successhl are manipulations of threat- and coping-appraisal variables in bringing about cognitive change?
Method
Data Collection and Selection

Published studies were identified using the computer databases PsychLit, Medline, and Social Sciences Citation Index (BIDS [Bath Information Data Service]). A keyword search was conducted using the expressions protection and motivation to identify all papers including these words in their title or abstract. TO ensure that no applications of PMT were missed using this search strategy, an additional search was carried out to identify all papers citing Rogers (1975, 1983). PMT review papers and book chapters were also consulted to identify any papers that were missed from the computer databases. Unpublished research was also included.
Criteria for Inclusion

The following criteria were used to select studies suitable for inclusion in the analysis:

PROTECTION MOTIVATION THEORY

113

1. The study must be an empirical application of PMT.3

2. There must be a measure of behavioral intention, and concurrent or subsequent behavior included in the analysis.
3. The behavior used in the study must be a health-related behavior (e.g., breast self-examination, smoking cessation, adopting a healthy diet). Tjpes of Studies Included There are two main types of health-related behavior explored in the studies collected: detection behaviors and prevention behaviors (cf. Maddux, 1993). Detection behaviors are conducted to enable an individual to discover whether he or she has a specific condition that could be a threat to health (e.g., breast selfexamination, mammography, Pap test, testicular self-examination). Detection behaviors influence health only if the individual takes further preventive action after learning the result of the detection behavior (Maddux, 1993). Prevention behaviors are behaviors that an individual adopts or ceases in the belief that doing so will reduce the risk of developing disease in the future. Examples of these behaviors include increased exercise, smoking cessation, sunscreen use, moderate alcohol consumption, and dietary improvements. These behaviors may play a preventive role in a range of illnesses such as stroke, heart disease, and cancer. PMT studies can also be characterized according to research design. For the purpose of the analysis, studies were classified according to three main types of research design. Correlational design. This type of design measures an individuals cognitions that have been established through general life experience. Here, all participants receive a questionnaire measuring PMT variables. No experimental manipulation is involved, and no prior information concerning the health threat or protective behavior is provided. Rather, scores on PMT variables are correlated with intentions, behavior, or both. Health-education intervention (health education vs. no education). In this design, variables are measured after one group of participants (the experimental group) has received information about the health threat and recommended protective action (Le., health education). This information addresses several PMT variables and presents facts in such a way that protection motivation will be encouraged. The control group does not receive this information. For example, Seydel, Taal, and Weigman (1990) showed an experimental group an educational-television film about cancer, while the control group watched a program on an unrelated topic.
3Although other studies have looked at associations between variables included in PMT with intention and behavior, the present review included only PMT studies in order to ensure that all components were operationalized and used according to the framework of the model.

114

MILNE ET AL.

Experimental manipulations ojspecific PMT variables. In this design, particular PMT variables are manipulated (high vs. low) in a communication prior to their measurement. For example, Stanley and Maddux (1986) gave participants a written communication containing a combination of (a) high or low self-efficacy information (i.e., information regarding the individuals ability to complete an exercise program), and (b) high or low response-efficacy information (i.e., information about the effectiveness of such an exercise program in enhancing health and physical attractiveness). There are two distinctions between these studies and the general manipulations of health-education intervention studies. First, experimental manipulations are more direct in that they manipulate information about one or more particular PMT variables. Health-education studies present general factual information about a health threat and coping response, based on the PMT variables. Each individual variable is not directly addressed. Second, experimental manipulations present information that is manipulated in a high or IOU, manner. For example, one group receives information designed to increase the strength of a belief, such as vulnerability (high vulnerability), while the other group receives information designed to decrease vulnerability (low vulnerability). Comparisons are made between the high-vulnerability group and the lowvulnerability group. For example, Wurteles ( 1988) high-vulnerability group was given information on the incidence of bone loss in young women, along with several reasons why younger women may be at risk from osteoporosis. The low-vulnerability group was given information stressing that younger women are at low risk from osteoporosis. In most experimental manipulation studies, there is no control group that does not receive any information regarding the target variable. The exception to this is Wurtele and Maddux ( 1987) who gave their participants written communications containing high information on four, three, two, one. or none of the PMT variables. In this case, there was a control group that was given a passage on an unrelated topic. No groups were given low information on the PMT variables.4 In the health-education intervention studies, the difference between the two subject groups is that one group receives general information about a health threat, and the other group receives unrelated information; that is, a comparison is drawn between an experimental group that receives information and a control group that receives no information about the threat. There is no group receiving low information. Some experimental manipulation studies (e.g., Beck & Lund, 198 1 ; Wurtele, 1988) also included correlational measures of PMT variables that were not manipulated in the communication. These measures will be treated as correlational measures in the review.
4 A n ~ t h e exception r is a study by Sturges and Rogers ( I 996) that included a no-message control group in an experimental manipulation. However, this study is not included in the present review because the PMT variables were combined into threat and coping for the manipulation checks and data analysis. Analyses for individual PMT variables were not reported.

PROTECTION MOTIVATION THEORY

115

Details about all studies included in the analysis are shown in Table 1. Twenty-seven studies, with 29 independent samples and a total of 7,694 participants, fulfilled the criteria for inclusion. Of these studies, 15 involved correlational designs, 8 employed specific experimental manipulations, and 3 compared health education versus no education. There were 19 cross-sectional studies, 1 I of which measured intention only, 5 measured concurrent behavior only, and 3 measured both intention and concurrent behavior. Seven studies were longitudinal, and 6 of these included measures of both intention and subsequent behavior, and 1 measured subsequent behavior but not intention (Taylor & May, 1996). One study (Seydel et al., 1990) measured intention, concurrent behavior, and subsequent behavior. Most of the studies (14) used samples of high-school, college, or university students. General population samples were used in 7 studies, while 7 targeted specific groups (dental patients: Beck & Lund, 1981; homosexual men: Aspinwall et al., 1991; nurses: Millard, 1994; low- and high-risk drinkers: Ben-Ahron, White, & Phillips, 1995; parents of children with muscular dystrophy: Flynn, Lyman, & Prentice-Dunn, 1995; workers exposed to harmful noise levels: Melamed, Rabinowitz, Feiner, Weisberg, & Ribak, 1996; and patients of a university-based sports-injury clinic: Taylor & May, 1996).

