On the Nature of Reactance
and its Role in Persuasive
James Price Dillard & Lijiang Shen
Reactance theory might be proﬁtably applied to understanding failures in persuasive
health communication but for one drawback: The developer of the theory contends that
reactance cannot be measured. Rejecting this position, this paper develops four alternative
conceptual perspectives on the nature of reactance (i.e., combinations of cognition and
affect), then provides an empirical test of each. Two parallel studies were conducted, one
advocating ﬂossing ðN ¼ 196Þ; the other urging students to limit their alcohol intake
ðN ¼ 200Þ: In both cases, a composite index of anger and negative cognitions fully
mediated the effects of threat-to-freedom and trait reactance on attitude and intention.
The data showed that, in fact, reactance can be operationalized as a composite of self-report
indices of anger and negative cognitions. The implications for persuasive communication,
in general, are considered as well the speciﬁc ﬁndings for ﬂossing and drinking.
Keywords: Reactance; Anger; Threat to Freedom; Persuasion
Persuasive attempts of all sorts, including public health campaigns, often fail to
produce the desired effect (Foxcraft, Lister-Sharp, & Lowe, 1997; Wallack, 1981; Wilde,
1993). In fact, in some cases, they produce results directly at odds with their intent
(Guttman, Kegler, & McLeroy, 1996; Hornik, 2002; Stewart & Martin, 1994).
The theory of psychological reactance (Brehm, 1966; Brehm & Brehm, 1981;
Wicklund, 1974) provides one theoretical perspective through which these suasory
ISSN 0363-7751 (print)/ISSN 1479-5787 (online) q 2005 National Communication Association
James Dillard, Department of Communication Arts & Sciences, Pennsylvania State University, and Lijiang Shen,
Center for Communication Research, University of Wisconsin-Madison, Madison, Wisconsin, USA. We thank
Mike Stephenson for his reactions to an earlier version of this manuscript. Correspondence to: James Dillard,
Department of Communication Arts & Sciences, Sparks Building, Pennsylvania State University, University Park,
PA 16802, USA. Email: email@example.com
Vol. 72, No. 2, June 2005, pp. 144–168
miscarriages might be understood. The theory contends that any persuasive message
may arouse a motivation to reject the advocacy. That motivation is called reactance.
From its inception to the present, the theory has been called upon to explain resistance
to persuasion (Buller, Borland, & Burgoon, 1998; Burgoon, Alvaro, Grandpre, &
Voulodakis, 2002; Ringold, 2002).
The primary limiting factor in the application of reactance theory to persuasive
campaigns is the ephemeral nature of its central, explanatory construct. The creator of
the theory contends that reactance cannot be measured (Brehm, 1966; Brehm &
Brehm, 1981). Although that claim may have been accurate during the infancy of
reactance theory, advances in the study of persuasion since that time suggest that it
should be reconsidered.
The rationale for this project is presented in four parts. First, we provide a brief
overview of the theory of psychological reactance. Second, we examine the notion of
reactance itself in more detail. In part, this involves reconceptualizing reactance along
more contemporary lines. Third, we consider a subset of the many variables that are
likely to lead to reactance arousal and propose a corresponding series of hypotheses.
The general supposition is that the effect of these variables on attitude is mediated by
one of the alternative conceptualizations of reactance. Fourth, two studies, designed to
test that supposition, form the empirical core of this paper. These sections each speak
to the larger themes that motivated the research. That is, what is the nature of
reactance and what role does it play in persuasive health communication?
The Theory of Psychological Reactance
There are four essential elements to reactance theory: Freedom, threat to freedom,
reactance, and restoration of freedom. The notion of freedom is not freedom in
general terms; it is “not abstract considerations, but concrete behavioral realities”
(Brehm & Brehm, 1981, p. 12). However, the concept of free behaviors is deﬁned
broadly so as to include actions, as well as emotions and attitudes (Brehm, 1966;
Wicklund, 1974). Individuals possess freedoms only to the extent that they have
knowledge of them and perceive that they are capable of enacting the behavior.
Given that an individual perceives a speciﬁc freedom, any force on the individual
that makes it more difﬁcult for him or her to exercise that freedom constitutes a threat
(Brehm, 1966; Brehm & Brehm, 1981). Even an impersonal event, such as the weather,
can be viewed as a threat, if it renders more difﬁcult the exercise of a freedom.
