The Health Belief Model and smoking cessation behaviours
Meta-Analysis of the Relationship Between Risk Perception and Health Behavior: The Example of Vaccination
Noel T. Brewer University of North Carolina at Chapel Hill
Gretchen B. Chapman Rutgers University
Frederick X. Gibbons and Meg Gerrard Iowa State University
Kevin D. McCaul North Dakota State University
Neil D. Weinstein Rutgers University
Background: Risk perceptions are central to many health behavior theories. However, the relationship between risk perceptions and behavior, muddied by instances of inappropriate assessment and analysis, often looks weak. Method: A meta-analysis of eligible studies assessing the bivariate association between adult vaccination and perceived likelihood, susceptibility, or severity was conducted. Results: Thirty-four studies met inclusion criteria (N � 15,988). Risk likelihood (pooled r � .26), susceptibility (pooled r � .24), and severity (pooled r � .16) significantly predicted vaccination behavior. The risk perception� behavior relationship was larger for studies that were prospective, had higher quality risk measures, or had unskewed risk or behavior measures. Conclusions: The consistent relationships between risk perceptions and behavior, larger than suggested by prior meta-analyses, suggest that risk perceptions are rightly placed as core concepts in theories of health behavior.
Keywords: perceived likelihood, perceived severity, perceived susceptibility, vaccination, influenza
Risk perceptions (i.e., beliefs about potential harm) are compo- nents of most theories of health behavior, but the strength of the relationships between these perceptions and behavior is unclear. Obtaining a better understanding of the size of these relationships can inform health behavior theory and guide intervention devel- opment. This article describes a meta-analysis of the associations between risk perceptions and behavior for one particular health- protective action: vaccination against infectious disease. Because we are aware of no experimental studies that have examined how manipulating perceived risk affects vaccination, all the data used in the meta-analysis are correlational (i.e., cross-sectional and longitudinal).
Reasons for Uncertainty of the Risk Perception�Behavior Relationship
The role of risk perceptions in shaping health behaviors is a fundamental, undecided issue in health psychology. Neither theo- ries of heath behavior nor empirical studies appear to agree about the importance of these perceptions. Risk perception is central to most health-specific behavioral theories (for reviews, see Sutton, 1987; Weinstein, 1993) including the health belief model (Rosen- stock, 1974), protection motivation theory (Rogers, 1975), and the extended parallel process model (Witte, 1992). Similarly, the self- regulation model (Leventhal, Meyer, & Nerenz, 1980) includes several constructs important to risk perception (Cameron, 2003). Many general behavioral theories are frequently applied to health behaviors (e.g., the theory of reasoned action, Fishbein & Ajzen, 1975; the theory of planned behavior, Ajzen, 1985; subjective expected utility theory, Ronis, 1992). These general theories posit that the likelihood and magnitude of potential outcomes (including non-health costs and benefits) shape behavior, but studies testing these theories assess the anticipated likelihood and magnitude of potential health-specific harms (i.e., risk perceptions) only if par- ticipants in pilot studies mention them.
Although the majority of empirical studies find positive associ- ations between risk perceptions and behaviors, as many theories suggest, individual studies report all types of relationships: posi- tive, negative, and none. In meta-analyses, the effect sizes found for risk perceptions tend to be significant but small. For example, in a review of 17 studies based on the health belief model,
Noel T. Brewer, Department of Health Behavior and Health Education, School of Public Health, University of North Carolina at Chapel Hill; Gretchen B. Chapman, Department of Psychology, Rutgers University; Frederick X. Gibbons and Meg Gerrard, Department of Psychology, Iowa State University; Kevin D. McCaul, Department of Psychology, North Dakota State University; Neil D. Weinstein, Department of Human Ecol- ogy, Rutgers University.
Portions of this article were presented at the 2005 Annual Conference of the Society for Behavioral Medicine in Boston, MA. Planning for this meta-analysis was supported by the Behavioral Research Program, Na- tional Cancer Institute. The study was also supported by National Cancer Institute Grant K05 CA92633 awarded to Kevin D. McCaul. We wish to thank Fred Wolf for his help with the statistics.
Correspondence concerning this article should be addressed to Noel T. Brewer, UNC School of Public Health, 306 Rosenau Hall, CB#7440, Chapel Hill, NC 27599. E-mail: ntb1@unc.edu
Health Psychology Copyright 2007 by the American Psychological Association 2007, Vol. 26, No. 2, 136 –145 0278-6133/07/$12.00 DOI: 10.1037/0278-6133.26.2.136
136
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
Harrison, Mullen, and Green (1992) reported an effect size r of .15 (.10�.20, p � .01) for the relationship of perceived likelihood to health behavior and of .08 (.01�.19, p � .01) for the relationship of perceived illness severity to health behavior, where the range in parentheses indicate the range of effect sizes across five categories of health behaviors and several different research designs. Floyd, Prentice-Dunn, and Rogers (2000), reviewing studies related to protection motivation theory, reported an effect size r of .20 for 25 studies measuring perceived likelihood and .19 for 21 studies measuring perceived severity, both significant at p � .001. How- ever, nearly half of the studies in this meta-analysis used intentions as the outcome variable, not behavior. In another meta-analysis guided by protection motivation theory (Milne, Sheeran, & Orbell, 2000), the effect size r for 8 studies using cross-sectional/ retrospective designs was .13 (ns) for perceived likelihood and .10 (ns) for perceived severity. For 5 studies using prospective de- signs, these authors reported an effect size r of .12 ( p � .01) for perceived likelihood and stated that the effect for perceived sever- ity was not significant. McCaul, Branstetter, Schroeder, and Glas- gow (1996) conducted a meta-analysis of mammography screening and reported an effect size r of .16 ( p � .001) for the 19 studies examining the perceived likelihood�behavior relationship. Janz and Becker (1984) reported that 30 out of 37 studies based on the health belief model found significant effects for perceived likeli- hood. They also reported significant effects for perceived severity for 24 out of 30 studies. Looking at such data, some researchers (e.g., Leventhal, Kelly, & Leventhal, 1999) have argued that risk perceptions may have little impact on health behavior.