Data Analysis
Two methods of data analysis were used: meta-analysis (Rosenthal, 1984) and vote count (Cooper, 1986). Meta-analysis was used to examine the overall strength of associations between threat- and coping-appraisal variables and intention, and all PMT variables with behavior. Vote counts were used to examine how often these associations were significant across the individual studies. For metaanalysis, all associations between variables needed to be bivariate relationships that could be converted to a single effect size (viz. Pearsons correlation). Many studies reported multivariate relationships, such as regression analysis. Authors of such papers were contacted, and the relevant correlation matrices were requested. There were four replies. Those studies where bivariate relationships were available were included in the meta-analysis. All studies were included in the vote counts. Meta-analyticprocedure. All hypotheses from the studies reporting bivariate relationships or where authors subsequently provided such data were included in the meta-analysis, which was conducted using Schwarzers (1988) computer program Meta. Statistics such as F and t values were converted to r values. The weighted average correlation (r+) was then calculated by multiplying each r-to-ztransformed correlation by its sample size and dividing by the total sample size. Thus, correlations from larger samples were given more weight in the metaanalysis than were those from smaller samples. The average correlation gives an overall effect size. Where a study addressed more than one hypothesis using the

Table 1

Applications of Protection Motivation Theory to Health-Related Behaviors

i rn
rn
-4

> r
No

Authors Yes No

Study design

Sample details

Focal behavior

Measure of Measure of Measure of concurrent subsequent intention behavior behavior

Abraham et al. ( 1994)

Aspinwall et al. (1991) Yes

Correlational, cross- N = 507; male and Condom use, HIV sectional study female school test, reducing pupils above 16; M sexual partners age = 16.8 years Correlational, N = 389; Reducing number of longitudinal study homosexual men; sexual partners, age range = 18-57 reducing number years, M age = 33 of anonymous years sexual partners, and reducing number of partners for unprotected anal and receptive intercourse Yes

No

Beck & Lund (1981)Experimental manipulation of PMT variables, cross-sectional study


N

Yes

Yes
No

Yes

No

No

No

Yes
L
=!
0

s
(fablecontinues)

;
I

= 80; male and Tooth flossing and female patients of brushing, and university dental plaque disclosure clinic; age range = 17-81 years; random assignment into experimental conditions Ben-Ahron et al. Correlational, cross- N = 196; male and Moderate drinking (1 995) sectional study female high-risk and lower risk drinkers HIV-related Bengel et al. (1996) Correlational, cross- N = 468; sectional study heterosexual preventive adults, aged 20-45 behavior; sexual years behavior, condom use, reduction of sexual partners and behavior change

<

-.
00

-L

Table 1 (Continued)

5 2
--I rn

Authors Yes

Study design

Sample details

Focal behavior

Measure of Measure of Measure of concurrent subsequent intention behavior behavior

D
!-

Boer & Seydel ( 1996)

No

Yes

Eppright et al. ( 1994)


No

Health-education N = 360; females, Breast-cancer manipulation (pro- invited to screening vision of health participate by education to exper- letter; age range = imental group); 50-70 years longitudinal study over 2 years Correlational, cross- N = 33 I ; male and AIDS-preventive sectional study female university behavior; adaptive students information seeking, reducing sexual partners, selecting infectionfree partners, avoidance of sharing body fluids, condom use Yes

No

Yes Yes

No

Yes
No

No

Yes

Yes

No

Yes

No

Yes

Flynn et al. (1995) Correlational, cross- N = 115; families Parents compliance sectional study attending clinics with childs sponsored by the muscular Muscular Dysdystrophy trophy Association physiotherapy Fruin et al. ( 1991) Experimental N = 6 15; male and Exercise manipulation of female high-school PMT variables, students; age cross-sectional range = 13- 17 years, M age = 14 study years 5 months; random assignment into experimental groups Hayes (1 996) Correlational, cross- N = 74; male and Sunscreen use and sectional study female underuse of sungraduate psyprotective clothing chology students Hodgkins & Orbell Correlational, N = 89; female Breast self( 1998) longitudinal study psychology examination over 1 month students and staff; age range = 17-40 years, M age = 2 1 years
--I

I rn 0 R

< (ruble continues)

-L -L

Table 1 (Continued)
.
h)

Authors
N
=

Study design
153; male and

Sample details Smoking cessation

Focal behavior

Measure of Measure of Measure of concurrent subsequent intention behavior behavior

z r
z
rn rn
4

Yes
No No

Maddux & Rogers ( 1983)

Experimental manipulation of PMT variables, cross-sectional study


No

D
!-

Melamed et al. ( 1996)

Yes

No

Millard ( 1 994)

female undergraduate students; random assignment into experimental conditions Correlational. cross- N = 28 1 ; Israeli men Use o f a hearing sectional study exposed to harmful protection device noise levels at work Correlational, N = I05 male and Nurses safety belongitudinal study female quali tied havior in relation and unqualified to AIDS and Hepanurses titis-B prevention; taking the sharps bin to the patient, recapping needles after use, renewing sharps bin when it reaches its maximum fill, wearing gloves when giving in.jections, etc. Yes

No

Yes

Plotnikoff & Higginbotham (1 995) Yes Yes


No Yes No No

No

Yes

No
T J

3
Yes Yes Yes

Correlational, cross- N = 800; males and Reduce dietary fat sectional study females identified from Federal Electoral Roll Rippetoe & Rogers Experimental N = 163; female Breast selfundergraduate examination (1987) manipulation of PMT variables, psychology cross-sectional students; random study assignment into experimental groups Ronis et al. (1 996) Correlational, cross- N = 622; adults over Frequencies of sectional study 18 years in Detroit toothbrushing, tooth flossing, and dental check-ups Seydel et al. (1990) Health-education N = 124 males and Checking for cancer, manipulation 132 females going to see doctor applied to take part when suspecting (provision of health education to following cancer, Pap test, experimental newspaper ad; age breast selfexamination, group); range = 19-73 4 age = 38 ordering leaflets on longitudinal study years, A (subsequent years cancer behavior based on orders of leaflets) (table continues)

Table 1 (Continued)

N N

-.

E
rn rn
I

Authors Yes

Study design

Sample details

Focal behavior

Measure of Measure of Measure of concurrent subsequent intention behavior behavior

Sheeran & Orbell ( 1996)

No

No

! -

Yes

No

No

Correlational, cross- N = 200; male and Tooth flossing and sectional study female condom use undergraduate psychology students N = 195; male and Signing up for an Stanley & Maddux Experimental ( 1986) manipulation of female exercise program PMT variables, undergraduate cross-sectional psychology study students; random assignment into experimental conditions Steffen (1990) Health-education N = 198; male Testicular selfmanipulation collegepsychology examination (provision of students; M age = health education to 2 1.2 years; random experimental assignment into group) experimental conditions Yes No

No

Tanner et al. (1989) Experimental manipulation of PMT variables


No No

N = 202; male and female college students


Condom use Yes

No

No

Taylor & May ( 1996) Yes

Yes

NO

No

Correlational, N = 62; students Compliance with the longitudinal study involved in sports recreational or physiotherapists competitive sports prescribed and patients of a modalities and university-based prescribed rest sports-injury clinic; 68% male; M age = 2 1.7 years Correlational, cross- N = 885; male and Substance use/ Umeh (in press) sectional study female secondary misuse, exercise, school females; fatty food age range = 13- 17 consumption, years, M age = 14 unprotected sex years van der Velde & van Correlational, cross- N = 23 I ; 147 male Condom use der Pligt ( 1991) sectional study homosexuals and 84 male and female heterosexuals; age range = 18-30 years Yes
No

No

(tclhle continues)

Table I (Continued)

5 rn
rn
--I

Authors Yes
No

Study design

Sample details

Focal behavior

Measure of Measure of Measure of concurrent subsequent intention behavior behavior Yes

D ! -

Wurtele ( 1 988)