However, social inﬂuence as a threat is most pertinent to questions of persuasive
health communication. In fact, one of the basic claims of the theory is that high-
pressure communicators are likely to be seen as threats to freedom (Brehm & Brehm,
1981; Wicklund, 1974).
Psychological reactance is “the motivational state that is hypothesized to occur
when a freedom is eliminated or threatened with elimination” (Brehm & Brehm, 1981,
p. 37). The theory contends that when a perceived freedom is eliminated or threatened
with elimination, the individual will be motivated to reestablish that freedom.
Direct restoration of the freedom involves doing the forbidden act. In addition,
freedoms may be restored indirectly by increasing liking for the threatened choice
(Brehm, Stires, Sensenig, & Shaban, 1966; Hammock & Brehm, 1966), derogating the
source of threat (Kohn & Barnes, 1977; Schwarz, Frey, & Kumpf, 1980; Smith, 1977;
Worchel, 1974), denying the existence of the threat (Worchel & Andreoli, 1974;
Worchel, Andreoli, & Archer, 1976), or by exercising a different freedom to gain feeling
of control and choice (Wicklund, 1974). Although all of these means for reducing
reactance have been the focus of research, reduced or boomerang attitude change has
captured the lion’s share of attention.
The Nature of Reactance
To clarify the role that reactance might play in the processing of health messages, it is
useful to examine the construct from both a conceptual and an operational
standpoint. Conceptually speaking, reactance has been deﬁned primarily in terms of
its antecedents and outcomes. For example, as expressed in the quotation above, the
proximal cause of reactance is a perceived threat to freedom: Reactance is “the
motivational state that is hypothesized to occur when a freedom is eliminated or
threatened with elimination” (Brehm & Brehm, 1981, p. 37). Thus, reactance is the
force that prompts certain outcomes, particularly, efforts to re-establish the threatened
freedom by either direct or indirect means. Apart from a brief mention of the
possibility that individuals “may be aware of hostile and aggressive feelings”
(Brehm, 1966, p. 9), if the level of reactance arousal is high, the nature of reactance
itself remains remarkably underdetermined. Brehm’s apparent reluctance to provide
greater conceptual explication of the principal mechanism of the theory may arise
from his belief concerning the potential for measurement of it. According to Brehm
and Brehm (1981), “reactance has the status of an intervening, hypothetical variable
... We cannot measure reactance directly, but hypothesizing its existence allows us to
predict a variety of behavioral effects” (p. 37, our emphasis). If one of the purposes of a
conceptual deﬁnition is to provide guidance on operationalization (Chaffee, 1991),
such guidance becomes superﬂuous when operationalization is viewed as impossible.
Since the time of Brehm’s writings, scholars have applied ideas drawn from
reactance theory to a number of different domains. And, in the course of analyzing
and extending the theory, reactance has been implicitly and explicitly deﬁned in
several different ways. Building on work of earlier authors, we discern four distinct
means of characterizing reactance. In the ﬁrst, reactance is viewed as purely cognitive.
Research in the cognitive response tradition clearly adopts this perspective
(e.g., Petty & Cacioppo, 1986), as does work on the clinical manifestations of
reactance (Kelly & Nauta, 1997). One advantage to conceiving of reactance in this way
is that it immediately becomes measurable through a variety of self-report techniques.
The means that is most obviously relevant to questions of persuasion is the widely
used thought-listing technique (Cacioppo & Petty, 1981). This purely cognitive view
suggests that reactance can be conceived of and operationalized as counter-arguing.
Second, citing similarities between antecedents of reactance and cognitive appraisals
that lead to anger, some writers suggest that reactance might be considered, in whole
146 J. P. Dillard & L. Shen
or in part, as an emotion (Dillard & Meijenders, 2002; Nabi, 2002). Certainly, this
claim aligns well with Brehm’s description of reactance as the experience of hostile and
aggressive feelings (Seltzer, 1983; White & Zimbardo, 1980; Wicklund, 1974). In this
view then, reactance might be considered more or less synonymous with the family of
concepts that index varying degrees of anger (e.g., irritation, annoyance, and rage).
From this perspective, reactance might be operationalized in various ways including
asking individuals to make a judgment on a close-ended scale regarding the degree to
which they are experiencing anger.
A third logical option holds that reactance might be considered as both affect and
cognition. Though unrelated to reactance per se, one example of this type of thinking
can be seen in Leventhal’s (1970) parallel processing model. He posits that individuals
have both cognitive and emotional reactions to persuasive health messages and that
those reactions have unique effects on message acceptance. Evidence consistent with
both points can be found in studies of cognitive and emotional responses to public
service announcements (Dillard & Peck, 2000, 2001; Dillard, Plotnick, Godbold,
Freimuth, & Edgar, 1996; Stephenson, 2003; Witte, 1994).