Risk Perception Dimensions
Before discussing the methodological issues that may have produced these relatively small effect sizes, we need to distinguish among three types of risk perceptions as shown in Table 1. Health hazards have many dimensions, but in describing the threat pre- sented by a hazard, nearly all theories focus on only two: the likelihood of harm if no action is taken and the severity of harm if no action is taken. The term likelihood is used interchangeably in this literature with probability, susceptibility and vulnerability. However, in this meta-analysis, we make a distinction between two logically distinct, though overlapping, concepts: likelihood of harm and susceptibility to illness. We define the first concept,
likelihood, as one’s probability of being harmed by a hazard under certain behavior conditions. It is represented by the question “What is the likelihood that you will get the flu this year if you don’t get a flu shot?” The term susceptibility is often used inter- changeably with likelihood, but we use it here to denote risk questions that appear to address different issues. These questions emphasize individual resistance or constitutional vulnerability, as in the questions “Do you get the flu easily?” and “Are you more likely to get the flu than other people?” Susceptibility to a disease should influence the likelihood of developing that disease, but being susceptible to an illness does not necessarily mean that the absolute probability of that illness is large. These two concepts are distinct from a third issue, “severity,” or “seriousness.” We define this third concept as the extent of harm a hazard would cause. It is represented by the question “How serious a disease is the flu?” Thus, our meta-analysis addresses three perceived risk dimensions: the likelihood of harm if no action is taken, susceptibility to harm if no action is taken, and the severity of harm if no action is taken.
An additional risk perception, the perceived risk if one does take some health-protective action, is also clearly relevant to health behavior. However, it reflects a combination of beliefs about both the likelihood of the risk if there is no action and the effectiveness of the precaution (Weinstein, 1993). Because beliefs about the risk given preventive action are seldom reported, we did not include this concept in our meta-analysis.
Conditioned Risk Questions
The empirical literature linking risk perceptions with behavior is compromised by methodological problems, some so severe that many studies need to be eliminated from meta-analyses (Brewer, Weinstein, Cuite, & Herrington, 2004; Weinstein & Nicolich, 1993; Weinstein, Rothman, & Nicolich, 1998). A major problem in testing whether risk perception motivates action is the failure to condition the risk question on not taking action. For example, if one is interested in testing the idea that a high perceived likelihood of getting influenza motivates influenza vaccination, one needs to know a person’s perception of what the probability would be if he or she does not get vaccinated. In a prospective study, when people are simply asked about their (unconditioned) probability of getting the flu, some may say that their risk is low because they never seem to get the flu. Yet, others may say that their risk is low
Table 1 Three Dimensions of Perceived Risk
Dimension Description Example
Perceived likelihood The probability that one will be harmed by the hazard
Imagine that the flu shot this year is unavailable and you are therefore unable to get the shot this fall. Given that you have had no shot, what would say is the likelihood that you would get the flu this winter? (Chapman & Coups, 2006)
If I don’t get immunized, there is a high chance of me getting flu [or pneumonia]. (Madhavan, Rosenbluth, Amonkar, Fernandes, & Borker, 2003)
Perceived susceptibility An individual’s constitutional vulnerability to a hazard
I get sick more easily than other people my age. (Nexoe, Kragstrup, & Sogaard, 1999)
Perceived severity The extent of harm a hazard would cause Influenza can cause death. (Nichol, Lofgren, & Gapinski, 1992) If I had influenza, I would not be able to manage daily activities.
(Zimmerman et al., 2003)
137RISK PERCEPTION AND VACCINATION
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
because they plan to get vaccinated, so their answer anticipates the effect of the vaccination on their risk. The risk perceptions de- scribed by the second group are the expected consequence of vaccination, not what they think their risk would be if they did not get vaccinated. Consequently, survey responses sometimes under- estimate what people who plan to act think their risk would be without action. As a result, the observed perceived risk�behavior association will underestimate the true association between per- ceived risk without behavior and the behavior itself. The latter is the quantity that should be examined.
A second serious methodological problem is the use of uncon- ditional risk questions in cross-sectional studies. Such studies compare the perceptions of people who have been vaccinated with the perceptions of people who have not. But the risk perceptions of the former group—in response to an unconditioned question such as, “What is your likelihood of getting the flu?”—will reflect their awareness of having received a vaccination (Brewer et al., 2004). If the vaccine is seen to be highly effective, the same theories that predict a positive association between risk likelihood and subse- quent action would predict a negative association in cross- sectional data (i.e., that people who have received the vaccine think their likelihood is lower than those who have not received the vaccine). This negative association would not mean that low risk likelihood motivates vaccination! Unconditioned risk questions in cross-sectional analyses will underestimate the relationship be- tween risk perceptions and behavior. (Because nearly all the stud- ies that were excluded from this meta-analysis for using uncondi- tioned risk questions also had other problems, we were unable to test this supposition.)
Whether or not risk questions need to be conditioned on not taking a precaution depends on whether it is expected to change one’s risk. Vaccination mainly changes likelihood. In contrast, mammography can change severity (by diagnosing breast cancer at an earlier stage), but it does not reduce the likelihood of disease. Consequently, perceived likelihood of the infectious disease (but not perceived severity) needs to be conditioned on not being vaccinated. Perceived severity of breast cancer (but not perceived likelihood) needs to be conditioned on not receiving a mammo- gram. Perceived susceptibility, as we are using the term, refers to beliefs about a general constitutional resistance that is independent of particular preventive actions, rather than a temporary state, so it does not need to be conditioned on no vaccination. (We located no studies in our search of the literature that conditioned perceived susceptibility on engaging in a health behavior.)
The meta-analyses carried out by Harrison et al. (1992) and Floyd et al. (2000) and the cross-sectional studies examined by Milne et al. (2000) included studies that should have used condi- tional questions to measure risk likelihood. The authors do not reveal which studies included in these reviews did or did not use conditional risk likelihood questions. Consequently, it is likely that they underestimated the risk likelihood�behavior relationship.
Other Risk Question Problems
Risk questions often have other weaknesses. One problem is that the risk question’s referent is ambiguous or that it refers to people in general (“How serious is the flu?”) rather than to the respondent (“How serious would it be if you got the flu?”). Social�cognitive theories of individual health behavior are constructed in terms of a
person’s beliefs about himself or herself, not a broader population category. Responses to a question referring specifically to the respondent are more likely to be associated with the respondent’s own behavior. Time frame is often missing from risk likelihood questions (i.e., “What is the chance that you will get the flu?” rather than “What is the chance that you will get the flu this year?”). This can result in added noise in responses if different respondents think about different time frames. Susceptibility ques- tions sometimes refer to illness in general (“I get sick more often than other people”) rather than to the specific illness under con- sideration (“I don’t seem to have much resistance to the flu”). More specific questions are more likely to be associated with the specific vaccination behavior under study.
In selecting studies for this meta-analysis, we excluded findings about risk likelihood if a cross-sectional study did not use a likelihood question conditioned on not being vaccinated because the finding is uninterpretable. (However, if the likelihood question referred to people in general— e.g., “A lot of people get the flu each year”—we did not require it to be conditioned because doing so would be unlikely to change response.) The other methodolog- ical issues mentioned earlier were incorporated into a study quality score rather than serving as exclusion criteria. Even with this restriction, risk perception measures in the literature take many forms, some of which more faithfully represent the concepts than others. Examples of the range of questions encountered in this review and included in this meta-analysis are shown in Table 1.