Experimental N = 89; female Taking calcium manipulation of psychology supplements and PMT variables, students; age increasing dietary longitudinal study range = 17-26 calcium intake over 1 week years, M age = 19.2 years; random assignment into experimental groups Wurtele & Maddux Experimental N = 160; female Aerobic exercise (1 987) manipulation of undergraduate PMT variables, students; random longitudinal study assignment into over 2 weeks experimental groups Yes
No

Yes

PROTECTION MOTIVATION THEORY

125

same sample, the average correlation within that study was employed. Guidelines developed by Cohen (1992) were used to interpret the effect sizes generated by the meta-analysis: r+ = .I0 is interpreted as a small association; rt = .30 is interpreted as a medium association; and rt = .50 is interpreted as a large association. A chi-square test of homogeneity was conducted (Hunter. Schmidt, & Jackson, 1982) to ascertain whether variation among the correlations was greater than chance. A significant result shows that the effect sizes show greater variability than chance. The degrees of freedom for the chi-square statistic are k - I , with k being the total number of independent correlations included in the analysis. The fail-safe N was also calculated. Thefail-safe N (FSN) is the number of studies giving null results that would be needed to change the conclusion that two variables are significantly correlated at p < .05. This allows the meta-analytic results to be considered in relation to unretrieved or future studies. Rosenthal (1991) suggests a tolerance level of 5k + 10. When the FSN exceeds the tolerance level, a finding can be seen as robust (i.e., resistant to unretrieved studies). Vote-count procedure. The vote count (Cooper, 1986) was conducted on all retrieved studies. Each variable was looked at in turn, and the number of times it significantly predicted intention (threat- and coping-appraisal variables) and behavior (all PMT variables) was counted. Each individual association between a variable and a dependent measure constitutes a hypothesis in this context. The ratio of times that a hypothesis was significant in the predicted direction to the number of times that it was tested was calculated as a percentage to produce a significance ratio. For this analysis, correlational studies, health-education intervention studies, and experimental manipulation studies were looked at individually in order to draw comparisons between them. Findings for different designs were also combined to produce an overall significance ratio. Results The results of the meta-analyses correlating threat- and coping-appraisal variables with intention and all PMT variables with concurrent behavior and subsequent behavior are shown in Table 2. The results of the vote count of these associations are summarized in Table 3.5 Relationship Between Threat- and Coping-Appraisal Variables a n d Intention Twenty-one studies with a total of 206 individual hypotheses were included in the vote count. Of these, 12 studies (1 3 individual samples) involving a total of 52 hypotheses were suitable for use in the meta-analysis.
5Tables of hypotheses, study details, and findings for relationships between PMT variables and intention, concurrent behavior. and subsequent behavior are available from the authors upon request.

126

MILNE ET AL.

Table 2

Meta-Analysis of Correlations Between PMT Variables and Intentions. Concurrent Behavior. and Subsequent Behavior
FSNi (tolerance value)

Variables Associations with intention Threat appraisal Vulnerability Severity Fear Coping appraisal Self-efficacy Response efficacy Response costs

t+

' x
30.78*** 30.15*** 3.02, ns 74.67*** 45.08*** 3.90. ns

10

9 4 13
1 1

1,366 .16*** 1,196 .lo*** 411 .20*** 2.181 .33*** 1,756 .79*** 631 -.34***

22 (60) 12 ( 5 5 ) 12 (30)

74 (75) 57 (70) 23 (30)

Associations with concurrent behavior Threat appraisal Vulnerability 5 6 Severity 1 Fear Coping appraisal Sel f-efficacy 7 Response efficacy 6 4 Response costs Intention 2 Associations with subsequent behavior Threat appraisal Vulnerability 4 Severity 4 1 Fear Coping appraisal Self-efficacy 5

1,211 .13*** 1,391 .lo*** 157 A**


1,507 1,426 1,186 257 .36*** .17*** -.32*** .87***

14.95** 17.71***
-

8 (35) 6 (40) 1(15) 42 (45) 15 (40) 18 (30) 31 (20)

76.81*** 22.82*** 2.96, ns 52.90***

401 .12** 372 .07, ns 194 -.04, ns 512 .22***

2.20, ns 9.29* 0.08, ns


0.98, ns

5 (30) 2 (30) 1(20) 17 (35)

(table continues)

PROTECTION MOTIVATION THEORY

127

Table 2 (Continued) FSNI (tolerance value)

Variables Response efficacy Response costs Intention

rt

x
10.26* 0.79,ns 3.21,ns

4 2 4

388 .09, ns 194 -.25*** 432 .40***

3 (30) 8 (20) 28(30)

Note. k = number of hypotheses. N = total number of subjects. r + = sample weighted average correlation (effect size). Chi square is a homogeneity statistic, providing a test of the null hypothesis that the average effect size is a result of sampling error alone (df= k - I). FSN = Failsafe N. A FSN value greater than the tolerance value (i.e., Sk + 10; cf. Rosenthal, 1991) indicates that the average correlation is likely to be resistant to unretrieved and future null results at p < .05. * p < . O S . * * p < . O I .***p<.OOl.

Meta-analyticfindings. All PMT variables were found to be significantly correlated with intention in the predicted direction (all p s < .001). The association between severity and intention was small (r+ = .lo). All other associations between threat-appraisal variables and intention were small to medium (vulnerability, r+ = .16; fear, r+ = .20). Both self-efficacy (r+ = .33) and response costs (r+ = -.34) had medium associations with intention, and at r+ = .29, the association between response efficacy and intention could be interpreted as medium. Unfortunately, the FSN for all variables failed to reach Rosenthals (199 1) tolerance level of 5k + 10. However, with an FSN of 74, with k = 13, the FSN for the association between self-efficacy and intention was only 1 away from reaching the tolerance level. This suggests that the association between self-efficacy and intention is the most robust of all associations between PMT variables and intention, and the one most likely to prove resistant to unretrieved and future null results. Vote-count results. Self-efficacy had a significance ratio of 70% and was the variable that was most often significantly associated with intention in the predicted direction across studies. Vulnerability ( 3 1%) and severity (23%) were least often associated with intention across studies. Threat-appraisal variables were less often associated with intention than were coping-appraisal variables in correlational studies. However, threat-appraisal cognitions proved to be more successhl in predicting intention in experimental studies. Fear and response efficacy were associated with intention in both health-education interventions testing these hypotheses. Self-efficacy was found to be significantly associated with intention in all hypotheses tested by experimentally manipulating variables. KeyJindings. Overall, coping-appraisal variables proved to be more strongly and consistently associated with intention than did threat-appraisal variables. Self-

128

MILNE ET AL.