The ﬁnal possibility also suggests that reactance has both cognitive and affective
components. However, unlike the previous position, which speciﬁed distinct effects, in
this fourth perspective cognition and affect are intertwined. In fact, they are
intertwined to such a degree that their effects on persuasion cannot be disentangled.
Such a view is most compatible with a conception of motivation as an alloy of its
components, rather than a simple sum of distinct elements (as is implied by the
previous position). Considered in unison, these four conceptions of reactance can be
expressed as a pair of research questions:
RQ1: Should reactance be conceptualized and operationalized as cognition, affect, or
RQ2: If reactance has both cognitive and affective components, how are the two
We should note some potential objections to thinking of reactance as either counter-
arguing or as anger. One reading of the theory suggests that reactance cannot be construed
as isomorphic with counter-arguing (or anger) because it is the cause of counter-arguing
(or anger). The theory might be interpreted this way, but to do so immediately returns us
to the circumstance in which reactance cannot be measured
. We see that as undesirable.
Another possible objection concerns the fact that counter-arguing (or anger) might be
prompted by message processing goals (e.g., defensive processing) or aspects of a message
(e.g., weak arguments). As this argument goes, if our proposed mediators are responsive
to other variables than those detailed by Brehm (1966; Brehm & Brehm, 1981), then they
are not unique. And, if they are not unique, they should not be considered reactance.
However, this argument overlooks the fact that all motivations implicate cognition, affect,
or both (e.g., Kuhl, 1986). That is, whether the motivation is one of achievement,
self-actualization, reactance, and so on, it will guide attention, inﬂuence thought
processes (Derryberry & Tucker, 1994), stimulate feelings, and direct behavior (Kuhl,
1986; Lang, 2000) Accordingly, we are inclined to give little weight to this position.
Reactance as a Mediator
Although one might wish for a more detailed explication of reactance itself, the theory is
quite explicit concerning the role of reactance in the persuasion process: Reactance is the state
that mediates the effects of threat to freedom on various outcomes such as attitude and
behavior. This point is of signal importance insofar as it suggests a strategy for addressing the
two research questions posed above. If it were possible to identify variables that were accepted
causes and consequences of reactance, then one could test the various combinations of
cognition and affect for their ability to mediate the antecedent-consequent relationship.
To the extent that mediation was observed, it would be reasonable to claim that one or more
of those combinations were functionally synonymous with reactance. To implement that
strategy we turn next to a brief and highly selective review of the precursors of reactance.
Two Antecedents of Reactance
Strength of the Threat to Freedom
The idea that threats to freedom vary in strength is broad enough to imply that
perceptions of threat are likely to be responsive to multiple message features. Given the
goals of this study, our aim was not to attempt to advance understanding of the impact of
speciﬁc message features on reactance. Rather, we sought (a) to create a strong
manipulation that (b) could be unequivocally judged as having construct validity by
anyone familiar with the theory. A review of the literature suggested that these aims could
be achieved by purposefully confounding intent to persuade and language intensity.
Several studies demonstrate that manipulations of perceived intent to persuade
produces results that are consistent with the theoretical predictions of reactance theory
(Heller, Pallak, & Picek, 1973; Kohn & Barnes, 1977; Worchel & Brehm, 1970). Strong
evidence of the negative impact of intent on persuasion can be found in Benoit’s
(1998) meta-analysis, which reported a homogeneous effect across 12 investigations.
Other research indicates that intense, forceful, or dogmatic language may increase the
magnitude of reactance (Bensley & Wu, 1991; Doob & Zabrack, 1971; Worchel & Brehm,
1970). For example, M. Brehm (reported in Brehm, 1966) conducted a study in which
students at the University of Kansas were exposed to a message that argued for a stronger
program in intercollegiate athletics. Students who received the version of the message that
ended with the sentence “You, as college students, must inevitably draw the same
conclusion” showed less agreement with the message than those for whom the sentence
was omitted. More recent work, conducted by Buller and hiscolleagues (Bulleret al., 1998,
2000), shows similar effects with some qualiﬁcations involving message format and prior
behavioral intentions. Overall, theory and previous ﬁndings led us to predict that:
H1: Strength of the threat to freedom is directly correlated with magnitude of reactance.