Risk Perceptions and Different Health Behaviors
The importance of risk perceptions to health behavior undoubt- edly varies across behaviors. Risk perceptions are probably more important for behaviors, such as sunscreen use, that are intended to reduce a specific health threat and are probably less important for behaviors, such as exercise and diet, that have a wide range of health and nonhealth consequences. Risk perceptions are probably more important when people make individual decisions about a behavior with relatively diffuse external influences, as in sun- screen use, than when strong external influences are present, as with physician recommendations for cancer screening tests. When it is easy to carry out the health behavior, there is likely to be a stronger association between perceptions and behavior than when it is difficult to carry out the behavior.
We selected adult vaccination against infectious disease as the focal health behavior for the current meta-analysis. It is a discrete behavior used to decrease the risk of a specific health threat. Strengthening the possible effect of risk perceptions on this be- havior, vaccination behavior is relatively easy to carry out. Weak- ening the risk perception�vaccination relationship, vaccination behavior sometimes reflects physician practice (the physician both recommends the vaccine and delivers it during a single patient visit) rather than the independent initiative of the recipient.
This meta-analysis was designed to test the hypotheses that higher perceived illness likelihood, perceived illness susceptibility, and perceived illness severity are associated with greater vaccina- tion behavior and to determine the strength of any associations that were found. In addition, we looked at several factors that might modify the strength of the associations, including both substantive issues (type of illness, population vaccinated) and methodological issues (study quality, ceiling or floor effects on variables). The
138 BREWER ET AL.
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
results of this meta-analysis will help us to understand better the role that risk perception should play in theories of health behavior.
Method
Study Selection
To identify articles to include in the meta-analysis, we used a two-stage screening process. In the first stage, we identified articles and papers that examined the risk perception and vaccination behavior of adults. We excluded childhood vaccination from the present analysis because it rep- resents a surrogate decision (parent judging risks and deciding about vaccination on behalf of the child). We searched PsycINFO and MEDLINE for articles published between the beginning of the databases and August 2004 whose title, abstract, or keywords included reference to both vacci- nation (i.e., vaccin* or immuniz* or innoculat* or shot) and perceived risk (perceived risk or risk perception* or perception of *risk or perceived likelihood or perceived susceptibility or perceived severity or attitude). We also searched the reference sections of the articles we obtained and circu- lated requests for unpublished studies among colleagues and on relevant electronic mailing lists.
In the second stage, we determined whether the studies satisfied our inclusion criteria. Studies included needed to assess one of three measures of perceived risk: perceived illness likelihood, perceived illness suscepti- bility, and perceived illness severity. Studies were excluded if they used only risk measures that conflated multiple risk constructs (e.g., likelihood and severity). Cross-sectional studies that examined perceived likelihood were required to condition the belief on not having been vaccinated (e.g., “How likely is it that you will get the flu given that you have not yet been vaccinated?”). Measures of susceptibility were acceptable only if they clearly assessed an individual’s resistance to disease (e.g., “I get sick more easily than other people my age.”). Simply mentioning the word suscep- tibility was not sufficient for a risk item to represent this construct. The criteria that applied to perceived severity were somewhat more relaxed and, in practice, many measures were included that might also be considered knowledge (e.g., “Influenza can cause death.”).
We required that the vaccination measure assess an actual behavior (not just a behavioral intention); vaccination measured by self-report was ac- ceptable. Articles in any language were accepted. Studies that only re- ported multivariate relationships were excluded because such statistics may understate the true relation of perceived risk to vaccination behavior. Studies were excluded from the meta-analysis if they did not report adequate statistical information to allow calculation of a bivariate effect size. When possible, study authors were contacted to obtain needed statis- tical information.
Of the 48 articles we excluded, the vast majority merely reviewed others’ data or did not report a quantitative assessment of the risk perception�vaccination relationship (n � 36). The others reported the relationship of risk perception to vaccination in a way that precluded calculating a standardized effect size (n � 2), multivariate but not bivariate risk perception�vaccination relationships (n � 2), unacceptable risk per- ception measures (n � 4), or risk measures whose appropriateness could not be assessed (n � 1).
Study Coding
Each risk measure was assigned a quality score. All risk measures received one point for each of the following good practices: The perceived risk question(s) concerned the individual’s own risk, the topic of the perceived risk question(s) matched the outcome, the perceived risk mea- sure was a composite of more than one perceived risk question (and, thus, was more likely to be a reliable measure), and categories of the response scale were not combined for analysis (a strategy that would reduce vari- ability in the predictor variable). Likelihood measures received an addi-
tional point for each of the following good practices: The underlying question(s) was conditioned on not having been vaccinated (for prospective studies), and the perceived risk question(s) specified the time frame for the illness. Based on a median split of the quality scores, risk measures were categorized as low or high quality.
For each study, we also coded sample size, whether the study was cross-sectional or prospective, the illness that the vaccine would protect against (i.e., influenza or other illness), skew of responses (i.e., whether more than 80% of respondents used a single perceived risk response category), vaccination rate (high or low, based on a median split), and population (i.e., whether respondents were healthy adults, medical person- nel such as doctors or nurses, or sick or high-risk persons). Coding of studies was performed by one of three judges using a standardized coding protocol and checked for accuracy by a second. Instances not covered by the protocol were discussed by the three judges to establish a standard policy that was then incorporated into the coding protocol. Calculation of effect sizes was performed by two judges independently. In the few cases in which judges’ assessments differed, they conferred with one another to resolve the discrepancy.
Data Analyses
For each study, up to three effect sizes (rs) were calculated, one for each of the perceived risk measures reported. The effect sizes were converted to Fisher’s zs (to allow us to combine the effect sizes properly). For each risk dimension, the zs were pooled across studies to create a single summary z, which was converted back into r (Rosenthal, 1994; Wolf, 1986). The single summary r for each of the three risk measures was tested for its difference from zero by t test. Variability among effect sizes might reflect important differences among subgroups. To investigate this possibility, the effect sizes were examined for heterogeneity by calculating the Q statistic. If Q was significant, we then examined six potential moderators of the risk perception�behavior relationship: design, quality score, illness, popula- tion, recruitment rate, and vaccination rate.
