33.41 .
‘3?
’v

. n.3,
M an is"
; 'W A
I I A
P .-

«X.

1‘

x k r- 5,
Lanafwgg;

A,”

I 95,.
Law *

. . a;
' ‘ , ,‘-

"x" , '. a; . .53;
mm? - r ”we.
39.23335] [24- 11‘“
W“? i % ,.
4M2“ .. ,7.“
$.11?“

fig,

 

 

A} .g.‘ ‘ .

- Ekfifa-WW‘

- "-3, '-5-‘,;':.-t"$,.‘

'1' ‘ ' 3

3.. {ft-W?" ‘V. ”‘
n.

 

Y
‘ 9.,
xii; Y1

ask.

 

"if. . ‘1; ',,
Q M ‘
"‘ M23373
" 9,9 3, 51);:
64 «gr.

i.

 

E
F”! 4‘

161.5"
‘5‘. .
l 6":

a illlifllliiiwWWWM ,

 

"Si“ LIBRARY
Michigan state
University

 

 

 

This is to certify that the

dissertation entitled

Confounds in Fear Arousing Persuasive Messages:
Do the Paths Less Traveled Make All the Difference?

presented by

Jose Ignacio Rodriguez

has been accepted towards fulfillment
of the requirements for

 

 

Doctorate (Ph.D.aeg1-ee in Communication

Major professor

Date 3-17—95

 

MS U is an Affirmative Action/Equal Opportunity Institution 0-12771

PLACE ll RETURN BOXtonmovothbchockoufmn yourncord.
To AVOID FINES return on or baton dd. duo.

DATE DUE DATE DUE DATE DUE

 

' 7!"!er-

 

 

 

 

NEW-303399 ‘ .

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

MSU It An Mandi” ActiorVEqud Opportunity intuition
Wan-9.1

CONFOUNDS IN FEAR AROUSING PERSUASIVE MESSAGES: DO THE PATHS
LESS TRAVELED MAKE ALL THE DIFFERENCE?

By

José Ignacio Rodriguez

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Communication

1995

ABSTRACT

CONFOUNDS IN FEAR AROUSING PERSUASIVE IVflESSAGES: DO THE PATHS
LESS TRAVELED MAKE ALL THE DIFFERENCE?

By

Jose' Ignacio Rodriguez

The purpose of this investigation is to explore potential confounds in fear arousing
persuasive messages. It is argued that these confounds moderate or mediate the
relationship between fear arousing messages and attitude change. A 3 X 3 X 2 Solomon
four-group design was employed. There were three message topics (bicycle helmets,
tetanus, and alcohol consumption), three fear conditions (high, low, and control), and the
presence or absence of a pretest. Subjects were assigned to conditions randomly with the
constraint that comparable numbers of subjects were in each cell. A series of causal
models were devised and tested. Path analyses indicated that perceived fear and perceived
argument quality mediated the relationship between fear arousing content and attitude
change. The exact structure of the path models varied as a function of message topic.

The implications of the findings and directions for future research are discussed.

ACKNOWLEDGMENTS

I am gratefirl to Dr. Frank Boster, Dr. Stan Kaplowitz, Dr. Sandi Smith, and Dr.

Robin Graham for their substantial contributions to this manuscript. I extend my deepest
appreciation to Dr. Boster for his willingness to share his ideas, insights, and exceptional
talents with me for the past four years. Frank and Linda Boster have also made my spouse
and I feel extremely welcome in their home, and we are thankful for their hospitality and
fiiendship. I also thank Dr. Kim Witte for sharing her impressive collection of fear
arousing persuasive messages with me. A special thanks to my spouse, Mercy Leon and
my good friend, Eric Brown for helping me collect and enter an enormous amount of data.
Finally, I thank the faculty and staff of the Department of Communication for their

encouragement and support.

iii

TABLE OF CONTENTS

LIST or TABLES

LIST OF FIGURES _

REVIEW OF THE LITERATURE
METHOD

RESULTS

DISCUSSION

APPENDIX A: CORRELATIONS AMONG THE VARIABLES FOR
THE ALCOHOL CONSUMPTION MESSAGE

APPENDIX B: CORRELATIONS AMONG THE VARIABLES FOR
THE BICYCLE HELMET MESSAGE

APPENDIX C: CORRELATIONS AMONG THE VARIABLES FOR
THE TETANU S MESSAGE

LIST OF REFERENCES

iv

vii

21

3O

59

65

67

69

71

Table

Table

Table

Table

Table

Table

Table

Table

Table

LIST OF TABLES

: Means and Standard Deviations for the Test of the

EPPM Using the Alcohol Consumption Message

: Means and Standard Deviations for the Test of the

ELM Using the Alcohol Consumption Message

: Descriptive Statistics for the Dependent Variables

of the Alcohol Consumption Message

: Correlations and Descriptive Statistics for the Perceived Fear

and Perceived Argument Quality Independence Model
for the Alcohol Consumption Message

: Means and Standard Deviations for the Test of the

EPPM Using the Bicycle Helmet Message

: Means and Standard Deviations for the Test of the

ELM Using the Bicycle Helmet Message

: Descriptive Statistics for the Dependent Variables

of the Bicycle Helmet Message

: Correlations and Descriptive Statistics for the

Perceived Argument Quality Model
for the Bicycle Helmet Message

: Means and Standard Deviations for the Test of the

EPPM Using the Tetanus Message

Table 10: Means and Standard Deviations for the Test of the

ELM Using the Tetanus Message

Table 11: Descriptive Statistics for the Dependent Variables

of the Tetanus Message

41

42

45

46

48

49

5]

52

53

54

56

Table 12: Correlations and Descriptive Statistics for the
Perceived Fear and Perceived Argument Quality
Independence Model for the Tetanus Message

58

Figure

Figure

Figure

Figure
Figure
Figure
Figure
Figure

Figure

8:

9:

Figure 10

Figure 11

Figure 12

LIST OF FIGURES

: A Diagram of a Causal Model Depicting the Set

of Relationships Predicted by PMT.

: A Diagram of a Causal Model Depicting the Set

of Relationships Predicted by EPPM when
Perceived Efiicacy and Threat are High.

: A Diagram of a Causal Model Depicting the Set

of Relationships Predicted by EPPM when
Perceived Eflicacy is Low and Threat is High.

: The Discrepancy Model.
: The Discrepancy and Credibility Model.
: The Outcome—Relevant Involvement Model.

: The Emotion-Cognition Independence Model.

The Emotion Antecedent to Cognition Model.
The Cognition Antecedent to Emotion Model.
The Perceived Fear and Perceived Argument
Quality Independence Model for the Alcohol

Consumption Message.

The Perceived Argument Quality Model for
the Bicycle Helmet Message.

The Perceived Fear and Perceived Argument

Quality Independence Model for the Tetanus Message.

vii

15

15

17

18

19

19

43

50

55

REVIEW OF THE LITERATURE

Researchers interested in fear appeals have attempted to demonstrate and explain
the causal relationship between the fear arousing content in a message, attitude, and
behavior change (Chu, 1966; Higbee, 1970; Janis & Feshback, 1953; Leventhal, 1970;
McGuire, 1969; Sutton, 1982). Although there are competing explanations for how and
why fear appeals impact attitudes, most models see the fear-arousing content of persuasive
messages as the initial causal variable followed by perceived fear, attitude toward the topic
of the persuasive message, and then behavior. Put another way, most proponents of fear
appeal models argue that fear arousing messages cause persons to perceive that they are
fearful which in turn causes attitude and then behavior change (i.e., fear arousing content
—> perceived fear —>attitude change —-) behavior change). In this way, messages high in
fear arousing content should produce substantial perceived fear, and thereby, attitude and
behavior change.

Considering the results of their meta-analysis, Boster and Mongeau (1984)
conclude that, on average, researchers are able to induce mag amounts of fear.
Specifically, Boster and Mongeau found that the weighted mean of the correlations
between the fear arousing content in a message and perceived fear is .36 (with a range
from .17 to .81). The weighted mean of the correlations between fear arousing content
and attitude change is .21 (with a range fiom -.25 to .63). Using the product rule, Boster

1

and Mongeau calculated the perceived fear-attitude change correlation (ryw) by dividing
the fear arousing content-attitude change (rxw or .21) correlation into the fear arousing
content-perceived fear (rxy or .36) correlation (i.e., rxw/rxy = ryw or .21/.36 = .58). Last,
the weighted mean of the correlations between attitude change and behavior is .10 (with a
range fi'om -.36 to .69). The regressions of the consequent variables onto the fear
arousing content were all linear. That is, there was no strong evidence of curvilinear or
quadratic efi‘ects across the studies included in the meta-analysis. Results from a more
recent meta-analysis are very similar (see Mongeau, in press). According to Boster and
Mongeau,

It is important to note that, according to most theories, perceived fear

mediates the relationship between manipulated fear and attitude. If these

theories are accurate, then it follows that if the fear appeal manipulation is

not strong (if the fear manipulation-perceived fear correlation is small),

then it is not possible for the fear manipulation to have a strong impact on

attitude (p. 362).
In other words, the results of this meta-analysis call into question the causal model
proposed by most fear appeal researchers. Boster and Mongeau argue that future
researchers should be aware of the potential problems associated with fear arousing
persuasive messages (i.e., confounds). In particular, Boster and Mongeau suggest that
researchers examine the degree to which highly fearfirl messages are significantly more

discrepant from participants' initial attitudes than messages low in fear arousing content.

That is, a plausible alternative hypothesis is that the relationship between fear arousing

content and attitude change is mediated by message discrepancy (i.e., fear arousing
content -—> message discrepancy -)attitude change).

In a similar way, Witte (1993) argues that “extra-message variables should be
identified and equated across fear appeal levels to prevent confounds” (p. 148). For
instance, Witte argues that variables such as source credibility should be equated across
levels of fear arousing content to reduce the possibility of confounding efl’ects. It is also
likely that a message feature such as argument quality, and an individual difference
variable like involvement may confound the relationship between fear arousing persuasive
messages and attitude change to some degree. Examining the potential confounding
effects of these variables is important because they are central to explaining the
relationship between persuasive messages and attitude change (i.e., Heesacker, Petty, &
Cacioppo, 1983; Petty & Cacioppo, 1979, 1984; Petty, Cacioppo, & Goldman, 1981;
Petty, Cacioppo, & Schumann, 1983). The purpose of this investigation is to determine
the extent to which these specific variables confound the relationship between fear
arousing messages and attitude change. The following literature review provides direction

for the study.

Fear Arousing Persuasive Messages: An Overview

 

Defining fear appeals. Designers of fear arousing persuasive messages
attempt to induce fear in the audience to change their attitudes in the direction
advocated by the source. To construct a fear arousing persuasive message
correctly, three components must be present. First, the message must demonstrate

that there is some threat to the audience. Second, the message must Show that the

audience is susceptible or vulnerable to the threat. Third, the message should
illustrate that persons can avoid the threat by adopting the attitudes, intentions,
and/or behaviors advocated in the message (Boster & Mongeau, 1984; Burgoon,
1989; Mongeau, in press; O’Keefe, 1990). Put another way, Boster and Mongeau
(1984) contend that fear arousing persuasive messages should present an argument
with the following components:

1) You (the listener) are vulnerable to a threat.

2) Ifyou are vulnerable, then you should take action to reduce

your vulnerability.

3) Ifyou are to reduce your vulnerability, then you must accept the

recommendations contained in this message.

4) Therefore, you should accept the recommendations contained in

this message. (p. 371)

Theoretical foundations. According to Dillard (1994) and Witte (1992) there are
three major theoretical approaches that have been employed to study fear arousing
messages. These theoretical approaches are drive models, the parallel response model,
and value expectancy theories. A recent addition to this theoretical montage is Witte's
(1992) extended parallel process model. Each of these approaches will be discussed
below.

Drive models. Early studies of fear arousing persuasive messages were guided by
drive models (i.e., Hovland, Janis & Kelly, 1953). According to the proponents of these
models, fear acts as a drive that causes individuals to accept the persuasive messages

advocated by a particular source. In this way, message acceptance or attitude change is

seen as a means of reducing fear. Central to these models is the assumption that high

amounts of fear arousing content should lead to high perceived fear and that high
perceived fear should cause persons to align their attitudes with the position advocated by
the source. Conversely, messages low in fear arousing content should lead to low
perceived fear, and thereby, low message acceptance. More recent investigations have
rejected the drive explanation and proposed more cognitive approaches for explaining the
impact of fear appeals on attitude change (Beck & Frankel, 1981; Rogers, 1983; Sutton,
1982)

MM response model. Leventhal's (1970, 1971) parallel response model
represents a cognitive approach to studying fear appeals. Leventhal argues that fear
arousing messages induce danger and fear control processes in the receiver. Danger
control refers to the problem-solving that persons do to deal with the threat presented by a
fearful message. Thus, danger control processes are assumed to be cognitive. Conversely,
fear control processes are presumed to be afl‘ective. Specifically, fear control is defined as
an afl‘ective reaction that receivers experience as a means of coping with fearful messages.
Like proponents of drive models, Leventhal (1970) predicts that,

with respect to main effects, the parallel response model clearly leads us to

expect that for the most part there will be positive associations between

fear and persuasion. (p. 127)

Blue expectancy theories. A third theoretical approach for studying fear arousing
messages is characterized by value expectancy theories (Rogers, 1975, 1983; Rogers &
Mewborn, 1976; Sutton, 1982). Rogers' (1975, 1983) Protection Motivation Theory

(PMT) is an example of a value expectancy theory. According to the PMT, four elements

of a message are responsible for producing cognitive control processes: (1) perceived
susceptibility to the threat in the message, (2) perceived severity of that threat, (3)
perceived response efiicacy based on proposed responses to the threat and (4) perceived
self-emcacy. Perceived susceptibility refers to a person’s beliefs about the degree to which
they are likely to be impacted by the threat(s) presented in a fear arousing persuasive
message. Perceived severity refers to a person’s beliefs about the degree to which the
threat(s) presented in a fear arousing persuasive message are intense or mild. Perceived
response efficacy refers to a person's perception of the relative effectiveness of the
response advocated in a message at reducing the threat. Perceived self-efiicacy is the belief
that one has the ability to carry out the suggested behaviors advocated in a particular
message. According to Rogers, when each of these four variables is high, then protection
motivation is also high and so is message acceptance or attitude change. A diagram of a

causal model depicting the set of relationships predicted by PMI‘ is shown in Figure 1.

 

Severity , Perceived

Manipulation Severity \

Efiicacy ——’ Perceived ———' Protection I Attitude

Manipulation Efficacy Motivation Change

Susceptibility Perceived
Manipulation Susceptibility

 

Figure 1. A Diagram of a Causal Model Depicting the Set of Relationships
Predicted by PMT.