Table 3

Vote Counr Significance Ratios for PMT Variables and Intentions. Concurrent Behavior: and Subsequent Behavior
Healtheducation interventions k Ratio Manipulated variables k Ratio k

Correlated variables Variables Threat appraisal Vulnerability Severity Fear Coping appraisal Self-efficacy Response efficacy Response costs +/-

Total Ratio

Ratio

Associations with intention

+
+ +

27 24 20 29 21 10

22 22
9

9 9 2
8

11 22 100 87
100
-

66 50 0 6 100 50 0

45 39 24 43 44 11

31 23 46 70 47 45

+
+

59

30 5

8
1

Associations with concurrent behavior Threat appraisal Vulnerability + 15 66 Severity + 20 20 Fear + 5 60 Coping appraisal Self-efficacya + 23 70 Response efficacy + 19 21 Response costs 5 100 Intention + 2 100 Associations with subsequent behavior Threat appraisal Vulnerability + 15 3 Severity + 13 8 Fear + 6 0

4 4

20 20

2 2

0 50
-

4
100

21 26 5

52 19 60 76 30 100 100

100 100
-

29 23 5 6

100
-

2
-

100

3 3

0 33

3 1
-

33 0
-

21 17 6

14 12 0

(table continues)

PROTECTION MOTIVATION THEORY

129

Table 3 (Continued) Healtheducation interventions


k

Correlated variables Variables


+/-

Manipulated variables
k

Total
k

Ratio

Ratio

Ratio 3 0

Ratio
9

Coping appraisal Self-efficacy + Response + efficacy Response costs Intention +

13

46

3
3 -

66 66 -

42 29 43 66

8 25 7 4 3 11 64

3 1

0 100

4
7 2

Note. k = Number of hypotheses. + =positive association; - = negative association. Ratio = significance ratio, the percentage of times a PMT variable was significantly associated with concurrent behavior in the predicted direction. aThe vote-count results for self-efficacy include the behavior-specific hypotheses tested by Ronis et al. (1996) only (e.g., findings for the regression of toothbrushing on selfefficacy and response efficacy to perform flossing are not included in the vote count).

efficacy had the most robust and the most consistent association with intention of all the PMT variables and was also one of the strongest correlates of this variable.

Relationship Between PMT Variables and Concurrent Behavior


Studies measuring behavior in a cross-sectional design, that is, using measures of behavior taken at the same time as measuring the PMT variables, were used to assess the relationship between PMT variables and concurrent behavior. Twelve studies (13 individual samples) with a total of 58 individual hypotheses were included in the vote count. Of these, 8 studies (9 independent samples) were suitable for use in the meta-analysis. Mefa-analyficfindings.All PMT variables were significantly correlated with concurrent behavior in the predicted direction at p < .001, with the exception of fear ( p < .01). The largest association was found between intention and concurrent behavior (r+ = .82). The FSN for this association was above Rosenthals (1991) 5k + 10 (FSN = 3 1) indicating the result to be robust and resistant to unretrieved results and future research. Response efficacy had a small-to-medium association (rt= .17) with concurrent behavior. The remaining coping-appraisal components had medium-to-large associations (r+ = .36 for self-efficacy and r+ = -.32 for response costs). Only one study tested the relationship between fear and concurrent behavior, finding a small-to-medium association (rt= .26). Both

130

MlLNE ET AL.

vulnerability ( r t = .13) and severity (I+ = . l o ) had small associations with concurrent behavior. With the exception of intention, the FSNs for all variables did not reach Rosenthals (1991) tolerance level, although self-efficacy (FSN = 42) was once again close to the stipulated level. The FSNs for all threat-appraisal components were very small, indicating that only a small number of null results would be needed to change the conclusion that these variables are significantly associated with concurrent behavior at p < .05. Vote-count results. Intention was found to be significantly associated with concurrent behavior in all hypotheses testing the relationship. Self-efficacy (76%), response costs ( 1 OO%), and fear (60%) were the threat- and copingappraisal variables that most often predicted concurrent behavior. Only one health-education intervention study (with four hypotheses) involved concurrent behavior, and only two hypotheses were tested in experimental studies. Keyfindings. Intention had the strongest, most robust, and most consistent association with concurrent behavior. Self-efficacy and response costs were the threat- and coping-appraisal variables most strongly and frequently associated with concurrent behavior. Relationship Between PMT Variables and Subsequent Behavior Eight studies with a total of 96 individual hypotheses were included in the vote count. Of these, 5 studies involving a total of 23 hypotheses were suitable for use in the meta-analysis. Meta-analytic.findings. Intention had the largest correlation with subsequent behavior (r+ = .40,p < .001). The association was moderate to strong, according to Cohens (1992) criteria. The FSN for this association was above Rosenthals 5k + 10 (FSN = 28) criterion, indicating the result to be robust and resistant to unretrieved results and future research. Perceived vulnerability was the only threat-appraisal variable to be significantly correlated with subsequent behavior (rf = .12, p < .01). This was a small association with a very small FSN (FSN = 5). Of the coping-appraisal components, self-efficacy (rt = .22) and response costs (r+ = -.25) were significantly correlated with subsequent behavior, both at p < .OO 1. Both associations were small to medium and unstable., with FSNs failing to reach Rosenthals (1991) tolerance level (FSN = 14 and 8 for self-efficacy and response costs, respectively). Vote-count results. Intention was the variable that was most often associated with subsequent behavior and had a significance ratio of 66%. Of the threat- and coping-appraisal variables, response costs and self-efficacy were most often associated with subsequent behavior across all of the studies, with significance ratios equal to 43% and 42%, respectively. Fear did not have statistically significant associations in any of the hypotheses tested, while vulnerability and severity

PROTECTION MOTIVATION THEORY

131

both had small overall significance ratios (14% and 17%, respectively). Only one study, generating three hypotheses, included measures of subsequent behavior following a health-education intervention; and one study, also involving three hypotheses, tested subsequent behavior following an experimental manipulation. Keyfindings. As PMT predicts, intention had the strongest, most robust, and most consistent association with subsequent behavior. Perceived vulnerability, self-efficacy, and response costs were the only threat- and coping-appraisal variables that were associated with subsequent behavior. Response costs and selfefficacy were the threat- and coping-appraisal variables that were most often significantly associated with subsequent behavior. Cognition Changes Following Manipulations of PMT Variables Experimental manipulations of PMT variables usually involve a written communication focusing on a particular PMT variable. There are usually several experimental groups-two (high or low) for each variable being manipulated. For example, if an experimenter wants to manipulate perceived vulnerability and perceived self-efficacy, he or she would use four experimental groups: one receiving high vulnerability information, one receiving low vulnerability information, one receiving high self-efficacy information, and one receiving low selfefficacy information. Manipulations in health-education studies are presented in a more realistic way. Information is presented in a manner that is more characteristic of health-education programs, for example, a television program (Seydel et al., 1990) or a health-education brochure or leaflet (Boer & Seydel, 1996; Steffen, 1990).6 Meta-analyticfindings. Findings from the meta-analysis of cognition changes following specific manipulations of threat- and coping-appraisal variables are shown in Table 4. Health-education intervention studies were not included in the meta-analysis since there were too few studies to permit meaningful inferences. With the exception of response costs (r+7 = .09, ns), all associations between manipulations of threat- and coping-appraisal variables and subsequent cognition change were significant at p < .001. Vulnerability (r+ = .63) and severity (r+ = .66) manipulations had large associations with subsequent cognition change (Cohen, 1992). Fear was manipulated twice, and a medium association was found between these manipulations and subsequent change in belief (r+ = .26). Self-efficacy manipulations were moderately associated with subsequent changes in this belief (r+ = .32). The remaining associations were medium to strong. Response costs were manipulated in just one study, and thus had a FSN of 1.
6A table showing the types of information used to manipulate PMT variables and their success in bringing about cognitive change is available from the authors upon request. 'Although the statistic d is the more commonly used measure of effect size in experimental studies, t+ was employed here for reasons of consistency and ease of interpretation.