Results
Thirty-four studies fulfilled our inclusion criteria. The included studies and their characteristics are summarized in Table A1 of the Appendix. The total number of subjects in the studies was 15,988 with a median of 374 per study and a range of 72 to 1,530. The studies were conducted between 1979 and 2004 (mdn � 1997). There were 28 cross-sectional studies and 6 prospective studies. Twenty-five studies concerned influenza vaccination and the re- maining studies concerned vaccination against hepatitis, pneumo- nia, or Lyme disease. Recruitment rates ranged from 25% to 99% (mdn � 63%) and vaccination rates ranged from 6% to 86% (mdn � 51%). Populations studied included healthy adults (9 studies), medical personnel (7 studies), and sick or high-risk pop- ulations such as older adults (18 studies).
Perceived Likelihood
The relationship of perceived likelihood to vaccination was examined in 12 studies with 6,958 participants and effect sizes ranging from �.12 to �.45 (see Figure 1). The pooled effect was moderate in size (r � .26) and significantly different from zero, t(6957) � 22.29, p � .001. Those perceiving a higher likelihood of getting the illness were more likely to be vaccinated. The pooled effect showed heterogeneity of variance (Q � 155.29, p � .001), suggesting the presence of one or more moderators.
139RISK PERCEPTION AND VACCINATION
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
Additional analyses showed significance for the five moderators we examined. Prospective studies yielded a pooled effect size (r � .29) larger than that for cross-sectional studies (r � .24), t(6956) � 3.56, p � .001. Higher quality risk measures yielded a pooled effect size larger (r � .28) than did lower quality risk measures (r � .23), t(6956) � 3.14, p � .01. Studies of influenza vaccina- tion yielded a larger pooled effect size than studies of other illnesses (r � .27 vs. .22), t(6956) � 3.36, p � .001. Studies with extreme vaccination rates yielded a smaller pooled effect size than did those with vaccination rates closer to 50% (r � .22 vs. .28), t(6591) � 3.67, p � .001. The pooled effect size was small for medical personnel (r � .07), moderate in size for healthy adults (r � .23) and large for sick and high-risk populations (r � .42), and all differed from one another (ts � 7.72, ps � .001). The effect sizes for all subgroups created by the moderators were signifi- cantly different from zero ( ps � .001). Extremity of response distribution was not examined as a moderator because no likeli- hood studies had extreme response distributions.
Perceived Susceptibility
The relationship of perceived susceptibility to vaccination was examined in 5 studies with 2,543 participants and effect sizes that
ranged from .15 to .36 (see Figure 2). The pooled effect size was moderate in size (r � .24) and significantly different from zero, t(2542) � 12.53, p � .001. Those who perceived themselves to be more susceptible to an illness were more likely to be vaccinated against it. The pooled effect showed heterogeneity of variance (Q � 16.99, p � .01), suggesting the presence of moderators. Because of the small number of studies that assessed susceptibility, however, it was not possible to examine moderators.
Perceived Severity
The relationship of perceived severity to vaccination was exam- ined in 32 studies with 13,945 participants and effect sizes that ranged from �.18 to .39 (see Figure 3). The pooled effect size was small to moderate in size (r � .16) and significantly different from zero, t(13944) � 19.43, p � .001. Those who perceived the severity of illness to be higher were more likely to be vaccinated. The pooled effect showed heterogeneity of variance (Q � 297.52, p � .001), suggesting the presence of moderators.
The effects of all six moderators proved to be statistically significant. Prospective studies yielded a pooled effect size (r � .23) larger than was yielded by cross-sectional studies (r � .15), t(13943) � 7.60, p � .001. Higher quality risk measures yielded a
Study z (SE) 95% CI Weight %
Beguin et al., 1998 0.0700 (0.0256) 21.67 Brewer, et al., 2004 0.1640 (0.0387) 9.48 Chapman & Coups, 1999 -0.1210 (0.1147) 1.08 Chapman & Coups, 2006* 0.2710 (0.0483) 6.09 Chapman & Coups, 2006** 0.1590 (0.0530) 5.06 Cummings et al., 1979 0.2100 (0.1231) 0.94 Madhavan et al., 2003 0.2730 (0.0328) 13.20 Nexoe et al., 1999 0.3970 (0.0291) 16.77 Rundall & Wheeler, 1979 0.4200 (0.0718) 2.75 Weinstein, 2004 0.3100 (0.0488) 5.96 Hamilton-West, 2006 0.2450 (0.0648) 3.38 Zimmerman et al., 2003 0.4489 (0.0323) 13.61
100.00
-0.5 -0.25 0 0.25 0.5
Less vaccination More vaccination
Figure 1. Perceived likelihood. Because our meta-analysis transformed the asymmetrical effect size r to the symmetrical effect size z, the figure shows the latter to more accurately depict confidence intervals (CIs). A positive z indicates that vaccination was more prevalent among participants reporting higher perceived likeli- hood. *Healthy adult sample. **Medical personnel sample.
)ES( z ydutS 95% CI Weight %
Armstrong et al., 2001 0.1470 (0.0455) 19.13 Nexoe et al., 1999 0.2300 (0.0292) 46.44 Roy et al., 1996 0.1540 (0.0825) 5.82 van Essen et al., 1997a 0.3780 (0.0781) 6.49 van Essen et al., 1997b 0.3650 (0.0423) 22.13
100.00
-0.5 -0.25 0 0.25 0.5
More vaccination Less vaccination
Figure 2. Perceived susceptibility. Because our meta-analysis transformed the asymmetrical effect size r to the symmetrical effect size z, the figure shows the latter to more accurately depict confidence intervals (CIs). A positive z indicates that vaccination was more prevalent among participants reporting higher perceived susceptibility.
140 BREWER ET AL.
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
pooled effect size larger (r � .20) than for lower quality risk measures (r � .11), t(13943) � 7.39, p � .001. Studies of influenza vaccination yielded a larger pooled effect size than studies of other illnesses (r � .18 vs. .07), t(13943) � 9.83, p � .001. Studies with extreme vaccination rates yielded a smaller pooled effect size than those with vaccination rates closer to 50% (r � .12 vs. .18), t(13578) � 4.28, p � .001. Studies with skewed risk response distributions yielded a smaller pooled effect size than those with more moderate distributions (r � .12 vs. .19), t(13213) � 5.86, p � .001. Pooled effect sizes were smaller for medical personnel (r � .14) and healthy adults (r � .11) than for sick and high-risk populations (r � .20, ts � 4.18, ps � .001). Effect sizes for all subgroups were significantly different from zero ( p � .001).