Trait Reactance Proneness
Although reactance was initially conceived as situation speciﬁc (Brehm, 1966;
Wicklund, 1974), Brehm and Brehm (1981) recognized that individuals may vary in
148 J. P. Dillard & L. Shen
their trait proneness to reactance arousal. Subsequently, several scholars developed
scales to measure persons’ transituational propensity to experience reactance (Dowd,
Hughs, Brockbank, & Halpain, 1988; Hong, 1992; Hong & Faedda, 1996; Hong & Page,
1989; Merz, 1983). Of the available options, Hong’s scale shows the best conceptual
correspondence with the reactance construct and the most favorable psychometric
properties. In line with the simple notion that persons high in trait reactance should be
more prone to experience reactance in response to a persuasive message, we
H2: Trait reactance proneness is directly correlated with magnitude of reactance.
The Combined Effects of Threat and Proneness
It is difﬁcult to derive an unequivocal hypothesis regarding the combined impact of
threat and trait proneness on the magnitude of reactance. To be sure, main effects-
only ﬁndings would pose no threat to the theory. However, an interaction between
the two antecedents might also be viewed as compatible with the theory assuming
that it was ordinal in form. For example, an interaction in which person high in trait
proneness showed greater reactivity to a strong threat than weak threat seems quite
plausible. Still, it is not clear that the theory demands such a pattern of data. Lacking
guidance from the theory, we asked simply:
RQ3: Is there an interaction between threat and reactance proneness on magnitude
of reactance arousal? And, if so, what is the form of that interaction?
Modeling the Reactance Process
As noted earlier, reactance theory speciﬁes a process in which antecedents of reactance
bring about reactance, which, in turn, prompts efforts to restore the threatened freedom.
Given four possible conceptions of reactance, four corresponding processes are possible.
An illustration of each is given in Figure 1. Because the ﬁrst model assumes that reactance
is a purely cognitive phenomenon, we label it the Single Process Cognitive Model.
For parallel reasons, the second model is termed the Single Process Affective Model. The
third option, which assumes that cognition and affect can be discriminated, is the Dual
Process Model. The Intertwined Process Model suggests that affect and cognition are so
closely interwoven that they are better thought of as indicators of an underlying concept
than as distinct phenomena.
To the process speciﬁed by Brehm (1966), we have added a causal link from attitude
to behavioral intention. This prediction, drawn from the theories of reasoned action
(Fishbein & Ajzen, 1975) and planned behavior (Ajzen, 1991), has been supported in
numerous studies (Kim & Hunter, 1993a, 1993b). The purpose in including it here is
to add a degree of complexity to the models under study. To do so increases the
difﬁculty of ﬁtting the models to the data and correspondingly enhances the degree of
conﬁdence that we might reasonably place in any of the models that successfully
reproduce the data. In addition, it extends the reach of reactance theory from attitude
to behavioral intention.
Two studies were conducted that were conceptual replications of one another. In each
case, respondents read one of two versions of a persuasive message that varied in
strength of threat, then provided data on their cognitive and affective responses, as well
at their attitude, behavioral intention, and trait reactance proneness. The studies
differed only in message topic: ﬂossing versus binge drinking. These two topics were
chosen to capture a variety of differences and thereby increase the generalizability of
the results. Speciﬁcally, one message promoted a private action (i.e., ﬂossing), while
the other advocated reducing a public behavior (i.e., drinking).
All participants were recruited from undergraduate classes in Communication or
Journalism at the University of Wisconsin-Madison. They received a small portion of
A Single Process Cognitive Model
A Single Process Affective Model
A Dual Process Cognitive-affective Model
Cognition Attitude Behavioral
Anger Attitude Behavioral
An Intertwined Process Cognitive-affective Model
Figure 1 Comparison of four models of reactance.
150 J. P. Dillard & L. Shen
extra credit for participating in the study. The 205 subjects in the drinking study were
between ages 18 and 30 ðM ¼ 20:79; SD ¼ 1:46Þ. White/Caucasian subjects accounted
for 90.2% of the sample, Asian descent for 5.9%, Hispanic descent for 2.4%, and other,
including African descent, 1.5%. The sample was 69.3% female and 30.7% male.
The 202 participants in the ﬂossing study were between ages 18 and 32 ðM ¼ 20:64;
SD ¼ 1:43Þ. White/Caucasian subjects accounted for 88.6% of the sample, Asian
descent for 6.9%, Hispanic descent for 3.0%, and other, including African descent,
1.5%. This sample was 72.3% female and 27.7% male. List-wise deletion reduced the
Ns to 196 (ﬂossing) and 200 (drinking).