132 MILNE ET AL. Table 4

Meta-Analysis of Changes in Cognitions Following Experimental Manipulations of PMT Variables


FSN (tolerance value)

Variables Threat appraisal Vulnerability Severity Fear Coping appraisal Self-efficacy Response efficacy Response costs

N
561 48 1 207 1,163 1.163 563

1 3

x2
11.78* 46.58*** 4.18* 83.52*** 54.83***
-

5 4
7

.63***

.66***
.26**

58 (35) 49 (30) 8 (20) 27 (35) 37 (35) I(15)

5 5
1

.32*** .42*** .09, ns

Note. k = number of hypotheses. N = total number of subjects. r+ = sample weighted average correlation (effect size). Chi square is a homogeneity statistic, providing a test of the null hypothesis that the average effect size is a result of sampling error alone (+ k - 1). FSN = Failsafe N. A FSN value greater than the tolerance value (i.e., 5k + 10; cf. Rosenthal, 1991) indicates that the average correlation is likely to be resistant to unretrieved and future null results a t p < .05. *p < .05. **p < . O l . ***p < ,001.

With the exception of self-efficacy (FSN = 27), the associations were all above Rosenthals ( 1991 ) tolerance level. Vote-count results. The results of the vote count of cognition changes following manipulations of threat- and coping-appraisal variables are shown in Table 5 . All variables had significance ratios greater than 50%. Experimental manipulations produced significant changes in belief in all of the hypotheses tested (with the exception of response costs, which has only been manipulated once [Fruin et al., 19911 and unsuccessfully). There were very few hypotheses involving health-education manipulations. Provision of health-education information did not lead to changes in any of the threat-appraisal variables. There was a change in belief in one of the hypotheses measuring response efficacy and in the only study of se I f-efficacy. Keyfindings. All threat- and coping-appraisal variables-with the exception of response cost-were successfully manipulated in experimental manipulations. The variables with the strongest relationships with intention and behavior were those that proved to be the hardest to manipulate to bring about belief change. The association between response costs and subsequent change in belief, although

PROTECTION MOTIVATION THEORY

133

Table 5
Vote Count Changes in Cognitions Following Experimental Manipulations of PMT Variables and Health-Education Interventions

Manipulated variables Variables Threat appraisal Vulnerabi 1ity Severity Fear Coping appraisal Self-efficacy Response efficacy Response costs
+/-

Healtheducation interventions
k

Total
k

Ratio 100 100 100


100 100

Ratio 2 2

Ratio

+ + +
+
+

5
4 2

0 0
-

7 6
4

57 66 100
70 86
100

5 5 1

100

2
-

50 0

43 7 1

Nofe. k = Number of hypotheses. + = positive association; - = negative association. Ratio = significance ratio, the percentage of times a PMT variable was significantly associated with concurrent behavior in the predicted direction.

only tested once, was nonsignificant; and of the significant belief changes, selfefficacy was the only one where the FSN failed to reach Rosenthals (1 991 ) tolerance level. Experimental manipulations appear to be more successful than healtheducation interventions in changing threat- and coping-appraisal cognitions. Discussion This is the first meta-analytic review to evaluate the success of PMT in the prediction of health-related intentions and behaviors. The questions addressed were: How well do threat appraisal and coping appraisal predict intention?; How well do PMT variables predict concurrent behavior and subsequent behavior?; and How successfi~l have manipulations of PMT variables been in bringing about changes in belief? The findings offer modest support for the threat- and coping-appraisal components of the model in predicting health-related intentions. All threat- and coping-appraisal variables were significantly associated with intention. However, coping-appraisal components of the model had stronger associations with intention than did the threat-appraisal cognitions. Although none of the associations had FSNs reaching Rosenthals (199 1) tolerance level, the association between

134

MILNE ET AL

self-efficacy and intention was very close to this threshold and is therefore more robust and resistant to future research than are the associations obtained for the other variables. PMT variables were all found to be significantly associated with concurrent behavior. The association between intention and concurrent behavior was the strongest and most consistent association found in the meta-analysis. The average correlations between threat-appraisal variables and concurrent behavior were small, with very small FSNs. At most, it would take only four studies finding nonsignificant relationships between perceived vulnerability. perceived severity, or fear to produce a nonsignificant average correlation. The coping-appraisal components were more strongly associated with concurrent behavior than were threat-appraisal variables, although with the exception of self-efficacy, these too had FSNs that were far from the tolerance level (Rosenthal, 1991). The vote count supported these findings, with coping-appraisal variables having much higher significance ratios than threat-appraisal cognitions. As predicted by PMT. health-related intentions were signifcantly associated with subsequent behavior. Intention had a medium-to-strong average correlation with subsequent behavior. This was a robust relationship, and the vote count showed good consistency in the relationship across PMT research. The strength of the association that was found between intention and behavior supports PMT since the model predicts that intention will be the best and most immediate predictor of behavior. The strength of the association between intention and behavior was consistent with findings from previous meta-analytic reviews. For example, Randall and Wolf (1 994) and Sheeran and Orbell (1998) found average correlations of .45 and .44, respectively, in their meta-analyses of the intentionbehavior relationship. The only threat- and coping-appraisal variables found to be significantly associated with subsequent behavior were perceived vulnerability, self-efficacy, and response costs. Although the average correlation between perceived vulnerability and subsequent behavior was significant, it was small and was not robust. The vote count showed that this association was not generally significant across the studies. The present study supports previous findings showing that threat appraisal is a poor predictor of intention and behavior (e.g., Abraham et al., 1994; Harrison et al., 1992; Hodgkins & Orbell, 1998; Maddux & Rogers, 1983). Difficulties with statistical interpretation and measurement may have been responsible for the weak association obtained between threat-appraisal components and intention and behavior. Weinstein and Nicolich (1993) highlight the problem of interpreting correlations between perceived personal risk and behavior in cross-sectional studies. This is because both positive and negative associations between risk and behavior are possible. If a person feels vulnerable to a health threat, they may decide to adopt a protective behavior, then a positive relationship between the two variables will be obtained. However, once the protective behavior has been