Relationships Among Moderator Variables
In the studies that assessed likelihood, no moderator variables were correlated with one another. Among the studies that assessed severity, however, several moderator variables were related to one another. First, quality of severity measures was related to study design: All 6 of the prospective studies that assessed severity used high-quality risk measures, compared with only 12 of the 26
cross-sectional studies (� � .42, p � .02). Second, population was also related to study design, �2(2, N � 32) � 7.07, p � .02.1 Of the 8 studies of healthy adults, 4 used prospective design, com- pared with only 1 of the 7 studies of clinicians and 1 of the 17 studies of high-risk adults. Finally, population was related to extremity of risk responses, �2(2, N � 32) � 7.22, p � .03. Of the 7 studies of clinicians, 5 showed extreme risk responses, compared with only 1 of 8 studies of healthy adults and 3 of 14 studies of high-risk adults. (Three additional studies of high-risk adults could not be coded for extremity of risk response because they did not report the relevant data.) Because several moderators were them- selves related, one cannot interpret their effects on the perceived severity�vaccination relationship as if the influence of other in- fluential moderators has been controlled for.
Discussion
Risk perceptions are central to many of the theories used to explain health behaviors but are less important to or ignored
1 The chi-square analyses should be viewed as descriptive, as the ex- pected values in several cells are less than 5.
)ES( z ydutS 95% CI Weight %
Aho, 1979 -0.0580 (0.0891) 0.91 Armstrong et al., 2001 0.1410 (0.0455) 3.49 Beguin et al., 1998 0.2410 (0.0256) 11.01 Brewer et al., 2004 0.0460 (0.0400) 4.51 Chapman & Coups, 2006 -0.1820 (0.1147) 0.55 Chapman & Coups, 2006* 0.3310 (0.0530) 2.57 Chapman & Coups, 2006** 0.2060 (0.1231) 0.48 Cummings et al., 1979 0.2080 (0.0594) 2.05 Ehresman et al., 2001 0.1000 (0.0535) 2.52 Evans & Watson, 2003 0.1330 (0.0261) 10.59 Gene et al., 1992 0.1050 (0.0778) 1.19 Hamilton-West, 2006 0.0160 (0.0639) 1.77 Hashimoto et al., 1988 0.0070 (0.0432) 3.87 Heimberger et al., 1995 0.0000 (0.0333) 6.51 Honkanen et al., 1996 0.1690 (0.0513) 2.74 Jacobsen et al. 1989 0.1480 (0.0488) 3.03 Lewis-Parmar & McCann, 2002*** 0.0000 (0.1005) 0.71 Lewis-Parmar & McCann, 2002**** 0.1930 (0.0606) 1.97 McCusker, et al., 1990 0.2340 (0.1204) 0.50 Nexoe et al., 1999 0.4150 (0.0295) 8.29 Nichol et al., 1996 0.2070 (0.0526) 2.61 Nichol et al., 1992 -0.0390 (0.0458) 3.44 Nichol & Hague, 1997 0.2540 (0.0507) 2.81 Opstelten et al., 2001 0.3950 (0.0398) 4.56 Pearson & Thompson, 1994 -0.1090 (0.0668) 1.62 Pregliasco, 1999 -0.0180 (0.0825) 1.06 Roy, 1996 0.0390 (0.0857) 0.98 Rundall & Wheeler, 1979 0.0770 (0.0747) 1.29 Stephenson et al., 2002 0.1500 (0.0410) 4.29 van Essen et al., 1997a 0.3300 (0.0803) 1.12 van Essen et al., 1997b 0.3900 (0.0428) 3.94 Weinstein, 2004 -0.0200 (0.0488) 3.03
100.00
-0.5 -0.25 0 0.25 0.5
Less vaccination More vaccination
Figure 3. Perceived severity. Because our meta-analysis transformed the asymmetrical effect size r to the symmetrical effect size z, the figure shows the latter to more accurately depict confidence intervals (CIs). A positive z indicates that vaccination was more prevalent among participants reporting higher perceived severity. *Healthy adult sample. **Medical personnel sample. ***High risk adult sample (diabetics). ****High risk adult sample (�75 years old).
141RISK PERCEPTION AND VACCINATION
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
altogether by others. The empirical literature, muddied by frequent inappropriate assessment and analysis, looks inconsistent. How- ever, we found a high degree of consistency and a strength of association between risk perceptions and behavior that is larger than had been suggested by prior meta-analyses. The present meta-analysis revealed that all three risk perception measures were related to vaccination behavior. The magnitudes of the associations with behavior were similar for perceived risk likelihood (r � .26) and susceptibility (r � .24) but somewhat smaller for severity (r � .16). The smaller effect size for perceived risk severity may reflect the larger variation of types of questions used to assess this construct, although lower predictive validity for perceived severity has been reported previously in the literature (e.g., Harrison et al., 1992).
A number of factors moderated the relationship of perceived risk likelihood and severity to vaccination behavior. Of note, the few prospective studies located yielded larger effect sizes than did cross-sectional studies. This moderation effect was especially large for risk severity measures, with the prospective studies yielding a pooled effect half again as large as that from the cross-sectional studies. To some extent, this may have been due to the fact that prospective studies tended to use higher quality severity measures. The moderating effect of prospective versus cross-sectional design is still present, however, for risk likelihood relationships, where design type is not confounded with risk measure quality.
Prospective studies are the preferred design because they assess risk perception before the respondent engages (or does not engage) in the health behavior. Consequently, prospective designs increase the plausibility that the risk perception motivates the behavior, rather than the reported risk perception being constructed to justify a behavior that has already taken place. The post hoc justification that can occur in cross-sectional studies (e.g., “I must be the sort of person who is at high risk because I just got vaccinated”) might artificially inflate the size of the risk perception�health behavior relationship, but the fact that effect sizes were no greater in cross-sectional than prospective studies (and indeed were smaller) suggests that this is not the case. It should be recognized, however, that if a behavior is repeated, as in annual influenza vaccination, perceptions may change over time to become consistent with past action, so the direction of causation can be unclear even in a prospective design. For this reason, first-time vaccination against an illness (as represented by the studies of hepatitis, pneumonia, and Lyme disease vaccination) may be the best indicator of the strength of causation (Weinstein, 2004), and these effects were smaller than those with influenza.
Another moderator revealed by the present analysis is quality of the risk measure. Likelihood and severity measures that scored lower on our quality scale showed weaker associations with vac- cination behavior. Poor quality measures may fail to capture the intended construct. If the measure assesses something other than the intended likelihood or severity or is statistically unreliable, it is not surprising for it to be unrelated to vaccination.
Studies of influenza vaccination yielded larger effect sizes than did studies of other types of vaccination. This may be because the non-flu category consisted of a mix of vaccine types (hepatitis, pneumonia, and Lyme disease) that resulted in additional noise across studies in risk and behavior measures, resulting in smaller effect sizes. It may also be due to the relative familiarity, acces- sibility, low cost, and habitual uptake of flu shots relative to these
other types of shots. Another possibility is that the hepatitis and Lyme disease vaccines require a series of three vaccinations and this additional hurdle may attenuate the risk perception�vaccination relationship.