Both messages followed the standard format for a fear appeal in that they consisted of a
threat-to-health component and an action or recommendation component (Rogers, 1983).
The threat-to-health portion of the messages discussed the negative consequences of not
ﬂossing and binge drinking respectively. A summary of the arguments is given in Table 1.
The strength of threat-to-freedom manipulation appeared in the action component
of each message. In each case, forceful language was used in the strong threat
condition and milder, more polite terms were chosen for the weak threat condition.
Tables 2 and 3 present the exact wording for each of the conditions in each message.
The anti-drinking appeal was a bit longer than the ﬂossing message (1167 words
versus 1025). The appeals also differed in that the ﬂossing message contained a color
picture of affected gums and a paragraph explaining how to ﬂoss just prior to the
action component (i.e., the threat-to-freedom manipulation). In all other respects, the
messages were fairly similar. Both were formatted on multiple pages and presented to
participants in small plastic notebooks.
When the participants arrived at the lab, they were told that they would be evaluating
some informational messages on a health topic that was relevant to college students.
They were assigned to either the high or low threat-to-freedom condition by
intermingling the two versions of the message booklets. Due to the sensitivity of the
binge drinking topic and the fact that most of the participants were underage,
anonymity was assured and achieved. First, subjects signed and dated the consent
forms, which were removed from the questionnaires, and passed to the experimenter
face down. Next, they were instructed to turn to a pre-message questionnaire that
contained a variety of items not relevant to the current report. At this point,
participants read the threat-to-health component of the message and provided data on
their reactions to it. When this was completed, they read the recommendation
component of the message, which contained the threat-to-freedom manipulation, and
provided data on their cognitive and affective responses. Only the cognitive and
affective reactions to the manipulation were used in subsequent analyses. Finally,
participants ﬁlled out a questionnaire that contained outcome measures such as
manipulation check, attitude toward the behavior, and behavioral intention, as well as
Hong’s reactance scale. The whole process took approximately 40 minutes.
Participants were thanked and their questions answered before they left the lab.
For all multi-item measures, principal axis factor analyses followed by varimax rotation were
employed to assess dimensionality. In every case, the scales appeared to be unidimensional.
Table 1 Summary of the Threat-to-health Portions of the Messages
The ﬂossing message The binge drinking message
1. Due to gum disease, the gums and bone
that support the teeth can become painfully
and irreversibly damaged. Left unattended,
teeth will eventually loosen and fall out.
1. Over-consumption is a big problem in 2001.
If you are a student at this university, there is a
two-thirds chance that you are a binge drinker.
2. Up to 30% of the population may be
genetically susceptible to gum disease. If
you are in this group, you are up to six
times more likely to develop some form of
2. There is clear evidence that alcohol
consumption can reduce your academic
3. There are other consequences of gum
disease: Bad breath, stroke, and pneumonia.
3. Too much alcohol increases the chance that you
will become a victim or an assailant in a forced
sex situation. It also increases the risk of STD.
4. Instructions on how to ﬂoss. 4. Binge drinking not only hurts your body, it
also affects the quality of life on campus and
it is everyone’s problem.
Table 2 Threat-to-freedom Manipulation in the Flossing Message
High threat Low threat
Regular ﬂossing can greatly reduce the
chances of gum disease. As any sensible person
can see, there is really no choice when it comes
to ﬂossing: You simply have to do it. In fact, the
scientiﬁc evidence showing a link between gum
disease and failure to ﬂoss is so overwhelming that
only a fool would possibly argue with it. And the
fact that gum disease can become the basis for
other severe problems such as stroke and
pneumonia makes it just stupid not to ﬂoss every
single day of your life.
Regular ﬂossing can greatly reduce the
chances of gum disease. And most people
would agree that ﬂossing is worthy of
serious consideration. There is pretty
compelling evidence showing a link
between gum disease and the failure to
ﬂoss. The majority of dentists view that
evidence as strong and clear. And, the
fact that gum disease can lead to other
severe problems such as heart disease,
stroke, diabetes, and pneumonia means
that you might want to think about
making ﬂossing a regular habit.
So, if you ﬂoss already, don’t stop even for a day.
And, if you haven’t been ﬂossing, right now is the
time to start. Today. Do it because you have to. Floss
every single day.
So, if you ﬂoss already, keep up the good
work. And if you haven’t been ﬂossing,
now might be a good time to start. In
fact, you may want to try it today.