The extendedfiparallel process model. Witte (1992) argues that the parallel
response model is useful, but it lacks precision (p. 333). Moreover, she argues that
Leventhal's (1970) model contains only general statements about fear control processes
and danger control processes. Witte's model specifies at which points each process is
likely to be invoked. In particular, Witte describes two general paths. She contends that if
perceived efficacy and perceived threat are high, then persons are likely to initiate danger
control processes. Danger control processes are assumed to be cognitive and related
positively to protection motivation and adaptive changes (attitude change). A second
route is taken if perceived efficacy is low and perceived threat is high. In this instance,
persons are likely to initiate fear control processes. Fear control processes are assumed to
be emotional and positively related to defensive motivation and maladaptive changes.
Witte's model represents some of the more recent theoretical work on fear appeals.
Diagrams of causal models depicting the set of relationships predicted by the EPPM are

shown in Figures 2 and 3.

 

(+) (+)
Danger '—"'_’ Protection _’ Attitude
Control Motivation Change

 

Figure 2. A Diagram of a Causal Model Depicting the Set of Relationships
Predicted by the EPPM When Perceived Efiicacy and Threat are High.

 

(+) (-)
F ear —’ Defensive —" Attitude
Control Avoidance Change

 

Figure 3. A Diagram of a Causal Model Depicting the Set of Relationships
Predicted by the EPPM When Perceived Efiicacy is Low and Threat
is High.

Witte’s (1994) recent test of the EPPM shows that the model was partially
consistent with the data. Specifically, Witte found that higher perceptions of efficacy were
related positively to attitude change. Equally important, in conditions of high efficacy,
perceived threat mediated the relationship between fear and behaviors. However, some of
Witte’s predictions were not consistent with the data. For example, in conditions of high

efiicacy, perceived threat did not mediate the relationship between fear and attitude

 

ch_anga. Witte also predicted that perceived fear and threat would increase more when
efiicacy was low than when efiicacy was high. Results showed that perceived fear and
threat c_lid_n9_t vary across levels of efficacy. For a more comprehensive summary of the
results and hypotheses, see Witte (1994, p. 124). The next section describes the research
on message discrepancy.
Massage Discrepancy: An Overview

Conceptualizing message discrepang. Fink, Kaplowitz, and Bauer (1983) make a
distinction between two types of discrepancy (positional and psychological). According to

Fink and his colleagues, message or positional discrepancy refers to the absolute difi‘erence

 

between a receiver‘s position on an issue and the position advocated in a message. Thus,
message discrepancy is the difference between two positions expressed in numerical units
(Fink, Kaplowitz, & Bauer, 1983, p. 415). Psychological discrepancy refers to the extent
(i.e., slightly, moderately, or extremely) to which a receiver perceives that his or her
position is difi'erent from the position advocated in a message. In other words, even
though the numerical difference between a receiver's position (i.e., 6) and a source's
position (i.e., 7) is l, a receiver may perceive that this difference is slightly, moderately, or
extremely discrepant from his or her position.

Theoreticg foundations. The linear discrepancy model predicts the relationship

 

between message discrepancy and attitude change. This model evolved from Information
Processing Theory (Anderson 1959; Anderson & Hovland, 1957). The basic assumption
of Information Processing Theory is that receivers respond to messages by dividing them
into elements such as arguments or assertions about the object that is being described by a
source (Hunter, Danes, & Cohen, 1984). The receiver then compares each element to her
or his corresponding belief(s) about the object, target or subject. Arguments or assertions
that advocate positions that are consistent with existing beliefs will have no impact on
attitudes toward the target. Conversely, if arguments advocate a position that is difl‘erent
(discrepant) from a receiver's beliefs then there may be attitude change. Hunter and his
colleagues (1984) provide two postulates that describe the relationship between message
discrepancy and attitude change: (1) the magnitude of [attitude] change is proportional to
the discrepancy between the receiver's attitude and the position advocated by the message

and (2) the [attitude] change is always in the direction of the message (p. 36). These

ID

postulates may be expressed by the following equation that is know as the linear
discrepancy model.

Aa=or(m-a), 0<or<1
In this model Aa represents attitude change, or represents a proportion of change, m
represents a message, and a represents initial attitude. The following section discusses
other prominent variables that may confound the relationship between fear arousing
messages and attitude change.

Involvement, Argament Qaality, and Credibility: An Overview
Theoretical foundations. The Elaboration Likelihood Model (ELM) devised by

 

 

Petty and Cacioppo (1986b) is one of two modern, dual-processing theories of attitude
change. The other is the Heuristic-Systematic Model (HSM) developed by Chaiken
(1980, 1987). The present discussion is limited to the ELM for two reasons. First, these
two theories attempt explain the relationship between persuasive messages and attitude
change in a similar fashion. Second, the ELM has generated a greater amount of research
than the HSM.

The ELM. According to the ELM, there are two general routes to persuasion.
There are the central (systematic) and peripheral (heuristic) routes to persuasion. When
processing via the central route, individuals are said to engage in “issue-relevant” thinking.
To do this type of thinking, persons must have the capacity and motivation (involvement)
to think about the issue in question. Moreover, the specific thoughts that persons
generate (i.e., favorable or unfavorable) are influenced (at least partially) by the strength

or quality of the arguments in a persuasive message. For instance, the ELM predicts that

11

strong arguments should lead to the generation of favorable thoughts. These thoughts
should then lead to persuasion in the direction the source intended. Conversely, weak
arguments should lead to the generation of unfavorable thoughts. These thoughts should
then lead to no persuasion. That is, there should be no change in the direction that the
source of the message intended when arguments are weak.

When processing via the peripheral route, individuals are said to engage in little or
no “issue-relevant” thinking. In this mode of processing, persons rely on simple decision
rules or mental heuristics like the attractiveness or credibility of the source. Put another
way, persuasion via the peripheral route occurs without scrutinizing the quality of the
arguments presented in the message, and is generally more ephemeral than persuasion via
the central route. The section below provides a more detailed discussion of the role of
involvement, argument quality, and source credibility in persuasion using the ELM
framework.

Involvement. Numerous studies demonstrate the impact of involvement on
attitudes (Brickner, Harkins, & Ostrom, 1986; Chaiken, 1980; Petty, Cacioppo, &
Goldman, 1981; Johnson & Eagly, 1989). For example, involvement in a specific task is
important because increased involvement is likely to reduce motivation losses (Chaiken,
1980; Petty, Cacioppo, & Goldman, 1981). In their meta-analysis, Johnson and Eagly
(1989) identify three types of involvement. They define value-relevant involvement as
“the psychological state that is created by the activation of attitudes that are linked to
important values” (p. 290). Outcome-relevant involvement refers to the degree to which

an individual believes that an attitude object is related directly to her or his personal goals.

12

Impression-relevant involvement refers to someone’s desire to express attitudes that are
socially appropriate. According to Johnson and Eagly, persons high in impression-
relevant involvement are concerned about “holding an opinion that is socially acceptable
to potential evaluators” (p. 292).

Johnson and Eagly (1989) summarize the results of their meta-analysis of the
relationship between involvement and persuasion by saying that:

We found that (a) with value-relevant involvement, high involvement

subjects were less persuaded than low-involvement subjects; (b) with

outcome-relevant involvement, high involvement subjects were more

persuaded than low-involvement subjects by strong arguments and less

persuaded by weak arguments; and (c) with impression-relevant

involvement, high-involvement subjects were slightly less persuaded than

low-involvement subjects. (p. 305)

Argament gualig. A message that contains strong arguments has been
characterized empirically as “one containing arguments such that when subjects are
instructed to think about the message, the thoughts they generate are predominantly
favorable.” Conversely, a message that contains weak arguments has been defined
empirically as one in which the arguments “are such that when subjects are instructed to
think about them, the thoughts they generate are predominantly unfavorable” (Petty &
Cacioppo, 1986a, p. 32). The Elaboration Likelihood Model (ELM) predicts that when
receivers are motivated and capable (high elaboration) of engaging in thinking related to a
particular issue, they are more likely to scrutinize the quality of the arguments in a

message than in conditions when motivation and capability are low (low elaboration). In

this way, persons who are motivated to think about a particular issue (highly involved)

13

should evaluate the relative quality of arguments, and therefore be less persuaded by poor
arguments (i.e., faulty reasoning, bad evidence) and most persuaded by arguments that are
strong. A number of studies have demonstrated this specific effect (i.e., Heesacker, Petty,
& Cacioppo, 1983; Petty & Cacioppo, 1979, 1984; Petty, Cacioppo, & Goldman, 1981;
Petty, Cacioppo, & Schumann, 1983). Using the conceptualizations of involvement
presented in Johnson and Eagly’s (1989) meta-analysis, one can make the distinction that
persons who are lligh in outcome-relevant involvement evaluate the relative quality of
arguments, and are therefore less persuaded by poor arguments and most persuaded by
strong arguments. With the other types of involvement (value and impression-relevant)
such an interaction has not been demonstrated.

Source credibility. Scholars have argued that there are two broad dimensions of
credibility (Miller, 1987; O’Keefe, 1990). One dimension is called trustworthiness, and
refers to the degree to which receivers perceive that a source is honest. The second
dimension is known as competence or expertise, and refers to the degree to which
receivers perceive that a source is knowledgeable. According to the Elaboration
Likelihood Model (ELM), source credibility impacts attitude change only when
involvement is low. In particular, the ELM predicts that when receivers are n_ot_ motivated
(i.e., low outcome-relevant involvement) to engage in thinking related to a particular
issue, they are l_e_sa likely to scrutinize the quality of arguments than in conditions when
motivation is high (i.e., high outcome-relevant involvement). In this way, persons who are
n_ot motivated to think about a particular issue should not evaluate the relative quality of

arguments, and therefore be most persuaded by heuristic cues (i.e., source credibility).

14

Consistent with this reasoning, a number of studies have demonstrated that as involvement
increases, persons are less persuaded by the credibility of the source (i.e., Johnson &
Scileppi, 1969; Kiesler & Mathog, 1968; Petty, Cacioppo, & Goldman, 1981; Ratneshwar
& Chaiken, 1986; Rhine & Severance, 1970).
Rationale

Discrepangz. Boster and Mongeau's (1984) meta-analysis calls into question the
causal model proposed by most fear appeal researchers (i.e., fear arousing content ——>
perceived fear —)attitude change —) behavior). One problem that Boster and Mongeau
note is that the relationship between fear arousing messages and attitude change may be
confounded by message discrepancy. That is, messages that are intended to induce high
fear may actually be more discrepant from the receiver's position than messages intended
to induce low fear. This greater discrepancy in high fear conditions may be inducing
greater attitude change than in low fear conditions. Recall that Boster and Mongeau
found no strong evidence of curvilinear or quadratic efl‘ects across the studies included in
the meta-analysis. The regressions of the consequent variables onto the fear arousing
content were all linear. This finding is consistent with the predictions of the linear
discrepancy model. That is, the linear discrepancy model predicts that highly discrepant
messages (high fear) are more persuasive than messages low in discrepancy (low fear).
Using this prediction, it is argued that highly fearful messages may induce significantly
greater attitude change than messages low in fear because they are significantly more
discrepant from receiver's initial attitudes than messages low in fear. Figure 4 presents the

predicted path model.

15

 

(+) (+)

Fear arousing content —) Message discrepancy —)Attitude change

 

Figure 4. The Discrepancy Model.

Another theoretical possibility is that discrepancy could influence perceived fear.
In particular, the more discrepant the message, the less it may be believed (i.e., source is
perceived as low in credibility/trustworthiness). Therefore, less fear is perceived and less

attitude change. Figure 5 presents the predicted path model.

 

(+) H (+) (+)
Fear arousing content —> Message discrepancy ->Credibility—>Perceived fear—>Attitude
change

 

Figure 5. The Discrepancy and Credibility Model.

Although this model iS possible, there are two reasons why it is unlikely to be
consistent with the data. First, multiplying the valences in the predicted model yields a
negative relationship between fear arousing content and attitude change. In other words,
this model predicts that as fear arousing content increases attitude change decreases. This
prediction is inconsistent with much of the literature on fear appeals. Second, the model
presents a long causal string. Consequently, the multiplication of the path coefficients is

likely to yield a small relationship (close to zero) between fear arousing content and

16

attitude change. This result would also be inconsistent with the meta-analytic data
reported by Boster and Mongeau (1984).

Involvement Ed Argumenpgaality. Using Johnson and Eagly’s ( 1989) definitions
of involvement along with the ELM, one can argue that persons who are high in outcome-
relem involvement evaluate the relative quality of arguments, and are therefore less
persuaded by poor arguments and most persuaded by strong arguments. Applying this
argument to the study of fear arousing persuasive messages, it is further argued that highly
fearful messages may induce significantly greater attitude change than messages low in
fear because they induce significantly more outcome-relevant involvement than messages
low in fear. That is, highly fearfirl persuasive messages may be more efl‘ective at inducing
attitude change than messages low in fear because they present receivers with significantly
more negative and graphic consequences or outcomes than low fear messages.

Essentially, the negative and graphic depictions in highly fearful messages may induce
greater outcome-relevant involvement by focusing the receiver’s attention on undesirable
consequences or outcomes that are likely to occur if corrective action is not taken. With
this heightened level of outcome-relevant involvement, persons are more likely to evaluate
the quality of arguments critically, and thereby, be more influenced by those arguments
that they mail/a as strong than by those they m as weak. Figure 6 presents the

predicted path model.

17

 

Perceived argument quality

Fear arousing content —) Outcome-relevant involvement —)Attitude change

 

Figure 6. The Outcome-Relevant Involvement Model.

Involvement and Source Credibflg. Consistent with the predictions of the ELM,
credibility is more likely to impact attitudes in the direction the source intended when
receiver involvement is low rather than high. Thus, one would expect that persons should
be persuaded by the credibility of the source only in conditions where outcome-relevant
involvement is low. That is, when outcome-relevant involvement is low, the credibility of
the source should impact attitude change positively.

The variables discussed above can be incorporated into three large path models
that describe the relationship between fear arousing content and attitude change in terms
of emotion and cognition. The first model predicts that emotion and cognition work
independently. That is, fear arousing content causes discrepancy, perceived fear, and
outcome-relevant involvement independently. These variables then cause attitude change
with credibility and perceived argument quality moderating the relationship between
involvement and attitude change. Figure 7 presents the predicted path model. The second
model predicts that emotion is antecedent to cognition. Here, fear arousing content
causes involvement and perceived fear independently, perceived fear (emotion) then

causes discrepancy (cognition). Figure 8 presents the predicted path model. The third

18

model predicts that cognition is antecedent to emotion. In this model, fear arousing
content causes involvement and discrepancy independently, discrepancy (cognition) then

causes perceived fear (emotion). Figure 9 presents the predicted path model.