Studies of medical personnel yielded a smaller effect size than studies of sick and/or high-risk adults. The effect size for healthy adults was in between these two groups for risk likelihood and equivalent to medical personnel for risk severity. Why would effect sizes be smaller for medical personnel? This result may indicate that perceived risk to self is less of a motivator for health care providers than it is for patients. Providers may be primarily motivated by concerns specific to their job role, such as a desire not to spread infection to patients. Alternatively, the smaller effect sizes may be related to some clinicians having been required by their employers to be vaccinated (e.g., against the flu and hepatitis B) or to receiving these services at their worksite. These two factors could undermine the influence of perceived risk by elimi- nating the volitional aspect of the behavior for some or by making vaccination so simple for others that the remaining unvaccinated persons reflect an unusual subset.
The distributions of both the risk measure and the behavior measure were both associated with effect size, an indication that perceived risk may have greater importance than suggested by the effect sizes we report. When an extreme percentage of respondents used a single risk response category or when an extreme percent- age of respondents was in a single behavior category (vaccinated or not), effect sizes were smaller. If an outcome is dichotomous and rare, an effect can appear to be quite small when reported as a correlation coefficient (or effect size r) even though it is a clinically important effect that would be large when reported as an odds ratio (Rutledge & Loh, 2004). The implication is that the meta-analysis may slightly understate the meaningfulness of the relation of perceived risk to vaccination.
Only five studies of susceptibility were included in the present analysis. As a consequence, although we could detect that suscep- tibility is reliably associated with vaccination behavior, we could not explore moderators of this relationship. Susceptibility appears to be an under-studied aspect of risk perception. Research is needed to determine whether this construct improves the predic- tion of health behaviors beyond that provided by perceptions of risk likelihood. Several other perceived risk measures, such as perceived disease prevalence and believing oneself to be a member of a high risk group, have also received relatively little attention in the research literature (Brewer & Hallman, 2006).
In summary, the occasional study showing no relationship be- tween perceived risk and vaccination, or a small negative relation- ship, can be viewed as part of a larger distribution of effect sizes around a positive and significant mean effect size. We find strong evidence that perceived likelihood, susceptibility, and severity are reliably associated with vaccination and that the relationships are at least small to moderate in size. When one considers the meth- odological weaknesses that suppress the size of this relationship, the effect may be more accurately characterized as moderate, with major applied implications. Many of these findings rely on data from cross-sectional studies, but larger effects are found in longi- tudinal studies giving us confidence in our conclusion. However, experiments that manipulate risk perception are needed to provide a more definitive confirmation of the causal relationship between risk perception and preventive action. Additional research is also
142 BREWER ET AL.
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
needed to determine whether the relationship between risk percep- tion and health behavior varies cross-culturally.
To illustrate the applied importance of an effect size that cor- responds to a small portion of the total variance in behavior, we converted a correlation of .28 (corresponding to the value found for high-quality studies of risk likelihood) to a contingency table. We assumed a vaccination rate of 50% and an equal distribution between high and low perceived risk, both parameters close to the median found in our meta-analysis. In this situation, the correlation of .28 corresponds to a vaccination rate of 36% in the low per- ceived risk group and 64% in the high perceived risk group. Assuming that the observed correlations represent the causal im- pact of risk perceptions, it appears that raising risk perceptions from low to high would have a major effect on vaccination behavior. The present meta-analysis demonstrates that hazard- specific risk perceptions are predictors of vaccination behavior. This finding supports the inclusion of risk perceptions in health behavior-specific theories such as protection motivation theory (Rogers, 1975) and the extended parallel process model (Witte, 1992). It also suggests that the more general behavioral theories, such as the theory of planned behavior (Ajzen, 1985), may im- prove their ability to predict health behavior if they too incorporate these constructs.
References
References marked with an asterisk (*) indicate studies included in the meta-analysis.
*Aho, W. R. (1979). Participation of senior citizens in the swine flu inoculation program: An analysis of Health Belief Model variables in preventive health behavior. Journal of Gerontology, 34, 201–208.
Ajzen, I. (1985). From intentions to actions: A theory of planned behavior. In J. Kuhl & J. Beckman (Eds.), Action control: From cognitions to behaviors (pp. 61– 85). New York: Springer.
*Armstrong, K., Berlin, M., Schwartz, J. S., Propert, K., & Ubel, P. A. (2001). Barriers to influenza immunization in a low-income urban pop- ulation. American Journal of Preventive Medicine, 20, 21–25.
*Beguin, C., Boland, B., & Ninane, J. (1998). Health care workers: Vectors of influenza virus? Low vaccination rate among hospital health care workers. American Journal of Medical Quality, 13, 223–227.
Brewer, N. T., & Hallman, W. K. (2006). Subjective and objective risk as predictors of influenza vaccination during the vaccine shortage of 2004 – 2005. Clinical Infectious Diseases, 43, 1379 –1386.
*Brewer, N. T., Weinstein, N. D., Cuite, C. L., & Herrington, J. (2004). Risk perceptions and their relation to risk behavior. Annals of Behavioral Medicine, 27, 125–130.
Cameron, L. D. (2003). Conceptualizing and assessing risk perceptions: A self-regulatory perspective. Presented at the National Cancer Institute’s “Workshop on Conceptualizing (and Measuring) Perceived Risk.” Ac- cessed April 3, 2005, from http://dccps.nci.nih.gov/brp/presentations/ cameron.pdf
*Chapman, G. B., & Coups, E. J. (1999). Predictors of influenza vaccine acceptance among healthy adults. Preventive Medicine, 29, 249 –262.
*Chapman, G. B., & Coups, E. J. (2006). Worry, risk, and influenza vaccination. Health Psychology, 25, 82–90.
*Cummings, K. M., Jette, A. M., Brock, B. M., & Haefner, D. P. (1979). Psychosocial determinants of immunization behavior in a swine influ- enza campaign. Medical Care, 17, 639.
*Ehresmann, K. R., Ramesh, A., Como-Sabetti, K., Peterson, D. C., Whit- ney, C. G., & Moore, K. A. (2001). Factors associated with self-reported pneumococcal immunization among adults 65 years of age or older in
the Minneapolis-St. Paul metropolitan area. Preventive Medicine, 35, 409 – 415.
*Evans, M. R., & Watson, P. A. (2003). Why do older people not get immunised against influenza? A community survey. Vaccine, 21, 2421– 2427.
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, 407– 429.