Flossing: It’s easy. Do it because you have to! Set
a goal for yourself to ﬂoss everyday during the
next week (starting today)!
Flossing: It’s easy. Why not give it a try?
Set a goal for yourself to ﬂoss everyday
during the next week (starting today)!
152 J. P. Dillard & L. Shen
Perceived threat to freedom was measured by four Likert-type scale items: “The message
threatened my freedom to choose,” “The message tried to make a decision for me,” “The
message tried to manipulate me,” and “The message tried to pressure me.” Participants
were presented with a 5-point responsescale in which 1 ¼ strongly disagree, 2 ¼ disagree,
3 ¼ neutral/don’t know, 4 ¼ agree, and 5 ¼ strongly agree. The alpha reliability of the
scale was .83 in the ﬂossing data and .87 in the drinking data.
Anger was measured using four items that had been validated in previous studies:
irritated, angry, annoyed, and aggravated (Dillard & Peck, 2000; Dillard et al., 1996).
The 5-point response scale was anchored at 0 ¼ “none of this feeling” and 4 ¼ “a great
deal of this feeling.” Alpha reliability was .92in the ﬂossing data and .94 in the alcohol data.
Participants were asked to write out whatever was in their minds when they ﬁnished
reading the action component of the message. The resulting data were coded in
a four-step sequence by three coders working in inter-locking pairs. First, the coders
segmented the data into psychological thought units. Agreement among the three
pairs of coders was 97%, 96%, and 95%.
Second, because we viewed open-ended reports of affect as redundant with the
close-ended reports, affective responses were identiﬁed and removed. To assist
with this step, coders relied on a list of feeling terms compiled by Shaver, Schartz, Kirson,
Table 3 Threat-to-freedom Manipulation in the Drinking Message
High threat Low threat
Responsible Drinking: You Have to Do It Consider Responsible Drinking
The previous pages make it crystal clear:
There is unequivocal evidence that over-
consumption of alcohol is implicated in
reduced school performance, sexual
violence, secondary effects on others and
physical harm to the drinker. In fact, any
reasonable person has to agree that over-
consumption of alcohol is a serious
campus problem that demands
immediate attention. No other conclusion
makes any sense. Stop the denial. There
is a problem and you have to be
part of the solution.
As the previous pages tried to show, there is
pretty compelling evidence that over-
consumption of alcohol is implicated in
reduced school performance, sexual violence,
secondary effects on others and physical harm
to the drinker. In fact, most people agree that
over-consumption of alcohol is a serious
campus problem that needs to be addressed
right away. It’s a sensible conclusion and one
that is hard to deny. There is a problem and
you have a chance to be part of the solution.
So if you drink, drink responsibly. Three
drinks is a safe, reasonable, and responsible
limit and it’s the limit that you need to
stick to. Do it.
So if you drink, think about drinking
responsibly. Perhaps three drinks is a safe,
reasonable, and responsible limit and it’s a
limit you can live with. Why not give
responsible drinking a try?
& O’Connor (1987). A unit was classiﬁed as affective whenever those words appeared
and cognitive otherwise (k ¼ 1.00, 1.00, and 1.00).
Third, coders evaluated whether or not the cognitive responses were relevant to the
message. The purpose of this step is to eliminate irrelevant cognitions and thereby
reduce the level of noise in the data (k ¼ 0.94, 0.91, and 0.89).
Finally, the remaining data were coded either as (a) supportive thoughts, (b) neutral
thoughts, or (c) negative thoughts. Supportive thoughts were deﬁned as responses that
expressed agreement with the message, self-identiﬁcation, and positive thoughts toward
the message, the message source, or the advocacy; and intention to comply with the
advocacy in the message, etc. Negative thoughts were deﬁned as responses that expressed
disagreement with the message,negative intention to comply with theadvocacy, intention
to engage in the risky behavior, derogations of the source, etc. Neutral thoughts were
deﬁned as non-evaluative responses to the message, such as “This message is printed on
blue paper” (k ¼ 0.76, 0.74, and 0.70). Only the negative cognitions were used in
subsequent data analyses. In the ﬂossing data, participants generated an average of 1.47
negative thoughts ðSD ¼ 1:85Þ in response to the action component of the message. The
corresponding numbers for the drinking data were 1.47 ðSD ¼ 1:76Þ:
Participants’ attitudes toward the message advocacy (i.e., “Flossing regularly is” and
“To limit one’s alcohol consumption to three drinks or less is”) were measured by
seven 7-point semantic differential questions. The word pairs used were: bad/good;
foolish/wise; unfavorable/favorable; negative/positive; undesirable/desirable; unne-
cessary/necessary; and detrimental/beneﬁcial. Alpha reliabilities were .84 in the
ﬂossing data and .89 in the drinking data.