 

Discrepancy

/

Fear arousing content -—+ Perceived fear —+ Attitude change

Involvement /\

Perceived
argument quality

Credibility

 

Figure 7. The Emotion-Cognition Independence Model.

19

 

Perceived fear ——-—r Discrepancy

1

Fear arousing content Attitude change
Involvement \
Perceived
argument quality
Credibility

 

Figure 8. The Emotion Antecedent to Cognition Model.

 

Discrepancy ——’ Perceived fear

1 \

Fear arousing content Attitude change
Involvement
Perceived
argument quality
Credibility

 

Figure 9. The Cognition Antecedent to Emotion Model.

20

Evaluating the models. There are two criteria for evaluating the fit of the path
models (Boster, 1993). First, one assesses the size of the path coefficients. Ifthey are not
substantial, given what is considered substantial in fear appeal research, then the model is
rejected. Ifa particular model is rejected, then possible sources of error are diagnosed.
For example, the model may contain an insignificant mediating variable. If the path
coefiicients are substantial, then one uses a second criterion. Specifically, the extent to
which the predicted (reproduced) correlations deviate from the obtained (actual)
correlations is determined. Ifthese deviations are substantial, then the model is rejected.

One fails to reject the model, if the deviations are not substantial.

METHOD

Mgr; The experiment employed a 3 X 3 X 2 Solomon four-group design. There
were three message topics (bicycle helmets, tetanus, and alcohol consumption), three fear
conditions (high, low, and control), and the presence or absence of a pretest. Three
message topics (devised for previous studies) were used to improve the generalizability of
the present investigation. Subjects were assigned to treatment conditions randomly with
the constraint that comparable numbers of participants were included in each cell (n = 30-
32).

With random assignment to conditions, the Solomon four-group design has several
strengths. First, pretesting of groups helps assess the comparability of the experimental
and control groups. At the same time, this design assesses both the main effects of
pretesting and the interaction between the experimental induction and the pretest
(Campbell & Stanley, 1963; Smith, 1988). Boster and Mongeau (1984) found that when
correlating the type of design with the size of the fear arousing content-attitude change
correlation, there was “a tendency for larger correlations in posttest-only designs than in
pretest-posttest designs” (p. 348). They attributed this result to the relative unreliability of
change scores when compared to static scores. The Solomon four group design allows a

comparison between posttest-only and pretest-posttest designs. Thus, it is possible to

21

22

determine if the results of the present study are consistent with Boster and Mongeau’s
meta-analysis.

Consistent with the definition of fear arousing persuasive messages discussed
previously, the efiicacy of all the messages was high. Severity and susceptibility were high
in high fear conditions and low in low fear condition. Thus, measures of perceived
eflicacy, severity, and susceptibility were used to see if the inductions worked as expected.
Equally important, these measures were used to examine if the data were consistent with
the predictions of Protection Motivation Theory (PMT) and the Extended Parallel Process
Model (EPPM).

Procedures. Subjects were told that the researchers were interested in " getting
your opinions and reactions to some educational materials." In the pretest conditions
participants were asked to report their initial attitudes about one of the three topics as well
as their outcome, value, and impression-relevant involvement. All participants were asked
to report their initial position on one of the topics, and then presented with a
corresponding message that varied in fear arousing content (high, low, or control). Afier
reading the message, participants were asked to indicate what position was advocated by
the author. They were also asked to report their perceived efiicacy, fear, severity,
susceptibility, perceived involvement, and attitude. In addition, participants’ perceptions
of argument quality and source credibility were assessed. Before participants were
dismissed, the purpose of the research was explained in detail. Any questions that
participants had were answered. Participants were also asked not to reveal the nature of

the research to anyone for the next six months.

23

Participants. Participants were recruited from introductory communication classes
at a large Midwestern University. These students signed up during regular class periods
and reported to the research room in groups of about 20. They received extra credit for
their participation in the study. A total of 555 subjects participated in the study.

Bigcle helmet messages. The messages concerning bicycle helmets were originally
devised and used by Witte (1991). The low fear message for the bicycle helmet portion of
the study included statistics that demonstrated that the injuries due to bike accidents, while
relatively fiequent, were not severe (e.g., “561,765 emergency room visits were due to
bike accidents, but all were minimal injuries and less than 0.002% of children in bike
accidents die”). This message also included a black and white picture of a child who
chipped her tooth during a bicycle accident.

The high fear message included harsh, abrupt declarations of death rates and risk
factors associated with not wearing bicycle helmets. Statistics include an emphasis on
death tolls (e.g., “1300 deaths per year, head traumas causing 85% of all bicycle
fatalities”). There were several gruesome pictures of bicycling accident victims. For
instance, one picture showed a woman whose skull was severed in halfas a consequence
of being hit by a car while riding her bicycle. Both the low and high fear messages were
revised slightly for the present study by making references to MSU (i.e., Spartan Village)
instead of Texas A&M where the messages were used originally. The position advocated
in p9t_h the high and low fear messages was that one should wear a bicycle hehnet when

riding a bicycle.

24
Tetanus messages. The messages concerning tetanus were adopted fi'om the work

of Leventhal, Jones, and Trembly (1966, p. 390). Similar to the messages presented above,
the high fear material differed fi'om the low in that it was designed to create a strong sense
of personal vulnerability to tetanus. The magnitude of danger was presented in a vivid
manner through graphic language and pictures. For example, the high fear message said
that, “The danger of tetanus is a good deal more widespread than most of us realize. One
of the most fiightening aspects of the disease is that it is always possible for you to
contract it.” One of the pictures included in this message showed a “patient who
experienced massive complications during a tracheotomy that was performed to
counteract the suffocating efi’ects of a tetanus infection.” In contrast, the low fear message
said that, “Generally, the bleeding that accompanies a wound is sufficient to flush the area
free of infection-producing bacteria. This cleansing process and the relative infi'equency of
oxygen-flee environments greatly reduce the ability of the tetanus bacilli to produce the
toxin.” The low fear message included a microscopic picture of the tetanus virus. These
messages were also revised slightly for the present study by including more recent
publication dates on the case studies. The position advocated in path the high and low fear
messages was that one should get a tetanus vaccination before infection occurs.

Alcohol mesaagas. The messages concerning alcohol consumption were created
and used by Kleinot (1980) and Kleinot and Rogers (1982). The high fear material
differed from the low in that it was intended to create a strong sense of personal
vulnerability to diseases because of alcohol consumption. The high fear message included

pictures of a heart, liver, artery, and brain that had been severely damaged by alcohol

25

consumption. This message made references to serious health problems and death. For
instance the said that, “Drinking is involved in several serious health concerns including
harmful effects on the liver, the heart, the brain, and often death.” Conversely, the low
fear message said that, “Drinking is involved in several health concerns, including effects
on the liver and the heart. These effects are not that serious for someone who is an
average drinker, but it might be of benefit to be aware of them.” The position advocated
in po_tl_r the high and low fear messages was that one should drink in moderation (less than
two times a week) and not get intoxicated.

Control ggoup message. All control groups read an article about a national teacher
corps program. The article described the program and told the story of several college
graduates who decided help some of America’s troubled schools by becoming teachers.
_In_strumentation

Perceived fear. The perceived fear scale was composed of six items with Likert-
type response options. Participants responded to items like, "This message makes me feel
fiightened." Response options ranged from 1(disagree strongly) to 7 (agree strongly).
These items were used by Witte (1991, 1992, 1994) and reliable from .86 to .97.

Positional and psychological discrepangl. Positional and psychological
discrepancy were measured as in previous research (of, Fink et al., 1983; Kaplowitz,
Fink, Mulcrone, Atkin, & Dabil, 1991). In the present study, it was assumed that
positional and psychological discrepancy correlated in the same manner (positively and
Significantly) as in the work of Fink and his colleagues. It is profitable to attempt to

measure each type of discrepancy independently to assess any potential context effects.

26
To measure positional discrepancy subjects were asked to report their initial

position on each of the topics presented (i.e., helmets, tetanus, alcohol). For example,
subjects were asked to report what percentage of the time they think thg should wear a
bicycle helmet. After, reading the helmet message, they were asked to indicate what
percentage of the time the author of th_e mesgge says to wear a bicycle helmet. A
discrepancy scale was constructed by subtracting subjects’ initial position on the
discrepancy items from their perception of the author’s position.

To measure psychological discrepancy, subjects were asked to think back to the
percentage of time they thought they should wear a bicycle helmet before they read the
message, and to rate how difi‘erent they perceive their initial position was in contrast to the
position advocated in the message. Like the instructions used in previous research, the
subjects were told (c.f., Kaplowitz, et al., 1991):

Imagine that 100 is moderately difi‘erent fiom your own view. If you think

that the difference between your view and the view expressed in the

message is twice as much as the moderate difi’erence, rate it as a E. If

you think that the difi‘erence between your view and the view expressed in

the message is half as much as the moderate difference, rate it as a 5_0. If

the view expressed in the message is NOT AT ALL different from your
view, rate it 0 (zero). You may use any number you wish, from zero on

up.

Involvement. Consistent with Johnson and Eagly’s (1989) conceptualizations of
involvement, participants were asked to report their value, outcome, and impression-
relevant involvement. Each scale was composed of four items with Likert-type response
options. For example, to measure value-relevant involvement, respondents were asked to

respond to items like, “Bicycle safety is something that I value highly.” To measure

27

outcome-relevant involvement, respondents were asked to respond to items like, “Bicycle
safety has little impact on my life.” To measure impression-relevant involvement,
respondents were asked to respond to items like, “I would wear the proper safety
equipment when riding my bike ifI thought that others would think better of me for it.”
Participant response options for all involvement items ranged from 1(disagree strongly) to
7 (agree strongly). Changes in involvement were calculated by subtracting pretest scores
from posttest scores.

Perceived argament gualig. It is important to note that argument quality was not
induced in this study. In other words, messages with strong and weak arguments were not
created for this study. Instead, existing fear arousing persuasive messages were used, and
subjects were asked to indicate to what degree (i.e., high-low) they perceived that these
messages were compelling. A perceived argument quality scale was used because it is a
more direct indicator of argument quality than the generation of positive or negative
thoughts. Recall that the ELM posits that positive and negative thoughts are alleged
consequences of argument quality and mt direct indicators of the construct.

Perceived argument quality was assessed with one five-item semantic difi‘erential
scale. Participants were asked to evaluate the grimy of the arguments presented in the
message by responding to items such as high-low, outstanding-poor, superior-inferior,
great-lousy, irnpressive-unimpressive, marvelous-terrible. Participant response options for
these items ranged fi'om 1 (low) to 7 (high).

Source credibilig. Source credibility was assessed with two seven-item semantic

differential scales. One scale measured competence, and the other measured

28

trustworthiness. The competence dimension was indexed by items such as experience-
inexperienced, informed-uninformed, trained-untrained, qualified-unqualified, skilled-
unskilled, intelligent-unintelligent, expert-novice. Participant response options for these
items ranged fi'om 1 (inexperienced) to 7 (experienced). The trustworthiness dimension
was indexed by items such as honest-dishonest, trustworthy-untrustworthy, reliable-
unreliable, responsible-irresponsible, corrupt-uncompt, just-unjust, fair-unfair. Participant
response options ranged from 1 (untrustworthy) to 7 (trustworthy).

Ema. Participants were asked to report their perceived response efiicacy and
self-emcacy. These items were composed of questions with Likert-type response options.
For example, to measure perceived response eflicacy, respondents were asked to respond
to items like, “I believe that drinking alcohol in moderation is an efl’ective way of
decreasing alcohol related diseases.” To measure perceived self-efficacy, respondents
were asked to respond to items like, “I believe that drinking alcohol in moderation is easy
for me.” Participant response options for these items ranged from 1(disagree strongly) to
7 (agree strongly).

Severity and susceptibilig. Participants were asked to report the degree to which
they perceive that the threat presented in the message was severe and the degree to which
they believe they were susceptible to that threat. These items were composed of questions
with Likert-type response options. For example, to measure perceived severity,
respondents were asked to respond to items like, “I believe that alcohol consumption
results in significant diseases.” To measure perceived susceptibility, respondents were

asked to respond to items like, “I believe that my chances of having a health-related

29

problem because of alcohol consumption are high.” Participant response options ranged
fi'om 1(disagree strongly) to 7 (agree strongly).

Att_itu_d_e. Attitude toward message recommendations was assessed with one nine-
item semantic difi‘erential scale. Participants were asked to report their attitude toward the
position presented in the message (i.e., “I think that getting a tetanus shot before infection
occurs is:”) by responding to items such as good-bad, positive-negative, important-
unirnportant, smart-stupid, pleasant-unpleasant, easy-hard, nerdy-cool, right-wrong,
rewarding-unrewarding. Participant response options ranged fi'om 1 (bad) to 7 (good).
The attitude change scale was devised by subtracting pretest scores from posttest scores.

Statistical analyses. The path models were tested using ordinary least squares
(OLS) estimates of the parameters (McPhee & Babrow, 1987; Kenny, 1979: Pedhazur,
1982). This procedure generates a predicted matrix that can be compared to the obtained
values of the correlations. The difference is tested using a chi-square statistic. A
significant chi-square indicates that the predicted model departs significantly from the

obtained data.

RESULTS

Preliminm Analyses

Confirmatory factor analyses (Hunter & Gerbing, 1982) were employed to test the
dimensionality of all of the measures. This procedure Showed that the measurement model
for the scales varied as a fimction of message topic. For this reason, the results of the
factor analyses are presented by topic. The measurement models for the alcohol
consumption message are presented first followed by the bicycle hehnet and tetanus
messages.

Alcohol consumption message. Confirmatory factor analyses resulted in the
retention of four attitude items, four perceived fear items, six discrepancy items, and five
perceived argument quality items. The measurement models for all of the other
instruments were unchanged. The reliabilities of the measures were estimated by
coefficient alpha and found to be .92 for the perceived fear scale, .86 for the discrepancy
scale, .84 for the posttest outcome-relevant involvement scale, .68 for the outcome-
relevant involvement change scale, .91 for the posttest value-relevant involvement scale,
.78 for the value-relevant involvement change scale, .74 for the posttest irnpression-
relevant involvement scale, .28 for the impression-relevant involvement change scale, .85

for the susceptibility scale, .90 for the severity scale, .89 for the response eficacy scale,

30

31
.81 for the self-emcacy scale, .82 for the attitude posttest scale, .57 for the attitude change

scale, .90 for the perceived argument quality scale, .91 for the trustworthiness scale, and
.92 for the expertise scale.