PROTECTION MOTIVATION THEORY

135

adopted, the individual may no longer feel vulnerable to the threat, therefore the association between perceived vulnerability and behavior will be negative. One possible explanation for the small overall average correlation between perceived vulnerability and concurrent behavior in the individual PMT studies may be that both positive and negative relationships between the two variables exist among the data. However, all of the associations between perceived vulnerability and concurrent behavior in the studies included in the vote count were positive. Thus, it is unlikely that this is the case in the present study. A related difficulty with cross-sectional studies is that it cannot be determined whether perceived-vulnerability beliefs influence behavior or vice versa. Weinstein and Nicolich (1993) maintain that only in certain circumstances can a causal inference be made about the correlation between vulnerability and behavior; that is, when future behavior is measured in a longitudinal study and perceived vulnerability is measured immediately after the threat communication has been presented to the participants, before they have an opportunity to change their behavior. Two studies have tested the association between perceived vulnerability and future behavior under these conditions (Wurtele, 1988; Wurtele & Maddux, 1987). Wurtele found a positive significant relationship, as predicted, while Wurtele and Maddux found a nonsignificant association. Therefore, this relationship needs to be tested further under the conditions stipulated by Weinstein and Nicolich before a conclusion can be reached about the ability of perceived vulnerability to predict behavior. Weinstein ( 1988) also highlights problems in measuring perceived vulnerability when participants are asked to estimate their personal vulnerability to a threat but are not given an option to say that they are unaware of the threat in question. All of the PMT studies included in the present review have measured vulnerability in this problematic way. This traditional operationalization of vulnerability assumes the variable to be static. Weinstein proposes that this static conceptualization is misguided and that perceived vulnerability develops in a series of three stages. The individual hears of the threat and becomes aware that it exists. This is followed by an assessment of how dangerous the threat is and how many people are likely to be affected. It is not until the final stage that the threat is personalized and the individual can give an estimate of his or her own personal perceived vulnerability. One possible reason why the average r+ and significance ratios for severity were poor may be that it is often very difficult to obtain variability in the data for perceived severity (Harrison et al., 1992; Janz & Becker, 1984). For example, few people would disagree that contracting cancer or AIDS would be serious for them. However, severity can be seen as a multidimensional construct. As well as the physical severity of the disease, such as pain or premature death, perceived severity can be assessed in terms of psychosocial severity-for example, how much the disease is likely to interfere with social roles that are important to the

136

MILNE ET AL.

individual, such as the ability to work (Sheeran & Abraham, 1996). Perceived severity has been operationalized in terms of both of these dimensions in applications of PMT to health-related behavior. For example, Abraham et al. (1994) operationalized perceived severity in terms of evaluating the fatality of AIDS: How many people who get the AIDS virus actually die of it? Hodgkins and Orbell ( 1 998) used items including those measuring psychosocial severity (e.g., Developing breast cancer would force me to change my goals in life). Other dimensions of severity have been highlighted, including the immediacy of the onset of the disease (near vs. distant), the visibility of symptoms (high vs. low), and rate of onset (gradual vs. sudden; Smith-Klohn & Rogers, 1991). Further research is needed in order to determine whether multidimensional measures of severity would be better predictors of intentions and behavior. The rewards component of PMT was not included in the present review because only one study was obtained that included this variable (Abraham et al., 1994). Abraham et al. had difficulties in trying to operationalize rewards within the context of giving up unprotected sex because the conceptual distinction between the reward value of a risk behavior and cost of a preventative measure may not be clear (Abraham et al., 1994, p. 271). The fact that sex is less exciting without a condom could also be viewed as a response cost. This difficulty in operationalizing rewards may be the reason why it appears to have been neglected in most PMT research. Ronis and Hare1 ( 1 989) found that the effects of severity on intention were not direct but were mediated by another variable. Their research combined components of the health belief model and subjective utility theory in a study of breast self-examination. They found that the effects of severity were entirely mediated by the component of the model concerned with benefits of adopting the recommended response. it may be that the effects of threat-appraisal variables are mediated by other components of PMT. Another possibility is that perceived vulnerability, perceived severity, and fear appraisal may be more strongly and consistently associated with intention and behavior for some people but not for others. For example, Brouwers and Sorrentino ( 1993) found that an individual difference variable-uncertainty orientation-moderated the impact of perceived threat. A significantly greater effect of perceived threat was obtained for uncertainty-oriented compared to certainty-oriented participants. It has also been suggested that the severity of the disease and the complexity of the behavior may influence the role of perceived vulnerability (Montgomery et al., 1989). Where the threat is perceived as severe and the behavior is complex (e.g., an individual may perceive AIDS to be a serious disease and perceive that using a condom is a complex behavior), the role of perceived vulnerability may be diluted. This suggests that intervening variables may be involved in the relationship between vulnerability and intention and behavior. Given the diffculties associated with measuring and interpreting findings for threat-appraisal

PROTECTION MOTIVATION THEORY

137

variables, it would be desirable for further research to establish whether threat appraisal is of limited predictive utility or whether its effects are being masked by mediating or moderating variables. Coping appraisal was found to be of greater utility than threat appraisal in the prediction of health-related intentions and behavior. Meta-analysis showed that self-efficacy was the variable that was most strongly related to intention and concurrent behavior, and, with the exception of the association between intention and behavior, this was the only average correlation that was close to being robust. The vote count also showed that self-efficacy was the PMT variable that was most consistently associated with intentions and subsequent behavior (while response costs were slightly more frequently associated with behavior than selfefficacy; these data come from just two studies). The present study thus adds further support to the growing evidence suggesting that self-efficacy is a major factor in determining both motivation and health-protective behavior (Schwarzer 8~ Fuchs, 1996). These findings support the view that health-education interventions should seek to enhance self-efficacy with regard to the behavior in question (Abraham et al., 1994). An important strength of PMT research has been that predictions have often been tested in experimental studies. This provides an opportunity to establish how successful experimental manipulations have been in bringing about changes in beliefs. There were too few studies involving health-education manipulations to be included in the meta-analysis. However, experimental manipulations of PMT variables were generally very successful in bringing about subsequent changes in belief. Although, response costs were only manipulated once (with unsuccessful cognition change), large and highly significant effects were found for manipulations of all other variables. It is interesting to note that manipulations of perceived severity and perceived vulnerability had stronger associations with subsequent changes in belief than did manipulations of coping-appraisal variables. This is important since coping appraisal was found to be more successhl in predicting health-related intentions and behavior than was threat appraisal. While self-efficacy was the best predictor of intention and behavior, self-efficacy manipulations had only a medium association with subsequent cognitive change. This suggests that future research is needed to establish the best way to enhance self-efficacy beliefs. Bandura (1991) suggests that the optimal strategy for increasing an individuals perceived selfefficacy is to provide direct experience with the behavior; for example, through the use of role play. Research to date has tended to rely on persuasive communication to increase self-efficacy. Future studies might profit from employing Banduras recommended strategy. The provision of general health education did not bring about subsequent cognitive change as often as did manipulations of specific variables. However, since there was only a small number of studies using health-education manipulations,