Behavioral intention was measured by a 100-point, single-item estimate of the
likelihood that participants would ﬂoss regularly in the following week or limit their
alcohol consumption to three drinks in the following week.
Hong’s (1992; Hong & Faedda, 1996; Hong & Page, 1989) reactance scale was used to
measure trait reactance proneness. Due to instability of the scale in some prior studies
(see Hong, 1992, vs. Hong & Faedda, 1996), a series of factor analyses were conducted
to evaluate the dimensionality of the scale. The results indicated ﬁrst-order multi-
dimensionality, but second-order uni-dimensionality. The details of these analyses are
presented in Shen and Dillard (in press). Thus, we treated the set of items as a single
measure of trait reactance proneness. The alpha reliabilities in the ﬂossing and
drinking studies were .79 and .83 respectively. Sample items from the Hong scale
include: “I consider advice from others to be an intrusion” and “I become frustrated
when I am unable to make free and independent decisions,” and “Advice and
recommendations usually induce me to do just the opposite.”
154 J. P. Dillard & L. Shen
Effects of the Manipulations
Perceived threat to freedom
In the ﬂossing study, high threat participants perceived signiﬁcantly stronger threat to
freedom ðM ¼ 2:98; SD ¼ 1:07Þ than did low threat participants ðM ¼ 2:31;
SD ¼ 0:95Þ, F ¼ 21:80; df ¼ 1; 200; p , :001;
¼ :32: Participants in the high threat
condition of the drinking study perceived signiﬁcantly stronger threat to freedom
ðM ¼ 3:11; SD ¼ 1:16Þ than participants in the low threat condition ðM ¼ 2:67;
SD ¼ 0:98Þ, F ¼ 8:67; df ¼ 1; 203; p , :01;
¼ :23: Both sets of results suggest that
the threat to freedom manipulations were successful.
Negative cognitive responses
In the ﬂossing study, high threat participants had signiﬁcantly more negative thoughts
ðM ¼ 1:76; SD ¼ 2:01Þ than participants in the low threat condition ðM ¼ 1:19;
SD ¼ 1:63Þ, F ¼ 4:81; df ¼ 1; 200; p , :05;
¼ :20: Participants in the high threat
condition of the drinking study had signiﬁcantly more negative thoughts ðM ¼ 2:10;
SD ¼ 2:02Þ than participants in the low threat condition ðM ¼ 0:82; SD ¼ 1:18Þ;
F ¼ 29:58; df ¼ 1; 203; p , :001;
In the ﬂossing study, high threat participants experienced stronger anger (M ¼ 0:90;
SD ¼ 1:09) than participants in the low threat condition (M ¼ 0:45; SD ¼ 0:73Þ, F ¼
11:34; df ¼ 1; 200; p , :01;
¼ :21: Participants in the high threat condition of the
drinking study reported signiﬁcantly stronger anger ðM ¼ 1:44; SD ¼ 1:19Þ than
participants in the low threat condition ðM ¼ 0:70; SD ¼ 0:96Þ; F ¼ 24:01; df ¼ 1; 203;
p , :001;
Results for the induction checks indicated that the threat manipulation successfully
induced differences in perceived threat to freedom (as indexed by the four-item scale).
Moreover, the threat manipulation produced differences in the number of negative
cognitions and the intensity of anger in both data sets. These latter ﬁndings do not
speak directly to questions regarding the nature of reactance. However, they do
demonstrate that the data exhibit one of the necessary conditions for a meaningful test
of mediation, that is, threat-induced variance in the proposed mediators.
Structural Equation Modeling
Input and model speciﬁcation
Tables 4 and 5 present the correlation matrix, means, and standard deviations of each
variable from the drinking and ﬂossing studies respectively. These summary data were
used as input to LISREL 8.30 (Jo
reskog & So
rbom, 1999). A covariance matrix was
constructed, which was then used to estimate the parameters of the models using
maximum likelihood procedures. To reduce heteroscadesticity, the sum of the 14
Hong reactance proneness items was divided by 14 and the 100-point behavioral
intention measure was divided by 10.