Bicycle helmet messaga. Confirmatory factor analyses resulted in the retention of
five attitude items, four perceived fear items, three discrepancy items, five perceived
argument quality items, six trustworthiness, and six expertise items. The measurement
models for all of the other instruments were unchanged. The reliabilities of the measures
were estimated by coefficient alpha and found to be .94 for the perceived fear scale, .89
for the discrepancy scale, .87 for the posttest outcome-relevant involvement scale, .65 for
the outcome-relevant involvement change scale, .92 for the posttest value-relevant
involvement scale, .87 for the value-relevant involvement change scale, .58 for the posttest
impression-relevant involvement scale, .10 for the impression-relevant involvement change
scale, .86 for the susceptibility scale, .94 for the severity scale, .86 for the response
efficacy scale, .81 for the self-efiicacy scale, .88 for the attitude posttest scale, .75 for the
attitude change scale, .94 for the perceived argument quality scale, .90 for the
trustworthiness scale, and .93 for the expertise scale.

Tetanus mesgge. Confirmatory factor analyses resulted in the retention of five
attitude items, four perceived fear items, four discrepancy items, five perceived argument
quality items, six trustworthiness items, six expertise items, and three items from the
susceptibility, severity, response efficacy, and self-eflicacy scales. The measurement
models for all of the other instruments were unchanged. The reliabilities of the measures

were estimated by coeficient alpha and found to be .92 for the perceived fear scale, .81

32

for the discrepancy scale, .72 for the posttest outcome-relevant involvement scale, .48 for
the outcome-relevant involvement change scale, .88 for the posttest value-relevant
involvement scale, .84 for the value-relevant involvement change scale, .66 for the posttest
impression-relevant involvement scale, .42 for the impression-relevant involvement change
scale, .88 for the susceptibility scale, .86 for the severity scale, .85 for the response
emcacy scale, .70 for the self-efiicacy scale, .90 for the attitude posttest scale, .79 for the
attitude change scale, .91 for the perceived argument quality scale, .90 for the
trustworthiness scale, and .92 for the expertise scale.

Pretest Effects

A preliminary concern was the threat of a pretest effect on participants’ posttest
attitude judgments. A series of analyses of variance were used to test the pretest efi‘ect
hypothesis. The pretest had no statistically significant or substantial impact on posttest
attitude scores for the alcohol Ms = 20.19 no pretest, 21.04 pretest, E (1, 180) = 1.09; p
> .05, n2 = .01, g = .07), bicycle hehnet Ms = 21.21 no pretest, 21.27 pretest, E (l, 179)
= .01; p > .05, n2 = .00, g = .00), or tetanus messages (Ms = 25.34 no pretest, 25.46
pretest, E ( 1, 174) = .07; p > .05, n2 = .00, g = .00). These results are not consistent with a
pretest effect hypothesis.

Another concern was the actual persuasiveness of the messages. The analyses of
variance found that subjects in the experimental conditions changed their attitudes more
than the control groups. These differences in attitude change, however, were not all
statistically significant. Specifically, subjects in the high fear condition changed their

attitudes toward drinking alcohol in moderation (Ms = -.07 control, .55 low fear, 3.00

33
high fear, E (2, 90) = 8.60; p < .001, 112 = .16) significantly more than the low fear and

control group. A Duncan multiple-range test indicated that the low fear mean was
significantly (p < .05) different fiom the high fear mean. Dunnett’s test to compare
experimental means with control means showed that attitude change in the high fear
condition differed significantly from the control group. Attitude change in the low fear
condition did not differ significantly fi'om the control group.

Subjects in the high fear group changed their attitudes toward wearing a bicycle
helmet while riding a bike Ms = .40 control, 2.52 low fear, 4.81 high fear, E (2, 89) =
7.11; p < .01, n2 = .14) significantly more than the control group. A Duncan multiple-
range test indicated that the low fear mean was not significantly (p >.05) different fi'om the
high fear mean. Dunnett’s test indicated that attitude change in the high fear condition
differed significantly fiom the control group. Attitude change in the low fear condition did
not differ significantly fi'om the control group.

The attitude change in the experimental groups toward getting a tetanus shot
before infection occurred (Ms = .63 control, 2.03 low fear, 1.33 high fear, E (2, 86) =
1.17; p > .05, n2 = .03) did not difl’er significantly from the control group. A Duncan
multiple-range test indicated that the low fear mean was not significantly (p >.05) different
from the high fear mean. Dunnett’s test revealed no significant differences between any of
the experimental group means and the control group.

Muction Checks
Variable analysis. The responses to the measures of perceived eficacy, severity,

and susceptibility are reported to see if the inductions worked as expected. Recall that

34

consistent with the definition of fear arousing persuasive messages discussed previously,
the eficacy of all the messages was high. Severity and susceptibility were high in high fear
conditions and low in low fear condition. Equally important, these measures were used to
determine if the data were consistent with the predictions of Protection Motivation Theory
(PMT) and the Extended Parallel Process Model (EPPM).

m. For the alcohol message, subjects in the high fear condition M = 26.34)
perceived that diseases resulting fi'om alcohol consumption were significantly more severe
(E (2, 185) = 7.52; p < .01, n2 = .08) than the low fear M = 24.24) and control group M
= 24.59). A Duncan multiple-range test indicated that the low fear mean was significantly
(p <.05) different fi'om the high fear mean. Dunnett’s test revealed that the difi‘erence
between the high fear group and the control group was significant. The difference
between the low fear mean and the control group mean was not statistically significant.

For the bicycle helmet message, subjects in the high fear condition M = 24.41)
perceived that injuries resulting fi'om bicycling accidents were significantly more severe (E
(2, 184) = 29.07; p < .001, n2 = .24) than the low fear M = 18.15) and control group M
= 20.87). A Duncan multiple-range test indicated that the low fear mean was significantly
(p <.05) different from the high fear mean. Dunnett’s test revealed that the differences
between the experimental groups and the control group were significant.

For the tetanus message, subjects in the high fear condition M = 19.98) perceived
that tetanus infections were significantly more severe (E (2, 177) = 38.59; p < .001, n2 =
.30) than the low fear M = 17.67) and control group M = 14.63). A Duncan multiple-

range test indicated that the low fear mean was significantly (p <.05) different from the

3 5
high fear mean. Dunnett’s test revealed that the differences between the experimental

groups and the control group were significant.

Susceptibilig. Subjects who read the alcohol message and were in the control
group M = 13.72) perceived that they were significantly more susceptible to diseases
resulting from alcohol consumption (E (2, 185) = 3.20; p < .05, n2 = .03) than the low fear
M = 11.44) but not the high fear group M = 11.89). Duncan’s multiple-range test
revealed that the difference between the high and low fear groups was not significant (p >
.05). Dunnett’s test indicated that the control group perceived great susceptibility than the
low fear group. The difference between the high fear group and the control group was
not statistically significant.

Subjects who read the bicycle helmet message and were in the control group M =
19.39) seemed to perceive that they were more susceptible to injuries resulting fiom
bicycling accidents than the low fear M = 17.65) and high fear group M = 17.46). This
difi‘erence, however, was not statistically significant (E (2, 184) = 1.89; p > .05, n2 = .02).
Duncan’s multiple-range test revealed that the difference between the high and low fear
groups was not significant (p > .05). Dunnett’s test indicated that the difi‘erences between
the control group and the experimental groups were not statistically significant.

Subjects who read the tetanus message and were in the high fear condition M =
9.63) perceived that they were significantly more susceptible to tetanus infection (E (2,
177) = 6.68; p < .01, n2 = .07) than the low fear M = 7.12) and control group M =
7.38). Duncan’s multiple-range test indicated that the difference between the high and low

fear groups was significant (p > .05). Dunnett’s test indicated that the difference between

36
the high fear group and the control group was statistically significant. The low fear mean

and the control group mean were not significantly different from one another.

Response efficacy. For the alcohol message, subjects in the experimental groups
(Ms = 2054 low fear, 21.97 high fear, E (2, 135) = 6.13; p < .01, n2 = .06) perceived that
drinking alcohol in moderation was a significantly more effective way of reducing diseases
related to alcohol consumption than the control group M = 18.30). Duncan’s multiple-
range test revealed that the difi‘erence between the high and low fear groups was
significant (p < .05). Dunnett’s tests indicated that the experimental groups perceived
great response eficacy than the control group.

For the bicycle helmet message, subjects in the high fear condition M = 24.71)
perceived that wearing a bicycle hehnet while riding a bicycle was a significantly more
efi‘ective way (E (2, 184) = 15.82; p < .001, n2 = .15) of reducing injuries from bicycling
accidents than the low fear M = 21.68) and control group M = 20.10). Duncan’s
multiple-range test revealed that the difference between the high and low fear groups was
significant (p < .05). Dunnett’s tests indicated that the high fear group perceived greater
response efficacy than the control group. The difference between the low fear group and
the control group was not statistically significant.

For the tetanus message, perceptions of response efficacy in the high fear group
difl‘ered significantly MS = 19.53 km fear, 20.17 high fear, E (2, 177) = 4.60; p <05, 112
= .05) from the control group M = 19.02). Duncan’s multiple-range test revealed that the
difference between the high and low fear groups was not significant (p > .05). Dunnett’s

tests indicated that the high fear group perceived greater response efficacy than the control

37
group. The difference between the low fear group and the control group was not

statistically significant.

Self-efficaay. Subjects in the experimental groups for the alcohol message Ms =
20.27 low fear, 21.39 high fear, E (2, 182) = 1.11; p >.05, n2 = .01) did not differ
significantly {Tom the control group M = 21.56) in their perceptions of self-efficacy.
Duncan’s multiple-range test revealed that the difl'erence between the high and low fear
groups was not significant (p > .05). Dunnett’s tests indicated that the differences
between the experimental groups and the control group were not statistically significant.

Subjects who read the bicycle helmet message and were in the high fear condition
(M = 20.55) perceived that wearing a bicycle helmet when riding a bicycle was easy for
them significantly more (1i (2, 183) = 22.81; 12 < .001, n2 = .20) than the low fear (M =
15.58) and control group M = 15.15). Duncan’s multiple-range test revealed that the
difi‘erence between the high and low fear groups was significant (p < .05). Dunnett’s tests
indicated that the high fear group perceived greater self-efficacy than the control group.
The difference between the low fear group and the control group was not statistically
significant.

Subjects who read the tetanus message and were in the high fear condition M =
18.32) perceived that getting a tetanus shot before infection occurred was easy for them
significantly more (15 (2, 177) = 3.74; p < .05, n2 = .04) than the low fear (M. = 17.13) and
control group M = 16.83). Duncan’s multiple-range test revealed that the difi’erence
between the high and low fear groups was significant (p < .05). Dunnett’s tests indicated

that the high fear group perceived greater self-efiicacy than the control group. The

38
difference between the low fear group and the control group was not statistically

significant.

Perceived fear. Subjects who read the alcohol message and were in the high fear
condition M = 13.89) perceived significantly more fear (E (2, 183) = 14.48; p < .001, n2
= .14) than the low fear M = 8.98) and control group M = 9.79). Duncan’s multiple-
range test showed that the difi‘erence between the high and low fear groups was significant
(p < .05). Dunnett’s tests indicated that the high fear group perceived greater fear than
the control group. The difference between the low fear group and the control group was
not statistically significant.

Subjects who read the bicycle helmet message and were in the high fear condition
(M = 18.92) perceived Significantly more fear (13(2, 183) = 52.16; p < .001, n2 = .36) than
the low fear M = 9.74) and control group M = 11.28). Duncan’s multiple-range test
revealed that the difference between the high and low fear groups was significant (p <
.05). Dunnett’s tests indicated that the high fear group perceived great fear than the
control group. The difl‘erence between the low fear group and the control group was not
statistically Significant.

Subjects who read the tetanus message and were in the high fear condition M =
17.45) perceived significantly more fear (E (2, 177) = 31.42; p < .001, n2 = .26) than the
low fear M = 13 .38) and control group M = 9.60). Duncan’s multiple-range test
revealed that the difl‘erence between the high and low fear groups was significant (p <
.05). Dunnett’s test indicated that the experimental groups perceived great fear than the

control group.

39
In general these data show that the inductions worked as expected. There were,

however, some effects that varied from one message topic to the other. Furthermore, the
measurement models for individual scales varied as a fiinction of message topic. For these
reasons it was determined that the tests of the predicted path models should be conducted
by message topic. The tests of the path models for the alcohol message are presented
first, followed by the bicycle helmet and tetanus messages.
Alcohol Consumption Mesgge

To assess the drinking behavior of the sample, subjects were asked how many
times and how much they drank on an average week. They were also asked how many
times and how much they drank “last week.” Subjects reported that they drank M = 2.5
times during an average week and that they had M = 7.40 drinks during an average week.
Subjects also reported that they drank M = 2.0 times last week and that they had M = 6.65
drinks last week. These data indicate that the sample was composed mostly of persons
who drank moderate to heavy amounts of alcohol. There were relatively few non-drinkers
and they were distributed proportionately across cells (4-8 per cell). Subsequent analyses
were run including and excluding the non-drinkers. No major difl‘erences were found
between these analyses. Therefore, the non-drinkers were included in the final analyses.

To see if these data were consistent with Rogers' (1975, 1983) Protection
Motivation Theory (PMT) or Witte’s (1992, 1994) Extended Parallel Process Model
(EPPM) scores on the perceived severity, susceptibility, response efiicacy, and self-
efiicacy scales were divided into high and low groups using a median split. Recall that

Rogers argued that when each of these four variables is high, then protection motivation is

40
also high and so is message acceptance or attitude change. The analysis of variance

showed that when these four variables were at a high level, their effect on attitude change
M = 1.75 high, 1.88 low) was neither significant nor substantial (_r = .00, E(1, 60) = .00,
p > .05). These results are not consistent with the PMT.

Witte (1992, 1994) contends that if perceived efiicacy and perceived threat are
high, then persons are likely to initiate danger control processes. Danger control
processes are assumed to be cognitive and related positively to protection motivation and
adaptive changes or attitude change. On the other hand, if perceived efiicacy is low and
perceived threat is high, persons are likely to initiate fear control processes. Fear control
processes are assumed to be emotional and positively related to defensive motivation and
maladaptive changes. This specific interaction was tested with attitude change serving as
the dependent variable. The analysis of variance indicated that the interaction between
perceived threat (severity and susceptibility) and perceived eflicacy (response and self-
efiicacy) was neither significant nor substantial (E(1, 58) = 1.65, p > .05, n2 = .03, g =
.17). These results are not consistent with the EPPM. Table 1 presents the obtained

means and standard deviations.