138 MILNE ET AL.


these findings can only be taken as suggestive. It seems that health education, in the form of providing balanced factual information, is not a successful method of bringing about cognitive change. Specific cognitive manipulations (e.g., telling one group of participants that they are very likely to contract lung cancer from smoking and another group that they are very unlikely to do so) are more successful in bringing about subsequent changes in belief. This procedure is antithetical to real-world health-education programs, not least because there is no information between high vulnerability and low vulnerability provided. In the specific experimental manipulation studies included in the review, there were generally no control groups that received no information, just high- versus low-information groups. Further research is needed to establish how best to influence cognitive change using more factual health-education information. A vote-count was conducted to compare the effectiveness of variables manipulated during experimental studies and those manipulated in health-education designs in changing intention and behavior. Threat-appraisal variables measured following specific manipulations generally predicted intention better than did those measured following a health-education communication or measurements of existing cognitions. Both coping-appraisal variables measured following specific manipulations and those measured following health education predicted intention better than did correlational designs. Unfortunately, since only two studies tested the ability of the model to predict behavior, comparisons between variables measured using the two types of experimental designs cannot be made with confidence. Wurtele and Maddux (1987) suggest that the persuasive messages used in experimental manipulations may be effective in enhancing intention to change behavior, but may be less useful in producing actual behavior change. Experimental studies have yet to explore how persistent the subsequent cognitive change is. Measures of cognitive change are invariably taken immediately after presenting the information. This may explain why manipulated variables are often better associated with intention than with behavior. If intention is measured in the laboratory shortly after the intervention, the effects of the manipulation are fresh in the minds of the participants. By the time the behavioral measure takes place, the effects of the intervention may have worn off (Weinstein, 1988). Future research will need to examine cognitive change over longer periods than immediate post-tests, and examine subsequent behavior over longer periods. The present review shows that the processes of threat and coping appraisal have modest utility in predicting intentions to protect oneself against a health threat. However, our findings also show that intentions are satisfactory predictors of health behaviors-as PMT proposes-although other PMT variables have only small-to-medium average correlations with precautionary actions. Overall, coping-appraisal variables had greater utility in the prediction of intention and behavior than did threat-appraisal variables. Relatively few studies examined how well PMT predicted subsequent behavior, perhaps because of problems in

PROTECTION MOTIVATION THEORY

139

operationalizing some of the models variables (e.g., rewards). Thus, further research is needed in order to clearly establish the utility of PMT in predicting behavior. Wurtele and Maddux (1987) have acknowledged the failure of PMT to produce consistent predictive associations, particularly for threat-appraisal variables. However, they suggest that the model is still useful in its contribution to understanding the arguments that should be contained in persuasive communications designed to produce belief change and subsequent changes in intentions and, perhaps, behavior. Experimental PMT studies have demonstrated that information can be manipulated so as to successfully change beliefs. Future research should try to examine how best to incorporate these findings into public-health intervention programs. References *Abraham, S. C. S., Sheeran, P., Abrams, D., & Spears, R. (1994). Exploring teenagers adaptive and maladaptive thinking in relation to the threat of HIV infection. Psychology and Health, 9. 253-272. Ajzen, I. (1988). Attitudes. personality, and behavior: Milton Keynes, UK: Open University Press. Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision Processes, 50, 179-21 1. Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predicting social behavior: Englewood Cliffs, NJ: Prentice Hall. Armitage, C. J., & Conner, M. (1996). Meta-analysis of the theory of planned behavior. Proceedings of the British Psychological Society, 4,48. *Aspinwall, L. G., Kemeny, M. E., Taylor, S. E., Schneider, S. G., & Dudley, J. P. (1 99 1). Psychosocial predictors of gay mens AIDS risk-reduction behavior. Health Psychology, 10,432-444. Axelrod, L. J., & Newton, J. W. ( 1 991). Preventing nuclear war: Beliefs and attitudes as predictors of disarmist and deterrentist behavior. Journal of Applied Social Psychology, 21,29-40. Bandura, A. (1977). Self-efficacy: Toward a unifying theory of behavioral change. Psychological Review, 84, 191-215. Bandura, A. (1986). Social foundations of thought and action. Englewood Cliffs, NJ: Prentice Hall. Bandura, A. (199 1). Social cognitive theory of self-regulation. Organizational Behavior and Human Decision Processes, 50, 248-287. *Beck, K. H., & Lund, A. K. (1981). The effects of health threat seriousness and personal efficacy upon intentions and behavior. Journal of Applied Social PSyChology, 11,401-415. Becker, M. H. (1974). The health belief model and sick role behavior. Health Education Monographs, 2,409-4 19.

140

MILNE ET AL.

*Ben-Ahron, V., White, D., & Phillips, K. (1995). Encouraging drinking at safe limits on single occasions: The potential contribution of protection motivation theory. Alcohol and Alcoholism, 30,633-639. *Bengel, J., Belz-Merk, M., & Farin, E. (1996). The role of risk perception and efficacy cognitions in the prediction of HIV-related preventive behavior and condom use. Psychology and Health, 11, 505-525. *Boer, H., & Seydel, E. R. (1996). Protection motivation theory. In M. Conner & P. Norman (Eds.), Predicting health behavior: Research and practice with social cognition models (pp. 95- 120). Buckingham, UK: Open University Press. Brouwers, M. C., & Sorrentino, R. M. ( 1 993). Uncertainty orientation and protection motivation theory: The role of individual differences in health compliance. Journal of Personality and Social Psychology, 65, 102- 1 12. Cohen, J. (1992). A power primer. Psychological Bulletin, 112, 155-1 59. Cooper, H. M. (1986). Integrating research: A guide for literature reviews. London, UK: Sage. Edwards, W. ( 1 954). The theory of decision making. Psychological Birlletin, 111, 380-417. *Eppright, D. R., Tanner, J. F., Jr., & Hunt, J. B. (1994). Knowledge and the ordered protection motivation model: Tools for preventing AIDS. Journal o f Business Research, 30, 13-24. Fishbein, M., & Ajzen, 1. (1975). Belief; attitude, intention, and behavior. New York, NY: John Wiley & Sons. *Flynn, M. F., Lyman, R. D., & Prentice-Dunn, S. (1995). Protection motivation theory and adherence to medical treatment regimens for muscular dystrophy. Journal of Social and Clinical Psychology, 14,6 1-75. *Fruin, D. J., Pratt, C., & Owen, N. (1991). Protection motivation theory and adolescents perception of exercise. Journal of Applied Social Psychology, 22, 55-69. Godin, G., & Kok, G. (1996). The theory of planned behavior: A review of its applications to health-related behaviors. American Journal of Health Promotion, 11, 87-98. 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, 107116. *Hayes, L. (1 996). Protection motivation theoty in an attempt to account,jorfactors related to sun-protection behavior Unpublished undergraduate dissertation, University of Shefield, Shefield, UK. *Hodgkins, S., & Orbell, S. (1998). Can protection motivation theory predict behavior? A longitudinal test exploring the role of previous behavior. Psychology and Health, 13, 231-25 1. Hovland, C., Janis, I. L., & Kelley, H. (1953). Communication and perxuasion. New Haven, CT: Yale University Press.