*Gene, J., Espinola, A., Cabezas, C., Boix, C., Comin, E., Martin, A., & Sanz, E. (1992). Do knowledge and attitudes about influenza and its immunization affect the likelihood of obtaining immunization? Family Practice Research Journal, 12, 61–73.
*Hamilton-West, K. E. (2006). [Predictors of influenza vaccination]. Un- published raw data.
Harrison, J. A., Mullen, P. D., & Green, L. W. (1992). A meta-analysis of studies of the health belief model with adults. Health Education Re- search, 7, 107–116.
*Hashimoto, F., Hunt, W. C., & Brusuelas, P. (1988). Physician acceptance of the hepatitis B vaccine at a university medical center. American Journal of Public Health, 78, 973–974.
*Heimberger, T., Chang, H., Shaikh, M., Crotty, L., Morse, D., & Birk- head, G. (1995). Knowledge and attitudes of healthcare workers about influenza: Why are they not getting vaccinated? Infection Control and Hospital Epidemiology, 16, 412– 414.
*Honkanen, P. O., Keistinen, T., & Kivela, S. (1996). Factors associated with influenza vaccination coverage among the elderly: Role of health care personnel. Public Health, 110, 163–168.
*Jacobson, J. J., Lang, W. P., Ybanez, M. S., Shipman, C., Jr., Johnston, F. K., & LaTurno, D. E. (1989). Acceptance of hepatitis B vaccine among dental health care workers. Journal of Public Health Dentistry, 49, 67–72.
Janz, N. K., & Becker, M. H. (1984). The health belief model: A decade later. Health Education Quarterly, 11, 1– 47.
Leventhal, H., Kelly, K., & Leventhal, E. A. (1999). Population risk, actual risk, perceived risk, and cancer control: A discussion. Journal of the National Cancer Institute Monographs, 25, 81– 85.
Leventhal, H., Meyer, D., & Nerenz, D. (1980). The common sense representation of illness danger. In S. Rachman (Ed.), Medical psychol- ogy (Vol. 2, pp. 7–30). New York: Pergamon.
*Lewis-Parmar, H., & McCann, R. (2002). Achieving national influenza vaccine targets—An investigation of the factors affecting influenza vaccine uptake in older people and people with diabetes. Communicable Disease & Public Health, 5, 19 –26.
*Madhavan, S. S., Rosenbluth, S. A., Amonkar, M., Fernandes, A., & Borker, R. (2003). Immunization predictors in rural adults under 65 years of age. Journal of Health Care for the Poor & Underserved, 14, 100 –121.
McCaul, K., Branstetter, A. D., Schroeder, D. M., & Glasgow, R. E. (1996). What is the relationship between breast cancer risk and mam- mography screening? A meta-analytic review. Health Psychology, 15, 423– 429.
*McCusker, J., Hill, E. M., & Mayer, K. H. (1990). Awareness and use of hepatitis B vaccine among homosexual male clients of a Boston com- munity center. Public Health Reports, 105, 59 – 64.
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, 106 –143.
*Nexoe, J., Kragstrup, J., & Sogaard, J. (1999). Decision on influenza vaccination among the elderly. A questionnaire study based on the health belief model and the multidimensional locus of control theory. Scandi- navian Journal of Primary Health Care, 17, 105–110.
143RISK PERCEPTION AND VACCINATION
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
*Nichol, K. L., & Hague, M. (1997). Influenza vaccination of healthcare workers. Infection Control & Hospital Epidemiology, 18, 189 –194.
*Nichol, K. L., Lofgren, R. P., & Gapinski, J. (1992). Influenza vaccina- tion. Knowledge, attitudes, and behavior among high-risk outpatients. Archives of Internal Medicine, 152, 106 –110.
*Nichol, K. L., MacDonald, R., & Hauge, M. (1996). Factors associated with influenza and pneumococcal vaccination behavior among high-risk adults. Journal of General Internal Medicine, 11, 673– 677.
*Opstelten, W., Hak, E., Verheij, T. J., & van Essen, G. A. (2001). Introducing a pneumococcal vaccine to an existing influenza immuni- zation program: Vaccination rates and predictors of noncompliance. American Journal of Medicine, 111, 474 – 479.
*Pearson, D. C., & Thompson. R. S. (1994). Evaluation of group health cooperative of Puget Sound’s senior influenza immunization program. Public Health Reports, 109, 571–578.
*Pregliasco, F., Sodano, L., Mensi, C., Selvaggi, M. T., Adamo, B., D’Argenio, P., et al. (1999). Influenza vaccination among the elderly in Italy. Bulletin of the World Health Organization, 77, 127–131.
Rogers, R. W. (1975). A protection motivation theory of fear appeals and attitude change. Journal of Psychology, 91, 93–114.
Ronis, D. L. (1992). Conditional health threats: Health beliefs, decisions, and behaviors among adults. Health Psychology, 11, 127–134.
Rosenstock, I. (1974). The health belief model and preventive behavior. Health Education Monographs, 2, 354 –386.
Rosenthal, O. (1994). Parametric measures of effect size. In H. Cooper & L. V. Hedges (Eds.), The handbook of research synthesis (pp. 231–244). New York: Russell Sage Foundation.
*Roy, G., Fradet, M., & Le Henaff, D. (1996). Vaccination against influ- enza in elderly persons: Some current presumptions. Canadian Journal of Public Health/Revue Canadienne de Sante Publique, 87, 298 –300.
*Rundall, T. G., & Wheeler, J. R. (1979). Factors associated with utiliza- tion of the swine flu vaccination program among senior citizens in Tompkins County. Medical Care, 17, 191–200.
Rutledge, T., & Loh, C. (2004). Effect sizes and statistical testing in the determination of clinical significance in behavioral medicine research. Annals of Behavioral Medicine, 272, 134 –145.
*Stephenson, I., Roper, J. P., & Nicholson, K. G. (2002). Healthcare
workers and their attitudes to influenza vaccination. Communicable Disease & Public Health, 5, 247–252.
Sutton, S. R. (1987). Social-psychological approaches to understanding addictive behavior: Attitude-behavior and decision-making models. British Journal of Addiction, 82, 355–370.
*van Essen, G. A., Kuyvenhoven, M. M., & de Melker, R. A. (1997a). Compliance with influenza vaccination: Its relation with epidemiologic and sociopsychological factors. Archives of Family Medicine, 6, 157– 162.
*van Essen, G. A., Kuyvenhoven, M. M., & de Melker, R. A. (1997b). Why do healthy elderly people fail to comply with influenza vaccination? Age & Ageing, 26, 275–279.
Weinstein, N. D. (1993). Testing four competing theories of health behav- ior. Health Psychology, 12, 324 –333.