41
Table 1

Means and Standard Degrations for the Test of the EEPM
Using the Alcohol Consumption Message

 

 

High Threat Low Threat
High Efficacy M = 1.75 M = 2.60

STD— 3 l7 STD=2 55

n=4 n= 15
Low Efiicacy M = 3.40 M= 1.09

S = 3 20 STD = 3 95

n =10 n = 33

 

The three way interaction predicted by the Elaboration Likelihood Model (ELM)
was also tested. The Elaboration Likelihood Model (ELM) predicts that when receivers
are motivated and capable (high outcome-relevant involvement) of engaging in thinking
related to a particular issue, they are likely to scrutinize the quality of the arguments in a
message, and therefore be less persuaded by poor arguments (i.e., faulty reasoning, had
evidence) and most persuaded by arguments that are strong. The ELM also predicts that
when receivers are E! motivated (i.e., low outcome-relevant involvement) to think about
a particular issue, they should not evaluate the relative quality of arguments, and therefore
be most persuaded by heuristic cues (i.e., source credibility).

To test the three way interaction predicted by the ELM, scores on the perceived
argument quality, outcome-relevant involvement, and expertise scales were divided into

high and low groups using a median split. To assess their impact on attitude change an

42
analysis of variance was conducted. This analysis indicated that the predicted three way

interaction was not significant (E (1, 54) = 1.68, p > .05). An examination of the means
found that the patterns of interactions were relatively meaningless and not consistent with

the predictions of the ELM. Table 2 presents the obtained means and standard deviations.

Table 2

Means and Standard Deviations for the Teg of the E_LM Using the
Alcohol Consumption Mesgge

 

 

High Credibility Low Credibility
High Involve. Low Involve. High Involve. Low Involve.
High Argu. Qua]. M = 2.62 M = 2.38 M = 3.50 M = .83
STD = 5.16 STD = 4.22 STD = 3.25 STD = 4.08
n = 13 n = 13 n = 2 n = 6
Low Argu. Qua]. M=3.17 M=-.25 M= .73 M= 86
STD = 4.24 STD = 4.09 STD = 2.87 S = 3.20
n = 6 n = 4 n = 11 n = 7

 

An examination of the correlation matrix generated to test the predicted models
showed that all of the models had some deficiencies (this correlation matrix is provided in
Appendix A). In other words, none of the predicted models fit the data well. For
instance, the relationship between fear arousing content and discrepancy was neither
substantial nor significant (; = .08, E(1, 56) = .32, p > .05). Thus, models that posited that

the fear arousing content-attitude change relationship was mediated by discrepancy were

43

rejected. Likewise the relationship between fear arousing content and outcome-relevant
involvement was neither substantial nor significant (1 = .08, E (1, 61) = .43, p > .05 when
the posttest was used, and I = .15, E(1, 61) = 1.36, p > .05 when the change scale was
used). For this reason, models that posited that the fear arousing content-attitude change
relationship was mediated by outcome-relevant involvement were rejected. Table 3
displays the descriptive statistics for the dependent measures of the alcohol consumption
message.

After the predicted models failed, the correlation matrix was explored to find and
test other theoretically meaningfirl causal models. The model that fit the data best was a
model that used perceived fear and perceived argument quality as independent mediators
of the relationship between fear arousing content and attitude change. An inspection of
scatterplots produced no evidence of nonlinearity, and no sign of extreme outliers.
Subsequent analyses showed no evidence of nonadditivity. The distribution of each
measure approximated the normal distribution closely. The resulting model, with path
coefficients, is presented in Figure 10. The path coefficients corrected for attenuation due

to error of measurement are in parentheses.

 

Perceived

argument quality
M 3 0) (.32) .23
Fear arousing conten Attitude change

.49 (.51) 689/
Perceived f .24

Figure 10. The Perceived Fear and Perceived Argument Quality Independence
Model for the Alcohol Consumption Message.

 

44

The fit of the model was assessed in two ways. First, predicted values were
generated fi'om both correlations not constrained to equal their obtained value, the
perceived argument quality-perceived fear correlation and fear arousing content-attitude
change correlation, to see if the predicted values were within sampling error of those
obtained. Both residuals (the difference between the obtained and the predicted
correlations) were small, .02 and .14 respectively, and within sampling error of zero.
Second, chi-square was employed as a global measure of fit. This model yielded a small
and insignificant chi-square (X30) = .61, p > .05 and X (2) = .34, p > .05 when the
correlations corrected for attenuation due to error of measurement were used). Both
methods lead to the same conclusion; namely, that the data are consistent with the model.
The path fi'om fear arousing content to perceived argument quality was significant t(60) =
2.19, p < .05, and so was the path from fear arousing content to perceived fear 1(60) =
4.05, p < .001.

A multiple regression analysis regressing attitude change onto perceived argument
quality and perceived fear showed that both variables impacted attitude change (E = .36,
E(2, 58) = 4.35, p < .05 and E = .50 when corrected for attenuation due to error of
measurement). The efi‘ect for perceived fear was substantial and significant (the
standardized regression coeflicient, E = .24, 1(58) = 1.96, p = .05), and the efl‘ect for
perceived argument quality was substantial and approached significance (the standardized
regression coeficient, E = .23, t(58) = 1.86, p = .07). The correlations among the

variables in this model are presented in Table 4 along with the reliabilities, means, and

45
Table 3

Descriptive Statistics for the Dengdent Vari_ables of the Alcohol Consumption Message

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

VARIABLE CONTROL GROUP LOW FEAR HIGH FEAR E
M .819 n M SE n .1! SE n

Perceived

Ee_ar 9.79 5.11 61 8.98 5.11 62 13.89 6.11 63 186

Discrapancy -15.88 40.46 59 -14.14 22.56 58 -24.48 42.50 62 179

Psychological

Discrepancy 60.07 56.39 61 70.54 59.70 61 60.56 67.57 63 185

Impression

Involvement

_Pasttest 17.95 5.46 59 17.67 5.17 63 17.25 5.16 64 186

Impression

Involvement

Ch_anga -.63 3.08 30 -1.06 3.10 31 .25 3.05 32 93

Outcome

Involvement

Posttest 9.80 5.11 61 10.59 4.93 63 11.55 5.26 64 188

Outcome

Involvement

Ch_ange .00 2.23 30 -.06 3.43 31 1.31 5.63 32 93

Value

Involvement

Posttest 15.20 7.14 60 15.73 5.77 63 17.69 6.71 64 187

Value

Involvement

Ch_ange -.27 3.67 30 -.26 3.37 31 2.91 5.29 32 93

Perceived

Argument

Qualifl 24.92 4.45 61 25.95 5.17 62 28.38 5.14 63 186

Expertise 40.59 6.26 61 40.75 8.47 61 43.55 4.89 64 186

Trust 39.85 6.57 61 38.69 9.44 62 42.17 6.23 63 186

Response

Efficacy 18.30 6.47 61 20.54 5.81 63 21.97 5.40 64 188

Self-Efficacy 21.56 4.61 59 20.27 5.56 62 21.39 5.33 64 185

Severity 24.59 3.48 61 24.24 3.66 63 26.34 2.61 64 188

Susceptibility 13.72 5.41 61 11.44 5.09 63 11.89 5.39 64 188

AttitudePosttest 18.84 6.20 61 21.08 5.06 61 21.88 5.01 64 186

AttitudeChang -.07 1.86 30 .55 2.61 31 3.00 4.25 32 93

 

46

standard deviations for the corresponding scales. The corrected correlations are reported

on the upper triangle of the matrix.

Table 4

Correlations and Descriptive Statistics for the Perceived Fear and
Perceived Argument Qualig Independence Model for the Alcohol

Consumption Message

 

 

FEAR PFEAR ARGUQ ATTCHN ALPHA NIEAN STD

 

FEAR 1.00 .51 .30 .42 1.00 -- -

PFEAR .49 1.00 .18 .39 .92 12.30 6.45
ARGUQ .28 .16 1.00 .38 .90 27.67 4.77
ATTCHN .32 .28 .27 1.00 .57 1.79 3.72

 

Bigcle Helmet Mesgge

To determine the extent to which the sample used bicycle helmets, subjects were
asked, “What percentage of the time do you actually wear a bicycle helmet when riding a
bicycle?” Response options ranged from 0 to 100. Subjects reported that they wore a
bicycle helmet M = 3.23% of the time and 87.2% of the sample reported that they wore a
bicycle helmet 0% of the time. These data indicate that the sample was composed mostly
of persons who did not wear a bicycle hehnet when riding a bicycle. There were relatively
few subjects who reported that they wore a bicycle helmet 100% of the time (1.6% of the
sample). Subsequent analyses were run including and excluding subjects who reported

that they wore helmets 100% of the time. No major differences were found between these

47
analyses. Therefore, those persons who reported that they wore helmets 100% of the time

were included in the final analyses.

To see if these data were consistent with Rogers' (1975, 1983) Protection
Motivation Theory (PMT) or Witte’s (1992, 1994) Extended Parallel Process Model
(EPPM) scores on the perceived severity, susceptibility, response efficacy, and self-
eflicacy scales were divided into high and low groups using a median split. When these
four variables were at a high level, their effect on attitude change (Mg = 4 .64 high, 4.25
low) was neither significant nor substantial (g = .05, E(1, 60) = .13, p > .05). The
interaction between perceived threat (severity and susceptibility) and perceived efiicacy
(response and self-efficacy) was neither significant nor substantial (E (1, 58) = .67, p >
.05, n2 = .01, g = .10). As explained in the previous section, the former results are not
consistent with PMT, and the latter are not consistent with the EPPM. Table 5 presents
the obtained means and standard deviations.

As in previous analyses, outcome-relevant involvement, perceived argument
quality, and expertise, did not impact attitude change as predicted by the Elaboration
Likelihood Model (ELM). Once again, scores on the perceived argument quality,
outcome-relevant involvement, and expertise scales were divided into high and low groups
using a median split. To assess their impact on attitude change an analysis of variance was

conducted.

Table 5

Means and Standard Deviations for the Test of the Efl’M

Using the Bicycle Helmet Messaga

 

 

Hrgh Threat Low Threat
High Efficacy M = 4.64 M = 6.25
S = 3.32 S = 3.79
n = 14 n = 12
Low Eficacy M=4 00 M= 3.55
S__ = 2.00 _S_ = 3.55
n = 3 n = 33

 

This analysis indicated that the predicted three way interaction was not significant

(E(1, 54) = 2.26, p > .05). An examination of the means found that the patterns of

interactions were relatively meaningless and not consistent with the predictions of the

ELM. Table 6 presents the obtained means and standard deviations.

An examination of the correlation matrix generated to test the predicted models

showed once again that all of the models had some deficiencies (this correlation matrix is

provided in Appendix B). For example, the relationship between fear arousing content and

discrepancy was neither substantial nor significant (1 = .02, E(1, 60) = .02, p > .05). Thus,

models that posited that the fear arousing content-attitude change relationship was

mediated by discrepancy were rejected.

Table 6

49

Means and Standard Deviations for the Test of the ELM Using the Bigcle Helmet

 

 

Message
High Credibility Low Credibility

High Involve. Low Involve. High Involve. Low Involve.

High Argu. Qua]. M = 4.30 M = 4.70 M = 5.22 M = 2.50
STD= 4.63 STD=3 67 STD= 5.82 STD=4.19
n = 20 n = 10 n = 9 n = 4

Low Argu. Qua]. M = 4.00 M = -.25 M = .83 M = 3.57
STD = 2.12 STD = 5.00 STD = 2.43 S = 5.88
n=2 n=4 n=6 n=7

 

In addition, the relationship between fear arousing content and outcome-relevant
involvement was neither substantial nor significant (1‘ = .16, E (1, 60) = 1.48, p > .05 when
the posttest was used, and I = .16, E(1, 60) = 1.56, p > .05 when the change scale was
used). Therefore, models that posited that the fear arousing content-attitude change
relationship was mediated by outcome-relevant were rejected. Table 7 displays the
descriptive statistics of the dependent measures for the bicycle helmet message.

After the predicted models failed, the correlation matrix was explored to find and
test other theoretically meaningful causal models. The model that fit the data best was a
model that used perceived argument quality as a mediator of the relationship between fear
arousing content and attitude change. An inspection of scatterplots produced no evidence

of nonlinearity, and no Sign of extreme outliers. Subsequent analyses showed no evidence

50

of nonadditivity. The distribution of each measure approximated the normal distribution
closely. The resulting model, with path coefficients is presented in Figure 11. The

corrected path coemcients are in parentheses.

 

.45 .23
Fear arousing content—v Perceived argument quality —*Attitude change
(.46) (.27)

 

Figure 11. The Perceived Argument Quality Model for
the Bicycle Helmet Message.

The fit of the model was assessed in two ways. First, a predicted value was
generated from the correlation not constrained to equal the obtained value, the fear
arousing content-attitude change correlation, to see if the predicted value was within
sampling error of the obtained. The residual (the difference between the obtained and the
predicted correlation) was small, .11, and within sampling error of zero. Second, chi-
square was employed as a global measure of fit. This model yielded a small and
insignificant chi-square(L2(1)= .30, p > .05 and L2 (1) = .39, p > .05 when corrected

correlations were used), indicating that the model fits the data.

Table 7

51

DescrErtive Statistics for the Dependent Variables of the Bigcle Helmet Mesgge

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

VARIABLE CONTROL GROUP LOW FEAR HIGH FEAR bl

M S_TD n M S_TD. n M SID n
Perceived
M 11.28 5.17 61 9.74 5.04 62 18.92 5.88 63 186
Discrapancy 170.13 97.96 62 201.26 103.70 62 188.43 115.62 63 187
Psychological
Discrepapicy 112.31 69.04 62 335.77155675 62 86.81 79.23 62 186
Impression
Involvement
Posttest 10.55 3.84 62 9.61 4.10 62 11.87 4.38 63 187
Impression
Involvement
§h_a_nge .30 2.85 30 .00 3.33 31 2.61 3.32 31 92
Outcome
Involvement
Posttest 15.98 7.06 62 15.94 6.61 62 16.92 6.28 63 187
Outcome
Involvement
Qhanga .00 2.35 30 1.64 4.74 31 3.19 4.94 31 92
Value
Involvement
Posttest 10.06 5.23 62 10.53 5.40 62 13.65 5.68 63 187
Value
Involvement
_Ch_ange -1.57 2.22 30 .58 5.03 31 4.58 5.66 31 92
Perceived
Argument
Qrglity 23.98 5.70 61 22.95 6.73 62 29.18 4.45 62 185
Expertise 34.49 5.10 61 34.48 6.46 62 35.92 5.25 63 186
Trust 34.37 5.47 60 35.26 6.35 62 36.18 6.09 62 184
Response
Efficacy 20.10 5.43 62 21.68 5.00 62 24.713.33 63 187
Self-Efficacy 15.15 5.58 62 15.58 4.67 62 20.55 4.53 62 185
Severity 20.87 4.65 62 18.15 5.16 62 24.413.96 63 187
Susceptibility 19.39 6.06 62 17.65 6.08 62 17.46 6.14 63 187
AttitudePosttgt_Z7.84 5.88 61 28.37 5.59 62 32.67 3.66 63 186
AttitudeCme .40 2.77 30 2.52 4.47 31 4.81 5.87 31 92

 

52
The path from fear arousing content to attitude change was significant t(60) =

3.95, p < .001, and the path from perceived argument quality to attitude change
approached significance 1(60) = 1.86, p = .07. The correlations among the variables in this
model are presented in Table 8 along with the reliabilities, means, and standard deviations
for the corresponding scales. The corrected correlations are reported on the upper

triangle of the matrix.