PROTECTION MOTIVATION THEORY

141

Hunter, J. E., Schmidt, F. L., & Jackson, G. B. (1982). Meta-analysis: Cumulating researchfindings across studies. Beverly Hills, CA: Sage. Janis, I. L. (1 967). Effects of fear arousal on attitude change: Recent developments in theory and experimental research. In L. Berkowitz (Ed.), Advances in experimental socialpsychology (Vol. 3, pp. 166-224). New York, N Y Academic Press. Janz, N., & Becker, M. H. (1984). The health belief model: A decade later. Health Education Quarterly, 11, 1-47. Kantola, S. J., Syme, G. J., & Nesdale, A. R. (1983). The effects of appraised severity and efficacy in promoting water conservation: An informational analysis. Journal of Applied Social Psychology, 13, 164-182. Leventhal, H. (1970). Findings and theory in the study of fear communications. In L. Berkowitz (Ed.), Advances in experimental social psychology (Vol. 5, pp. 1 19- 186). New York, NY: Academic. Maddux, J. E. (1993). Social cognitive models of health and exercise behavior: An induction and review of conceptual issues. Journal of Applied Sport Psychology, 5, 116- 140. *Maddux, J. E., & Rogers, R. W. (1983). Protection motivation theory and selfefficacy: A revised theory of fear appeals and attitude change. Journal of Experimental Social Psychology, 19,242-253. *Melamed, S., Rabinowitz, S., Feiner, M., Weisberg, E., & Ribak, J. (1996). Usefulness of the protection motivation theory in explaining hearing protection device among male industrial workers. Health Psychology, 15, 209-2 15. *Millard, H. (1994). AIDS and nurses 'protective behavior at work: Wh-vdo the). ignore the risks? Exploring their thinking using a combined model of the theories of protection motivation. reasoned action, and identity. Unpublished undergraduate dissertation, University of Shefield, Shetxeld, UK. Montgomery, S. B., Joseph, J. G., Becker, M. H., Ostrow, D. G., Kessler, R. C., & Kirscht, J. P. ( 1 989). The health belief model in understanding compliance with preventive recommendations for AIDS: How useful? AIDS Education and Prevention, 1, 303-323. Mulilis, J. P., & Lippa, R. (1 990). Behavior changes in earthquake preparedness due to negative threat appeals: A test of protection motivation theory. Journal of Applied Social Psychology, 20,619-638. *Plotkinoff, R. C., & Higginbotham, N. (1995). Predicting low-fat diet intentions and behaviors for the prevention of coronary heart disease: An application of protection motivation theory among an Australian population. Psychology and Health, 10, 397-408. Randall, D. M., & Wolf, J. A. (1994). The time interval in the intention-behavior relationship: Meta-analysis. British Journal of Social Psychology, 33,405-4 18. *Rippetoe, P. A,, & Rogers, R. W. (1 987). Effects of components of protection motivation theory on adaptive and maladaptive coping with a health threat. Journal of Personaliw and Social Psychology, 52, 596-604.

142 MILNE ET AL.

Rogers, R. W. (1975). A protection motivation theory of fear appeals and attitude change. The Joiirnal of Psychology, 91,93- 1 14. Rogers, R. W. (1983). Cognitive and physiological processes in fear appeals and attitude change: A revised theory of protection motivation. In B. L. Cacioppo & L. L Petty (Eds.), Social psychophysiologv: A sourcebook (pp. 153- 176). London, UK: Guilford. 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, NY: Plenum. *Ronis, D. L., Antonakos, C. L., & Lang, W. P. (1996). Usefulness of multiple equations for predicting preventive oral health behaviors. Health Education Quarterly, 23, 5 12-527. Ronis, D., & Harel, Y. (1989). Health beliefs and breast self-examination behaviors: Analysis of linear structural relations. Psychology and Health, 3,259-285. Rosenstock, I. M. (1966). Why people use health services. Millbank Metnorial Fund Quarterly, 4 4 9 4 - 124. Rosenthal, R. (1991). Meta-analytic procedures for social research (2nd ed.). Newbury Park, CA: Sage. Schwarzer, R. (1988). Meta: Programs for secondary data analysis. Berlin, Germany: Free University of Berlin. Schwarzer, R., & Fuchs, R. (1996). Self-efficacy and health behaviors. In M. Conner & P. Norman (Eds.). Predicting health behavior: Research andpr-nctice with social cognition models (pp.163-196). Buckingham, UK: Open University Press. *Seydel, E., Taal, E., & Wiegman, 0. (1990). Risk appraisal, outcome and selfefficacy expectancies: Cognitive factors in previous behavior related to cancer. Psychology and Health, 4,99- 109. Sheeran, P., & Abraham, C. (1996). The health belief model. In M. Conner & P. Norman (Eds.), Predicting health behavior: Research and practise with social cognition models (pp. 23-6 1 ). Buckingham, UK: Open University Press. *Sheeran, P., & Orbell, S. (1996). How confidently can we infer health beliefs from questionnaire responses? Psychology and Health, 11,273-290. Sheeran, P., & Orbell, S. (1998). Do intentions predict condom use? Metaanalysis and examination of six moderator variables. British Jourrial of'Social Psychology, 37,23 1-250. Sheppard, B. H., Hartwick, J., & Warshaw, P. R. ( 1 988). The theory of reasoned action: A meta-analysis of past research with recommendations for modifications and future research. Journal of Consumer Research, 15, 325-339. Smith-Klohn, L., & Rogers, R. W. (1991). Dimensions of severity ofhealth threats: The persuasive effects of visibility, time of onset, and rate of onset on young women's intentions to prevent osteoporosis. Health Psychology, 10, 323-329.

PROTECTION MOTIVATION THEORY

143

*Stanley, M. A., & Maddux, J. E. (1 986). Cognitive processes in health enhancement: Investigation of a combined protection motivation and self-efficacy model. Basic and Applied Social Psychology, 7, 101- 1 13. 'Steffen, V. J. (1990). Men's motivation to perform the testicle self-exam: Effects of prior knowledge and an educational brochure. Journal of Applied Social PSYC~O~OQ, 20,68 1-702. Sturges, J. W., & Rogers, R. W. (1996). Preventive health psychology from a developmental perspective: An extension of protection motivation theory. Health Psychology, 15, 158- 166. *Tanner, J. F., Jr., Day, E., & Crask, M. R. (1989). Protection motivation theory: An extension of fear appeals theory in communication. Journal of Business Research, 19, 267-276. *Taylor, A. H., & May, S. ( I 996). Threat and coping appraisal as determinants of compliance with sports injury rehabilitation: An application of protection motivation theory. Journal of Sports Sciences, 14,47 1-482. *Umeh, F. (in press). Protection motivation, coping styles and health-related behavioral intentions amongst adolescents. Psychology and Health. *van der Velde, F. W., & van der Pligt, J. (1991). AIDS-related health behavior: Coping, protection motivation, and previous behavior. Journal o f Behavioral Medicine, 14,429-45 1. Weigman, O., Taal, E., Van den Bogaard, J., & Gutteling, J. M. (1992). Protection motivation theory variables as predictors of behavioral intentions in three domains of risk management. In J. A. M. Winnubst & S. Maes (Eds.), Lifestyles, stress, and health (pp. 55-70). Leiden, The Netherlands: DSWO. Weinstein, N. D. (1988). The precaution adoption process. Health Psychology, 7, 335-386. Weinstein, N. D., & Nicolich, M. (1993). Correct and incorrect interpretations of correlations between risk perceptions and risk behaviors. Health Psychology, 12, 235-245. Wolf, S., Gregory, W. L., & Stephan, W. G. (1986). Protection motivation theory: Prediction of intentions to engage in anti-nuclear war behaviors. Journal of Applied Social Psychology, 16, 3 10-321. *Wurtele, S. K. ( I 988). Increasing women's 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 protection motivation theory components in predicting exercise intentions and behavior. Health Psychology, 6,453-466.
Note. Asterisk ( * ) indicates papers included in the meta-analysis and vote count.

S-ar putea să vă placă și