Weinstein, N. D. (2004). Misleading tests of health behavior theories. Unpublished manuscript, Department of Human Ecology, Rutgers Uni- versity, New Brunswick, NJ.
*Weinstein, N. D., McCaul, K., Gerrard, M., & Gibbons, F. X. (2004). Risk perceptions: Assessment and relationship to influenza vaccination. Un- published manuscript, Department of Human Ecology, Rutgers Univer- sity, New Brunswick, NJ.
Weinstein, N. D., & Nicolich, M. (1993). Correct and incorrect interpre- tations of correlations between risk perceptions and risk behaviors. Health Psychology, 12, 235–245.
Weinstein, N. D., Rothman, A. J., & Nicolich, M. (1998). Use of correla- tional data to study effects of risk perceptions on precautionary behavior. Psychology and Health, 13, 479 –501.
Witte, K. (1992). Putting the fear back into fear appeals: The extended parallel process model. Communication Monographs, 59, 329 –349.
Wolf, F. M. (1986). Meta-analysis: Quantitative Methods for research synthesis. Beverly Hills, CA: Sage.
*Zimmerman, R. K., Santibanez, T. A., Janosky, J. E., Fine, M. J., Raymund, M., Wilson, S. A., et al. (2003). What affects influenza vaccination rates among older patients? An analysis from inner-city, suburban, rural, and Veterans Affairs practices. American Journal of Medicine, 114, 31–38.
Appendix
Notes on the Calculation of Effect Sizes
The general analytic approach followed that suggested by Rosenthal (1994) and Wolf (1986). An effect size (r) was calcu- lated for each bivariate risk perception-vaccination behavior rela- tionship reported in a study. We report only on linear effects because no studies reported tests of curvilinear effects, and hardly any examined interactions. A sizeable number of studies reported contingency tables from which we were able to directly calculate r. The most common situation was a 2 � 2 table (low or high perceived risk, by vaccinated or not) from which we calculated a phi coefficient. In other cases, we transformed the statistic reported
(e.g., t) to r. If multiple relationships involving the same risk construct (e.g., two likelihood measures) were reported in the same study, these were combined by averaging the rs after a Fisher z transformation. If multiple relationships for different risk con- structs (e.g., likelihood and severity) were reported in the same study, these were included separately in the meta-analysis. In cases in which a study reported results separately for substantially dif- ferent populations (such as healthy adults and infirm adults), effect sizes were calculated separately for each population and were treated as being from separate studies.
144 BREWER ET AL.
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .
Table A1 Studies Included in Meta-Analysis
Study Design Illness Vaccination
Rate (%) Population r N Quality score
�80% in same response category
Perceived likelihood
Beguin et al., 1998 c f 32 m .07 1,530 2 n Brewer et al., 2004 p o 6 h .16 702 7 n Chapman and Coups, 1999 c f h �.12 79 5 n Chapman and Coups, 2006a p f 47 h .27 428 6 n Chapman and Coups, 2006b p f 67 m .16 428 6 n Cummings et al., 1979 p f h .21 286 3 n Hamilton-West, 2006 c o 34 h .25 241 7 n Madhavan et al., 2003 c o 20 h .27 931 5 n Rundall and Wheeler, 1979 c o 73 s .36 197 3 n Nexoe et al., 1999 p f 37 s .40 1,182 4 n Weinstein, 2004 p f 28 h .30 423 7 n Zimmerman et al., 2003 c f 79 s .45 959 3 n
Perceived susceptibility
Armstrong et al., 2001 c f 63 s .15 486 3 n Roy et al., 1996 c f 48 s .15 150 2 y van Essen et al., 1997a c f 66 s .36 167 3 n van Essen et al., 1997b c f 86 s .35 561 3 n Nexoe et al., 1999 p f 37 s .23 1,179 2 n
Perceived severity
Aho, 1979 c f 54 s �.06 129 3 n Armstrong et al., 2001 c f 63 s .14 486 2 y Beguin et al., 1998 c f 32 m .24 1,530 3 y Brewer et al., 2004 p o 6 h .05 627 2 n Chapman and Coups, 1999 c f h �.18 79 3 n Chapman and Coups, 2006a p f 47 h .32 428 3 n Chapman and Coups, 2006b p f 67 m .20 428 3 n Cummings et al., 1979 p f h .21 286 2 n Ehresmann et al., 2001 c o 59 s .10 353 1 Evans and Watson, 2003 c f 51 s .13 1,468 1 n Gene et al., 1992 c f 51 s .02 168 1 n Hamilton-West, 2006 c o 34 h .02 248 4 n Hashimoto et al., 1988 c o 57 m .01 538 1 y Heimberger et al., 1995 c f 16 m .00 904 1 y Honkanen et al., 1996 c f 51 h .17 383 1 y Jacobson et al., 1989 c o 61 m .15 423 1 y Lewis-Parmar and McCann, 2002c c f 67 s .00 102 1 Lewis-Parmar and McCann, 2002d c f 70 s .19 275 1 McCusker et al., 1990 c o 24 h .23 72 2 n Nexoe et al., 1999 p f 37 s .39 1,151 3 n Nichol et al., 1992 c f 47 s �.04 480 2 y Nichol et al., 1996 c o 74 s .20 364 2 n Nichol and Hague, 1997 c f 61 m .25 392 2 y Opstelten et al., 2001 c f 75 s .38 634 2 n Pearson and Thompson, 1994 c f 71 s �.11 227 0 n Pregliasco et al., 1999 c f 26 s �.02 150 1 n Roy et al., 1996 c f 47 s .04 139 1 y Rundall and Wheeler, 1979 c o 72 s .08 182 2 n Stephenson et al., 2002 c f 14 m .15 597 2 n van Essen et al., 1997a c f 66 s .32 158 1 n van Essen et al., 1997b c f 86 s .37 549 1 n Weinstein, 2004 p f h �.02 423 4 n
Note. c � cross-sectional design, p � prospective design. f � vaccination against influenza, o � vaccination against other illness. h � healthy adults, m � medical personnel, s � sick or high risk persons. n � no; y � yes. a Healthy adult sample. b Medical personnel sample. c High risk adult sample (diabetics). d High risk adult sample (�75 years old).
145RISK PERCEPTION AND VACCINATION
T hi
s do
cu m
en t i
s co
py ri
gh te
d by
th e
A m
er ic
an P
sy ch
ol og
ic al
A ss
oc ia
tio n
or o
ne o
f i ts
a lli
ed p
ub lis
he rs
. T
hi s
ar tic
le is
in te
nd ed
s ol
el y
fo r t
he p
er so
na l u
se o
f t he
in di
vi du
al u
se r a
nd is
n ot
to b
e di
ss em
in at
ed b
ro ad
ly .