Table 8

Correlations and Descriptive Statistics for the Perceived Argament Qaalig Model
for the Bicycle Helmet Message

 

FEAR ARGUQ ATTCHN ALPHA MEAN STD

 

FEAR 1.00 .46 .25 1.00 -- --
ARGUQ .45 1.00 .27 .94 25.56 6.19
ATTCHN .22 .23 1.00 .78 3.66 5.33

 

Tetanus Message

To see if these data were consistent with Rogers' (1975, 1983) Protection
Motivation Theory (PMT) or Witte’s (1992, 1994) Extended Parallel Process Model
(EPPM) scores on the perceived severity, susceptibility, response efficacy, and self-
efiicacy scales were divided into high and low groups using a median split. When these
four variables were at a high level, their effect on attitude change M = 1 .25 high, 1.79

low) was neither significant nor substantial (; = .06, E(1, 57) = .20, p > .05). The

 

53

interaction between perceived threat (severity and susceptibility) and perceived eficacy
(response and self-efficacy) was neither significant nor substantial (E (1, 55) = .06, p >
.05, g = .03). As explained in the analyses for the alcohol message, the former results are
not consistent with PMT, and the latter are not consistent with the EPPM. Table 9

presents the obtained means and standard deviations.

Table 9

Means and Standard Deviations for the Test of the Efl’M
Using the Tetanus Message

 

 

ngh Threat Low Threat
High Efficacy M= 1.25 M= 1.69

S =3.01 S = 1.30

n= 12 n=13
Low Efficacy M= 1.11 M= 2.08

8 =4.39 S =4.36

n=9 n=25

 

Outcome-relevant involvement, perceived argument quality, and expertise, did not
impact attitude change as predicted by the Elaboration Likelihood Model (ELM). Scores
on the perceived argument quality, outcome-relevant involvement, and expertise scales
were divided into high and low groups using a median split. To assess their impact on
attitude change an analysis of variance was conducted. This analysis indicated that the

predicted three way interaction was not significant (E(1, 50) = .18, p > .05). An

54

examination of the means found that the patterns of interactions were relatively

meaningless and not consistent with the predictions of the ELM. Table 10 presents the

obtained means and standard deviations.

Table 10

Means and Standard Deviations for the Test of the ELM Using the Team Message

 

 

High Credibility Low Credibility
High Involve. Low Involve. High Involve.‘ Low Involve.
High Argu. Qual. M: .69 M= 1.82 M: .00 M= 1.33
STD = 2.46 STD = 4.42 STD = 71 STD = 2.3
n = 13 n = 11 n = 5 n = 3
Low Argu. Qual. M = .00 M = .40 M = 1.75 M = 4.36
STD = .00 STD = 1.14 S ,= 2.43 S = 5.78
n=2 n=5 n=8 n=1l

 

An examination of the correlation matrix generated to test the predicted models
showed once again that all of the models had some deficiencies (this correlation matrix is
provided in Appendix C). For example, the relationship between fear arousing content and
discrepancy was neither substantial nor significant (1' = .08, E(1, 58) = .38, p > .05).
Therefore, models that posited that the fear arousing content-attitude change relationship
was mediated by discrepancy were rejected. Furthermore, the relationship between fear

arousing content and outcome-relevant involvement was neither substantial nor significant

55
(g = .04, E(1, 58) = .11, p > .05 when the posttest was used, and g = .18, E(1, 58) = 2.09,

p > .05 when the change scale was used). Thus, models that posited that the fear arousing
content-attitude change relationship was mediated by outcome-relevant involvement were
rejected. Table 11 presents the descriptive statistics for the dependent measures of the
tetanus message.

After the predicted models failed, the correlation matrix was explored to find and
test other theoretically meaningful causal models. To see if the tetanus message data were
consistent with the data for the other topics, a path model was tested using perceived fear
and perceived argument quality as independent mediators of the relationship between fear
arousing content and attitude change. An inspection of scatterplots produced no evidence
of nonlinearity. Two outliers were detected and deleted fi'om subsequent analyses. These
analyses showed no evidence of nonadditivity. The distribution of each measure
approximated the normal distribution closely. The resulting model, with path coefiicients,

is presented in Figure 12. The corrected path coefficients are in parentheses.

 

Perceived

argument quality
.26 (.28) (aw-.09
Fear arousing content Attitude change

.34 NE) (-.09)
Perceived fear /-07'

Figure 12. The Perceived Fear and Perceived Argument Quality Independence
Model for the Tetanus Message.

 

Table 11

56

Descriptive Matistics for the Dependent Variables of the Tetanus Mesgge

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

VARIABLE CONTROL GROUP LOW FEAR HIGH FEAR E
M SlQ n M $2 I! M SE n

Perceived

Eear 9.60 4.75 61 13.38 5.62 60 17.45 5.85 60 180

Discrepgq 78.95 91.78 60 107.98 131.95 60 102.23 129.91 60 180

Psychological

Discrepgcy 50.31 51.42 60 61.20 58.85 60 50.67 57.03 60 180

Impression

Involvement .

Posttest 9.92 4.18 60 10.57 4.86 60 11.18 4.19 60 180

Impression

Involvement

thnge -.53 3.81 30 2.07 4.66 30 1.30 3.62 30 90

Outcome

Involvement

Posttest 19.08 5.53 60 22.13 4.55 60 23.02 4.19 60 180

Outcome

Involvement

thnge -.83 3.59 30 2.50 4.93 30 4.10 3.52 30 90

Value

Involvement

Posttest 14.00 6.21 60 17.18 5.74 60 18.98 6.24 60 180

Value

Involvement

QMpgg -l.33 3.13 30 2.47 5.20 30 5.20 6.53 30 90

Perceived

Argument

QrLality 24.58 6.37 60 27.87 5.57 60 30.55 3.99 60 180

Expertise 34.88 4.84 59 37.46 6.39 59 39.12 4.14 60 178

Trust 34.93 5.13 59 36.59 5.73 59 37.83 5.13 60 178

Response

Efficacy 19.02 2.47 60 19.53 2.24 60 20.17 1.37 60 180

Self-Efficacy 16.83 3.24 60 17.13 2.90 60 18.32 3.27 60 180

Severity 14.63 4.17 60 17.67 3.47 60 19.98 2.27 60 180

Susceptibility 7.38 3.11 60 7.12 3.46 60 9.63 5.46 60 180

AttitudePosttest 33.03 2.86 60 32.78 3.45 60 34.10 2.21 60 180

AttitudeCthge .63 3.17 30 2.03 4.48 29 1.33 2.72 30 89

 

57

The fit of the model was assessed in two ways. First, predicted values were
generated from both correlations not constrained to equal their obtained value, the
perceived argument quality-perceived fear correlation and fear arousing content-attitude
change correlation, to see if the predicted values were within sampling error of those
obtained. Both residuals (the difi‘erence between the obtained and the predicted
correlations) were small, .10 and -.05 respectively, and within sampling error of zero.
Second, chi-square was employed as a global measure of fit. This mode] yielded a small
and insignificant chi-square (_X_2(2) = .29, p > .05 and L2 (2) = .37, p > .05 when corrected
correlations were used). The path from fear arousing content to perceived argument
quality was significant 1(58) = 2.09, p < 05, and so was the path fiom fear arousing
content to perceived fear t(58) = 2.79, p < .01. A multiple regression analysis regressing
attitude change onto perceived argument quality and perceived fear showed that these
variables did not impact attitude change significantly or substantially (E = .12, E(2, 56) =
.43, p > .05 and E = .15 when corrected for attenuation due to error of measurement).
The efl’ect for perceived fear was neither significant nor substantial (E = -.07, t(56) = -.55,
p > .05), and the effect for perceived argument quality was neither significant nor
substantial (E = -.09, t(56) = -.64, p > .05). Since the size of these path coefi'rcients was
extremely small and insignificant, the model was rejected. No other model fit these data.

An examination of the distribution of scores on the attitude pretest showed that the

sample had a positive attitude toward the position advocated in the message before the

 

induction. For example, 50% of the subjects in the high fear condition reported that they

had the most positive attitude possible (scale maximum = 35) toward getting a tetanus

58
shot before infection occurred. Similarly, 33.3% of the subjects in the low fear condition

reported that they had the most positive attitude possible (scale maximum = 35) toward
getting a tetanus shot before infection occurred. Since most of the subjects already had
the most positive attitude possible on the scale, the induction had little impact on attitude
change. The correlations among the variables used in the path model are presented in
Table 12 along with the reliabilities, means, and standard deviations for the corresponding

scales. The corrected correlations are reported on the upper triangle of the matrix.

Table 12

Correlations and Descriptive Statistics for the Perceived Fear and

Perceived Argpment Qaaligy Independence Model for the Tetanus Message

 

 

FEAR PFEAR ARGUQ ATTCHN ALPHA MEAN STD

 

FEAR 1.00 .35 .28 -.11 1.00 -- --
PFEAR .34 1.00 .21 -.11 .92 15.75 6.30
ARGUQ .26 .19 1.00 -.12 .88 29.08 4.13

ATTCHN -.10 -.09 -.10 1.00 .79 1.68 3.67

 

DISCUSSION

Overview

This study examined the impact of potential confounds in the relationship between
fear arousing persuasive messages and attitude change. A series of causal models were
developed and tested. The first set of models posited that discrepancy or outcome-
relevant involvement mediated the relationship between fear arousing content and attitude
change. These models were incorporated into three larger models that predicted that
discrepancy, outcome-relevant involvement, and perceived fear mediated the relationship
between fear arousing content and attitude change. The data were inconsistent with the
predicted models. These data were also inconsistent with Protection Motivation Theory
(PMT), the Extended Parallel Process Model (EPPM), and the Elaboration Likelihood
Model (ELM). A series of similar causal models were found to fit the data, however. In
these models perceived fear and perceived argument quality were examined as mediators
of the relationship between fear arousing content and attitude change. The results showed
that the structure of the models varied as a firnction of message topic. Both perceived
argument quality and perceived fear mediated the efi‘ect of fear arousing content on
attitude change for the alcohol and tetanus messages. This model was labeled the

Perceived Argument Quality and Perceived Fear Independence Mode]. Perceived

59

60
argument quality mediated the effect of fear arousing content on attitude change for the

bicycle helmet message. This model was called the Perceived Argument Quality Model.

Theoretical Implications

 

The Perceived Argument Quality and Perceived Fear Independence Model posits
that fear arousing content produces greater perceived fear and perceived argument quality
which, in turn, increase attitude change. Put another way, the Independence Model posits
that m perceived fear and perceived argument quality mediate the relationship between
fear arousing content and attitude change. The Perceived Argument Quality Mode] posits
that fear arousing content produces greater perceived argument quality which, in turn,
increases attitude change. In other words, this mode] posits that the relationship between
fear arousing content and attitude change is mediated S_(Lely by perceived argument
quality. The Perceived Argument Quality Mode] represents a cognitive route to attitude
change. The Perceived Argument Quality and Perceived Fear Independence Model
represents a cognitive and emotional route to attitude change.

The emotional route. The fear arousing content, perceived fear, and attitude
change causal chain illustrates the emotional route to attitude change. As other scholars
have argued (Burgoon, in press), the causal relationship between fear arousing content,
perceived fear, and attitude change can be explained most parsirnoniously by employing
drive-reduction theory (Higbee, 1970; Baron & Bryne, 1977). According to this model,
fear arousing content increases negative arousal (perceived fear). This arousal decreases
when an effective way of reducing it is presented (message acceptance). Thus, arousal

reduction is reinforcing, and likely to increase attitude change (Baron & Bryne, 197 7).

 

 

61
In this study fear arousing content had a substantial effect on perceived fear for all

message topics. The subsequent effect of perceived fear on attitude change depended on
the message topic, however. Recall that for the bicycle helmet message, perceived fear
did not mediate the relationship between fear arousing content and attitude change. When
perceived fear did impact attitude change, the efi‘ect was positive. In other words, there
was no evidence of a defensive avoidance efi‘ect. This finding is consistent with several
theories which predict that fear and persuasion are related positively. These data are not
consistent with the presence of a fear control process. There is no evidence that fear
arousing content induced fear control processes as described by Leventhal (1970), or
Witte (1992, 1994). Contrary to recent speculation by Dillard (1994), these data d_o
suggest that cognition may be more important than emotion in explaining the relationship
between fear arousing content and attitude change because most of the models had a
cognitive component. This cognitive route to attitude change posits that highly fearful
messages are perceived as more compelling than messages low in fear, and in tum,
increase attitude change.

The cogm'tive route. Boster and Mongeau (1984) argue that the relationship
between fear arousing content and attitude change may be confounded by message
discrepancy. These data are not consistent with the discrepancy mediation hypothesis, but
they are, however, consistent with the hypothesis that the fear arousing content-attitude
change relationship is confounded. Specifically, these data indicate that highly fearful
messages induce attitude change because they are seen as more compelling than those

messages low in fear. Put another way, these data Show that perceived argument quality

62
mediates the relationship between fear arousing content and attitude change. This causal

chain suggests a cognitive or information-processing route to persuasion.

The causal relationship between fear arousing content, perceived argument quality,
and attitude change can be explained by the efi‘ect of intense language on attitudes.
Burgoon (1989) argues that fear arousing persuasive messages may be conceptualized as a
special case of intense language. Researchers interested in the effect of intense messages
on attitude change argue that language high in intensity is vivid and specific (Hamilton, et
al., 1990; Hamilton & Stewart, 1993). These characteristics make intense messages
appear more informative than messages low in intensity. In other words, vividness and
specificity should clarify the position advocated in a message, and lead to greater
argument quality. The clarity associated with intense messages should then lead to greater
retention of the new information and attitude change. Consistent with this reasoning,
Hamilton and Stewart (1993) Show that message intensity increases attitude change by
increasing argument quality (p. 244).

In this study fear arousing content had a substantial efl‘ect on perceived argument
quality for all message topics. The subsequent effect of perceived argument quality on
attitude change was substantial for the alcohol and bicycle hehnet messages. The effect of
perceived argument quality on attitude change for the tetanus message was trivial because
of the relatively high pretest attitude judgments reported by subjects. In other words,
subjects could not adopt more positive attitudes because their initial attitudes were

substantially positive. For fear arousing message topics where initial attitudes are not

63
substantially positive, one would expect perceived argument quality to impact attitude

change substantially.

The perceived argument quality effect on attitude change has important
implications for the literature on fear arousing persuasive messages. Two of the messages
(alcohol and tetanus) in this study were used in published research. Thus, it is likely that
the perceived argument quality confound is present in other comparable messages. For this
reason, it is important to consider the potential confounding effect of perceived argument
quality when interpreting the results of previous literature. It is unlikely, however, that all
low fear messages are less compelling than high fear messages. In other words, it is
possible that other low fear messages may present more compelling arguments than the
ones used in this study. Nevertheless, the perceived argument quality confound may be
eliminated in future studies by creating low and high fear messages that are equally
compelling. Furthermore, perceived argument quality should be measured to index any
potential confounding efi’ects. In this way, one can benefit from using entire samples
instead of deleting large portions that do not meet certain criteria.

Methodological Implications

Some of the confusion in the literature on fear arousing persuasive messages may
be attributed to the use of posttest only designs. It is clear from the results of the study
that posttest scores are misleading because subjects respond difi‘erently on the pretest even
though they are assigned randomly to conditions, or because they are already at the upper
bound of the scale on the pretest. In sum, random assignment does not guarantee that the

comparability of subjects. For this reason it is recommended that subsequent studies

64

compute and examine change scores to determine if subjects change their attitudes
substantially or not at all. Of course, one problem is that change scores tend to be less
reliable than posttest scores. This fact implies the importance of careful pretesting to
obtain change scores with satisfactory reliability.

One common threat to the validity of pretest-posttest designs is that the pretest
may sensitize subjects to the nature of the experiment. Because the posttest scores of
subjects receiving and not receiving a pretest differed little, there is no evidence that this
factor threatened the validity of the data presented in this study.

Summary

The purpose of this investigation was to explore potential confounds in fear
arousing persuasive messages. It was argued that these confounds moderate or mediate
the relationship between fear arousing messages and attitude change. A 3 X 3 X 2
Solomon four-group design was employed. There were three message topics (bicycle
helmets, tetanus, and alcohol consumption), three fear conditions (high, low, and control),
and the presence or absence of a pretest. Subjects were assigned to conditions randomly
with the constraint that comparable numbers of subjects were in each cell. A series of
causal models were devised and tested. Path analyses indicated that perceived fear and
perceived argument quality mediated the relationship between fear arousing content and
attitude change. The exact structure of the path models varied as a firnction of message

topic. The implications of the findings and directions for future research were discussed.

CORRELATIONS AMONG THE VARIABLES FOR THE ALCOHOL

FEAR
PFEAR
DISCR
PSYDIS
IMPPOST
IMPCHN
OUTPOST
OUTCHN
VALPOST
VALCHN
ARGUQ
EXPERT
TRUST
REFF
SEFF
SEVERE
SUSCEP
ATTPOST
ATTCHN

FEAR
PFEAR
DISCR
PSYDIS
IMPPOST
IMPCHN
OUTPOST
OUTCHN
VALPOST
VALCHN
ARGUQ
EXPERT
TRUST
REFF
SEFF
SEVERE
SUSCEP
ATTPOST
ATTCHN

FEAR

1.0000
.4867
-.O725
-.1005
.1955
.3511
-.0208
.0785
.2090
.3418
.2803
.1719
.0145
.0250
.0829
.4411
.0962
-.1083
.2452

OUTPOST

.0208
.2140
-.2250

.1400
—.0408
-.0209
1.0000

.5546
-.0297
.2050
.0570
.0384
.1292
.0573
.3019
.1867
.3761
.1318
.1566

APPENDIX A

CONSUMPTION MESSAGE
PFEAR DISCR PSYDIS
.4867 -.0725 -.1005

1.0000 .0382 -.1164
.0382 1.0000 -.1122
-.1164 -.1122 1.0000
-.0461 .1626 -.2232
.2903 -.0635 -.0859
.2140 -.2250 .1400
.0339 -.2488 .1283
.0623 .2318 -.1935
.3608 -.0765 -.1340
.1212 -.0271 -.2298
-.1519 .0464 -.1525
-.1176 -.0005 -.1715
-.1624 .0743 .0522
-.2048 .2409 -.2647
.1500 -.0552 -.0177
.3386 .0127 -.0949
-.2292 .1356 -.0904
.2999 -.2253 -.1389
OUTCHN VALPOST VALCHN
.0785 .2090 .3418
.0339 .0623 .3608
-.2488 .2318 -.0765
.1283 -.1935 -.1340
.0163 .2470 -.0368
.0285 -.2283 .2422
.5546 -.0297 .2050
1.0000 .1820 .2494
.1820 1.0000 .2184
.2494 .2184 1.0000
.0753 .4002 .3564
.1000 .2855 .1110
.1992 .1331 .1291
.2023 .2394 -.0089
.0282 .5962 .0242
.1227 .1980 .1516
-.0405 .0929 .2349
.3209 .5189 .0098
.3305 .1379 .4817

65

IMPPOST

.1955
-.0461
.1626
-.2232
1.0000
.2898
-.0408
.0163
.2470
-.0368
.1649
.0304
.0923
-.0465
.4544
.1364
-.2085
.1805
-.0596

ARGUQ

.2803
.1212
.0271
.2298
.1649
.0413
.0570
.0753
.4002
.3564
1.0000
.5676
.4289
.3465
.2770
.3417
.0399
.3342
.2356

IMPCHN

.3511
.2903
-.0635
-.0859
.2898
1.0000
-.0209
.0285
-.2283
.2422
-.0413
-.1940
-.2177
-.0600
-.0547
-.0185
-.0247
-.2185
.1238

EXPERT

.1719
.1519
.0464
.1525
.0304
.1940
.0384
.1000
.2855
.1110
.5676
1.0000
.7388
.4486
.2911
.3762
.1410
.3585
.2293

66

TRUST REFF SEFF SEVERE SUSCEP ATTPOST
FEAR .0145 .0250 .0829 .4411 .0962 -.1083
PFEAR -.1176 -.1624 -.2048 .1500 .3386 -.2292
DISCR -.0005 .0743 .2409 -.0552 .0127 .1356
PSYDIS -.1715 .0522 -.2647 -.0177 -.0949 -.0904
IMPPOST .0923 -.0465 .4544 .1364 -.2085 .1805
IMPCHN —.2177 -.0600 -.0547 -.0185 -.0247 -.2185
OUTPOST .1292 -.0573 -.3019 .1867 .3761 .1318
OUTCHN .1992 .2023 .0282 .1227 -.0405 .3209
VALPOST .1331 .2394 .5962 .1980 .0929 .5189
VALCHN .1291 -.0089 .0242 .1516 .2349 .0098
ARGUQ .4289 .3465 .2770 .3417 -.0399 .3342
EXPERT .7388 .4486 .2911 .3762 -.1410 .3585
TRUST 1.0000 .3241 .2015 .2194 -.0599 .2797
REFF .3241 1.0000 .2603 .2241 -.2332 .4536
SEFF .2015 .2603 1.0000 .1457 -.3302 .3665
SEVERE .2194 .2241 .1457 1.0000 .0030 .0749
SUSCEP -.0599 -.2332 -.3302 .0030 1.0000 -.1954
ATTPOST .2797 .4536 .3665 .0749 -.l954 1.0000
ATTCHN .1492 .1689 .0448 .2250 .0416 .2897
ATTCHN
FEAR .2452
PFEAR .2999
DISCR -.2253
PSYDIS -.1389
IMPPOST -.0596
IMPCHN .1238
OUTPOST .1566
OUTCHN .3305
VALPOST .1379
VALCHN .4817
ARGUQ .2356
EXPERT .2293
TRUST .1492
REFF .1689
SEFF .0448
SEVERE .2250
SUSCEP .0416
ATTPOST .2897
ATTCHN 1.0000

N of cases: 55

APPENDIX B

CORRELATIONS AMONG THE VARIABLES FOR THE BICYCLE HELMET

MESSAGE

FEAR PFEAR DISCR PSYDIS IMPPOST IMPCHN
FEAR 1.0000 .6489 -.0180 -.0870 .1327 .3709
PFEAR .6489 .0000 -.0968 -.3920 .2652 .2312
DISCR -.0180 .0968 1.0000 .3845 -.1216 .0849
PSYDIS -.0870 .3920 .3845 1.0000 -.2731 .1099
IMPPOST .1327 .2652 -.1216 -.2731 1.0000 .4808
IMPCHN .3709 .2312 .0849 -.1099 .4808 .0000
OUTPOST .1553 .2949 -.3004 -.2408 .4111 .2837
OUTCHN .1605 .0887 .0989 .0915 .2127 .3028
VALPOST .3069 .5811 -.0549 -.3219 .3675 .2386
VALCHN .3551 .4606 .0547 -.1252 .2072 .3195
ARGUQ .4547 .5668 -.1723 -.3370 .3619 .1188
EXPERT .1809 .2719 -.2293 -.2932 .0933 .0897
TRUST -.0027 .3171 -.2514 -.4610 .1432 .0461
REFF .3370 .3509 -.1865 -.2955 .0629 .0726
SEFF .5750 .4930 -.2935 -.3350 .2304 .2259
SEVERE .5709 .5546 -.1961 -.4686 .1974 .2928
SUSCEP .0366 .1079 -.3709 -.1779 .0050 .1066
ATTPOST .3901 .5159 -.2891 -.4746 .2532 .1792
ATTCHN .2178 .1104 .2654 -.1220 .1264 .1998

OUTPOST OUTCHN VALPOST VALCHN ARGUQ EXPERT
FEAR .1553 .1605 .3069 .3551 .4547 .1809
PFEAR .2949 .0887 .5811 .4606 .5668 .2719
DISCR -.3004 .0989 -.0549 .0547 -.1723 .2293
PSYDIS -.2408 .0915 -.3219 -.1252 ”.3370 .2932
IMPPOST .4111 .2127 .3675 .2072 .3619 .0933
IMPCHN .2837 .3028 .2386 .3195 .1188 .0897
OUTPOST 1.0000 .3981 .4519 .1709 .3760 .2852
OUTCHN .3981 .0000 .1388 .3620 .2796 .2082
VALPOST .4519 .1388 1.0000 .7270 .5439 .2518
VALCHN .1709 .3620 .7270 1.0000 .4809 .2729
ARGUQ .3760 .2796 .5439 .4809 1.0000 .6161
EXPERT .2852 .2082 .2518 .2729 .6161 .0000
TRUST .2606 .0693 .3004 .2183 .5271 .6656
REFF .1266 .2655 .1233 .2846 .5742 .4307
SEFF .3874 .3923 .3569 .4018 .6434 .3694
SEVERE .2571 .1160 .2525 .2776 .5070 .3561
SUSCEP .3893 .1945 -.0450 -.2534 .0481 .0673
ATTPOST .4324 .2723 .2984 .2646 .6092 .4217
ATTCHN -.0031 .4043 .0926 .2953 .2339 .1844

67

68

TRUST REFF SEFF SEVERE SUSCEP ATTPOST
FEAR —.0027 .3370 .5750 .5709 .0366 .3901
PFEAR .3171 .3509 .4930 .5546 .1079 .5159
DISCR -.2514 -.1865 -.2935 -.1961 -.3709 -.2891
PSYDIS -.4610 -.2955 -.3350 -.4686 -.1779 -.4746
IMPPOST .1432 .0629 .2304 .1974 .0050 .2532
IMPCHN -.O461 .0726 .2259 .2928 -.1066 .1792
OUTPOST .2606 .1266 .3874 .2571 .3893 .4324
OUTCHN .0693 .2655 .3923 .1160 -.1945 .2723
VALPOST .3004 .1233 .3569 .2525 -.O450 .2984
VALCHN .2183 .2846 .4018 .2776 -.2534 .2646
ARGUQ .5271 .5742 .6434 .5070 .0481 .6092
EXPERT .6656 .4307 .3694 .3561 -.0673 .4217
TRUST 1.0000 .3951 .2663 .3084 -.0112 .4343
REFF .3951 1.0000 .6686 .5191 -.0881 .6769
SEFF .2663 .6686 1.0000 .5901 .0402 .7011
SEVERE .3084 .5191 .5901 1.0000 .1918 .6085
SUSCEP -.0112 -.0881 .0402 .1918 1.0000 .0553
ATTPOST .4343 .6769 .7011 .6085 .0553 1.0000
ATTCHN .1420 .2405 .2808 .1498 -.3783 .3718
ATTCHN
FEAR .2178
PFEAR .1104
DISCR .2654
PSYDIS -.1220
IMPPOST .1264
IMPCHN .1998
OUTPOST -.OO31
OUTCHN .4043
VALPOST .0926
VALCHN .2953
ARGUQ .2339
EXPERT .1844
TRUST .1420
REFF .2405
SEFF .2808
SEVERE .1498
SUSCEP -.3783
ATTPOST .3718
ATTCHN 1.0000

N of cases: 62

APPENDIX C

CORRELATIONS AMONG THE VARIABLES FOR THE TETANU S MESSAGE

FEAR
PFEAR
DISCR
PSYDIS
IMPPOST
IMPCHN
OUTPOST
OUTCHN
VALPOST
VALCHN
ARGUQ
EXPERT
TRUST
REFF
SEFF
SEVERE
SUSCEP
ATTPOST
ATTCHN

FEAR
PFEAR
DISCR
PSYDIS
IMPPOST
IMPCHN
OUTPOST
OUTCHN
VALPOST
VALCHN
ARGUQ
EXPERT
TRUST
REFF
SEFF
SEVERE
SUSCEP
ATTPOST
ATTCHN

FEAR

1.0000

.3378
-.0635
-.0642
-.0514
-.1216
.0681
.2706
.2497
.2670
.2557
.0381
.0417
.0628
.1877
.4019
.1347
.0957
.1050

OUTPOST

.0681
.1451

-.2677
-.2768

.1111
.1116
1.0000
.2479
.5817
.2945
.0983
.0349
.0318
.2707
.4640
.1322
.2649
.4075
-.2603

PFEAR

.3378
1.0000
-.0216
-.1603

.2421

.1926

.1451

.1390

.2937

.2861

.1699

.1184

.1589

.1547

.0719

.2111

.2983

.0641
-.1022

OUTCHN

.2706
.1390
.1631
.0620
.0035
.2464
.2479
.0000
.2094
.6670
.1441
.1726
.1444
.0045
.1180
.2451
.4318
.1076
.3806

DISCR

-.0635
-.0216
1.0000
.0427
.0403
.1613
-.2677
.1631
.2660
.2003
.1999
.1269
.1915
.1825
.3723
.0648
.1522
.2463
.2700

VALPOST

.2497
.2937
.2660
.2350
.4427
.1317
.5817
.2094
.0000
.5047
.3144
.0324
.1069
.2748
.5162
.2230
.1971
.3993
.1586

69

PSYDIS

-.0642
-.1603

.0427
1.0000
.1441
.2118
.2768
.0620
.2350
.0385
-.3519
-.l407
-.2935
-.2002
-.O950
-.2756
-.O938
-.3525

.1818

VALCHN

.2670
.2861
.2003
.0385
.2728
.2873
.2945
.6670
.5047
1.0000
.1758
.0470
.1111
.1447
.1448
.2086
.3319
-.0075
.2955

IMPPOST

-.0514
.2421
-.0403
.1441
1.0000
.6178
.1111
.0035
.4427
.2728
.0799
-.1100
-.0962
.0249
.0535
-.1856
.0524
.0149
-.0063

ARGUQ

.2557
.1699
.1999
-.3519
.0799
.0386
.0983
.1441
.3144
.1758
1.0000
.4915
.5020
.1876
.2171
.2775
.1707
.4802
-.1105

IMPCHN

-.1216
.1926
.1613
.2118
.6178

1.0000
.1116
.2464
.1317
.2873
.0386

-.0527

-.0218
.1291
.0761

—.0564
.0494

-.1023
.0780

EXPERT

.0381
.1184
.1269
.1407
.1100
.0527
.0349
.1726
.0324
.0470
.4915
.0000
.6583
.1546
.0603
.1187
.1337
.1952
.0324

TRUST
FEAR -.O4l7
PFEAR .1589
DISCR .1915
PSYDIS -.2935
IMPPOST -.0962
IMPCHN -.0218
OUTPOST .0318
OUTCHN .1444
VALPOST -.1069
VALCHN .llll
ARGUQ .5020
EXPERT .6583
TRUST 1.0000
REFF .1468
SEFF -.Ol77
SEVERE .1320
SUSCEP .2407
ATTPOST .2353
ATTCHN .0129

ATTCHN

FEAR -.1050
PFEAR -.1022
DISCR .2700
PSYDIS .1818
IMPPOST -.0063
IMPCHN .0780
OUTPOST -.2603
OUTCHN .3806
VALPOST -.1586
VALCHN .2955
ARGUQ -.1105
EXPERT .0324
TRUST .0129
REFF -.1575
SEFF -.2015
SEVERE -.1248
SUSCEP -.O365
ATTPOST -.3122
ATTCHN 1.0000

N of cases:

57

REFF

.0628
.1547
-.1825
-.2002
.0249
.1291
.2707
.0045
.2748
.1447
.1876
.1546
.1468
1.0000
.4339
.5658
.0410
.4267
-.1575

70

SEFF

.1877
.0719
-.3723
-.0950
.0535
.0761
.4640
.1180
.5162
.1448
.2171
.0603
-.0177
.4339
1.0000
.2486
.0840
.5183
-.2015

SEVERE

.4019
.2111
.0648
.2756
.1856
.0564
.1322
.2451
.2230
.2086
.2775
.1187
.1320
.5658
.2486
1.0000

.1449

.2438
-.1248

SUSCEP

.1347
.2983
.1522
-.0938
.0524
.0494
.2649
.4318
.1971
.3319
.1707
.1337
.2407
.0410
.0840
.1449
1.0000
.0502
-.O365

ATTPOST

.0957

.0641
-.2463
-.3525
.0149
.1023
.4075
-.1076

.3993
-.0075
.4802
.1952
.2353
.4267
.5183
.2438
.0502
.0000
.3122

H

LIST OF REFERENCES

Anderson, N. H. (1959). Test of a model for opinion change. Jourml of Abnormal and
Social Pachology, _5_9_, 371-381.

Anderson, N. H., & Hovland, C. (1957). The representation of order effects in
communication research. In C. Hovland (Ed), The order of presentation in persuasion
(pp. 158-169). New Haven: Yale University Press.

Baron, R., & Bryne, D. (1977). Social pgchology. Boston, MA: Allyn & Bacon.

Beck, K. H., & Frankel, A. (1981). A conceptualization of threat communications and
protective health behavior. Spcia] Paychology Quarterly, 1, 204-217.

Brickner, M., Harkins, S., & Ostronr, T. (1986). Personal involvement: Thought
provoking implications for social loafing. Jouma] of Personality and Social
Pachology, M, 763-769.

Boster, F. J. (1993). A recurs_ive mode] in which all variables m measured: The simple
causal string. Unpublished manuscript, Michigan State University, Department of
Communication, East Lansing.

Boster, F. J ., & Mongeau, P. (1984). Fear-arousing persuasive messages. In R. N.
Bostrum (Ed), Communication Yearbook. (Volume 8, pp. 330-375). Beverly Hills:

Sage.

Burgoon, M. (1989). Messages and persuasive efi‘ects. In J. J. Bradac (Ed), Message
effecta in communication science. (Vol. 17, pp. 129-164). Newbury Park, CA: Sage.

Burgoon, M. (in press). Language and social influence. In H. Giles & P. Robinson
(Eds), Handbook of langaage and social psychology. London: John Wiley.

Campbell, D. T. & Stanley, J. C. (1963). Experimental and quasi-experimental designs for
research. Chicago: Rand McNally.

Chaiken, S. (1980). Heuristic versus systematic information processing and the use of

source versus message cues in persuasion. Jourrgl of Personalig and Social
Psychology, M, 752-766.

71

72

Chaiken, S. (1987). The heuristic model of persuasion. In M. P. Zanna, J. M. Olson, &

C. P. Herman (Eds), Social influence: The Ontario Smposium (Vol. 5, pp. 3-39).
Hillsdale, NJ: Erlbaum.

Chu, G. C. (1966). Fear arousal, efiicacy and imnrinency. Journal of Personalig and
Social Psychology, é, 517-524.

Dillard, J. P. (1994). Rethinking the study of fear appeals: An emotional perspective.
Commugnication Theory. 4, 295-323.

Eagly, A. H., Wood, W., & Chaiken, S. (1978). Causal inferences about communicators
and their effect on opinion change. 101ml of Personality and Social ngchology, M,
424-435.

Fink, E. L., Kaplowitz, S. A., & Bauer, C. L. (1983). Positional discrepancy,
psychological discrepancy, and attitude change: Experimental tests of some
mathematical models. Communication Monoggaphs, M, 413-430.

Hamilton, M. A., Hunter, J ., & Burgoon, M. (1990). An empirical investigation of an
axiomatic model of the efi‘ect of language intensity on attitude change. Journal of
Langaage and Social Pychology, 2, 23 5-255.

Hamilton, M. A., & Stewart, B. L. (1993). Extending an information processing model of
language intensity efi‘ects. Communication Qaarterly, 4_1_, 231-246.

Heesacker, M., Petty, R E., & Cacioppo, J. T. (1983). Field dependence and attitude
change: Source credibility can alter persuasion by affecting message-relevant thinking.
Journal of Personality, i1, 653-666.

Higbee, K. L. (1970). Fifteen years of fear arousal: Research on threat appeals: 1953-
1968. Paychologica] Bulletin, 2, 426-444.

Hovland, C., Janis, 1., & Kelly, H. (1953). Communication and persuasion. New Haven,
CT: Yale University Press.

Hunter, 1. E., Danes, J. E., & Cohen, S. H (1984). Mathematical models of attitude
change: Change in single attitudes and coggu'tive structure. New York: Harcourt

Brace Jovanovich.

 

Hunter, J. E., & Gerbing, D. W. (1982). Unidirnensiona] measurement, second order
factor analysis and causal models. In B. M. Straw and L. L. Cummings (Eds),
Research in Organizational Behavior (Vol. 4), Greenwich, CT: JAI Press.

Janis, I. L., & Feshback, S. (1953). Effects of fear-arousing communication. The Journal
Abnorrna] and SocErl Psychology, M, 78-92.

73

Johnson, B. T., & Eagly, A. H. (1989). Efi‘ects of involvement on persuasion: A meta-
analysis. Psychological Bulletin. 106. 290-314.

Johnson, H. H., & Scileppi, J. A. (1969). Effects of ego-involvement conditions on
attitude change to high and low credibility communicators. Journal of Personalig and
Social Echology, 1;, 31-36.

Kaplowitz, S. A., Fink, E. L., Mulcrone, J., Atkin, D., & Dabil, S. (1991). Disentangling
the effects of discrepant and disconfirrning information. Social Paychology Mmly,
S_4, 191-207.

Kenny, D. (1979). Correlation and causalig. New York: Wiley.

Kiesler, S. B., & Mathog, R. B. (1968). Distraction hypothesis in attitude change:
Effects of effectiveness. Pachological Reports, 2;, 1123-1133.

Kleinot, M. C. (1980). Efl‘ects of threat noxiousness, vulnerability, and recommendation
eflicapy on intentions to consume alcohol. Doctoral dissertation, University of

Alabama.

Kleinot, M. C., & Rogers, R. W. (1982). Identifying effective components of alcohol
misuse prevention programs. Journal of Studies on Alcohol. M, 802-811.

Leventhal, H. (1970). Findings and theory in the study of fear communications. In L.

Berkowitz (Ed) Advances in experimental social paychology (V o]. 5, pp. 119-186)
New York: Acadenric Press.

Leventhal, H. (1971). Fear appeals and persuasion: The difi‘erentiation of a motivational
construct. American Journal of Public Health, Q1, 1205-1224.

Leventhal, H., Jones, S., & Trembly, G. (1966). Sex differences in attitude and behavior
change under conditions of fear and specific instructions. Journal of Eaperimental
Social Psychology, 2 387-399.

_’

McGuire, W. J. (1969). The nature of attitudes and attitude change. In G. Lindzey & E.
Aronson (Eds), The handboak of social psychology (Vol. 3., 2nd ed). Reading, MA:
Addison-Wesley.

McPhee, R., & Babrow, A. (1987). Causal modeling in communication research: Use,
disuse, and misuse. Communication Monogtaphs, 54, 344-366.

Miller, G. R. (1987). Persuasion. In C. R. Berger & S. H. Chafi‘ee (Eds), Handbook of
commanication science (pp. 446-483). Newbury Park, CA: Sage.

74

Mongeau, P. A (in press). Another look at fear-arousing persuasive appeals. In M. Allen
& R. W. Preiss (Eds), Persuasam: Advances through meta-analysis.

O’Keefe, D. J. (1990). Persuasion: Theory aad reseaLh. Newbury Park, CA: Sage.

Pedhazur, E. (1982). Multiple regtession in belraviora] reseaMh (2nd ed). New York:
Holt, Rinehart, & Wilston.

Petty, R. E., Cacioppo, J. T. (1979). Issue involvement can increase or decrease
persuasion by enhancing message-relevant cognitive responses. Journal of Personality

and Social Pachology, fl, 1915-1926.

Petty, R. E., Cacioppo, J. T. (1984). The effects of involvement on responses to
argument quantity and quality: Central and peripheral routes to persuasion. Mat
Personalig and S009] Psycholggy, M, 69-81.

Petty, R. E., Cacioppo, J. T. (1986a). Communication and persuasion: Central and
peripheral routes to attitude change. New York: Springer-Verlag.

Petty, R. E., Cacioppo, J. T. (1986b). The elaboration likelihood model of persuasion. In

L. Berkowitz (Ed), Advances in experimental social psychology (Vol. 19, pp. 123-
204). San Diego, CA: Academic Press.

Petty, R. E., Cacioppo, J. T., & Goldman, R. (1981). Personal involvement as a
determinant of argument-based persuasion. Joumal of Personality and Social
Psychology, 4_1_, 847-855.

Petty, R. E., Cacioppo, J. T., & Schumann, D. (1983). Central and peripheral routes to
advertising effectiveness: The moderating role of involvement. Journal of Consumer
Research, 1_Q, 135-146.

Ratneshwar, S., & Chaiken, S. (1986). When is the expert source more persuasive? A
heuristic processing analysis. In T. A. Shimp, S. Shanna, G. John, J. A. Quelch, J. H.
Lindgren, Jr., W. Dillon, M. P. Gardner, & R. F. Dyer (Eds), 1986 AMA educator’s
proceedings (p. 86). Chicago: American Marketing Association.

Rhine, R. J., & Severance, L. J. (1970). Ego-involvement, discrepancy, source credibility,
and attitude change. Journal of Personality and Social Psychology, _1_6_, 175-190.

Rogers, R. W. (1975). A protection motivation theory of fear appeals and attitude
change. Journal of Psychology, fl, 93-114.

Rogers, R. W. (1983). Cognitive and psychological processes in fear appeals and attitude
change: A revised theory of protection motivation. In J. Cacioppo & R. Petty (Eds),
Social psychophysiology (pp. 153-176). New York: Guilford.

75

Rogers, R. W. & Mewbom (1976). F ear appeals and attitude change: Efi‘ects of a
threat’s noxiousness, probability of occurrence, and the efiicacy of the coping
response. Journal of Personaligz and Social Psycholagy, 35, 54-61.

Smith, M. J. (1988). Contemporm communication research mathoda. Belmont, CA:
Wadsworth.

Sutton, S. R. (1982). Fear-arousing communications: A critical examination of theory

and research. In J. R. Eisner (Ed), Social pfichology and behavioral medicine (pp.
303-337). New York: John Wiley.

Witte, K. (1994). Fear control and danger control: A test of the extended parallel process
model (EPPM). Communication Monogtaphs, _6_1, 113-134.

Witte, K. (1993). Message and conceptual confounds in fear appeals: The role of threat,
fear, and emcacy. Southern Communication Journal, 5_8, 147-155.

Witte, K. (1992). Putting the fear back into fear appeals: The extended parallel process
model. Communication Monogtaphs, 3, 329-349.

Witte, K. (1991). The biaycle helmet project. Unpublished manuscript. Texas A&M
University.