All of the variables in the models were treated as latent constructs
proneness and attitude were corrected for measurement error by ﬁxing the error term of
the corresponding manifest variable at 1-a
times its variance (Bollen, 1989). This
procedure was followed for anger and negative cognitions in the Dual Process Model.
However, these two variables were treated as manifest in the Intertwined Process Model
and their measurement paths were estimated from the data. The measurement paths for
threat and behavioral intention were set at 1.00.
Criteria for evaluating the models
Because the Dual Process Model contained the two Single Process alternatives, it was not
necessary to run all four of the competing models. Rather, examination of the Dual
Table 4 Correlations, Means, and Standard Deviations for the Flossing Data ðN ¼ 196Þ
Variable Mean SD 1234567
1. Threat 0.00
2. Proneness 2.98
2 0.03 1.00
3. Interaction 2 0.23
0.01 2 0.06 1.00
4. Anger 0.68 0.95 0.24 0.29 0.15 1.00
1.48 1.85 0.14 0.28 0.08 0.43 1.00
6. Attitude 4.66 0.49 2 0.11 2 0.15 2 0.11 2 0.22 2 0.27 1.00
7. Intention 7.21
0.03 2 0.16 2 0.04 2 0.15 2 0.16 0.45 1.00
High threat was coded as 1 and low threat as 2 1.
Proneness was transformed by dividing HPRS score by 14 and intention was transformed by
dividing response by 10 to minimize heteroscadesticity.
The interaction term was mean centered.
Table 5 Correlations, Means, and Standard Deviations for the Drinking Data ðN ¼ 200Þ
Variables Mean SD 1234567
1. Threat 0.00
2. Proneness 2.97
2 0.14 1.00
3. Interaction 2 1.13
2 0.01 0.05 1.00
4. Anger 1.06 1.14 0.32 0.15 0.01 1.00
1.47 1.76 0.37 0.16 0.09 0.45 1.00
6. Attitude 4.49 0.63 0.03 2 0.13 0.07 2 0.12 2 0.06 1.00
7. Intention 6.03
2 0.04 2 0.14 2 0.01 2 0.27 2 0.24 0.06 1.00
High threat was coded as 1 and low threat as 2 1.
Proneness was transformed by dividing HPRS score by 14 and intention was transformed by
dividing response by 10 to minimize heteroscadesticity.
The interaction term was mean centered.
156 J. P. Dillard & L. Shen
Process Model versus the Intertwined Process Model was sufﬁcient to exhaust all logical
possibilities. If the data supported the Single Process Cognitive Model, we should observe
a signiﬁcant path to and from negative cognition and non-signiﬁcant paths to and from
anger. The reverse pattern should obtain for the Single Process Affective Model.
For each model, strength of threat, reactance proneness, and a mean-deviated product
term were speciﬁed as exogenous variables that inﬂuenced either affect and cognition (in
the Dual Process Model) or a latent variable indexed by affect and cognition (in the
Intertwined Process Model). These mediating variables were speciﬁed as the immediate
causal antecedent of attitude, which in turn caused behavioral intention.
The models were evaluated on two criteria: signiﬁcance of the path coefﬁcients and
overall ﬁt. We planned to run each model, eliminate non-signiﬁcant paths, and then
repeat the analysis on the modiﬁed model with an eye toward changes in the goodness-of-
ﬁt indices. To evaluate the overall ﬁt of the models to the data, we considered three ﬁt
indices. First, the Adjusted Goodness of Fit Index (AGFI) produces values ranging from 0
to 1 with values in excess of .90 indicating good ﬁt. Second, Browne and Cudeck (1993)
contend that values of the Root Mean Square Error of Approximation (RMSEA) of .08 or
lower indicate reasonable ﬁt, though values of .06 or below should be preferred. Third, the
Bayesian Information Criterion (BIC; Raftery, 1995) is constructed such that negative
values provide evidence of model ﬁt, while positive BIC values suggest problematic model
ﬁt. One exceptional virtue of the BIC is its ability to compare the ﬁt of non-nested models.
Differences in BIC of 2 or more provide evidence favoring one model (with the smaller
BIC value) over another; 6 or more provide strong evidence; and 10 is taken to be very
strong evidence for model improvement (Raftery, 1995).
All of the predicted paths were signiﬁcant in initial analyses of the ﬂossing data.
Consequently, we turned our attention to the ﬁt indices (see Table 6), which uniformly
preferred the Intertwined Process Model: AGFI (.94 vs. .87), RMSEA (.006 vs. 110), BIC
(2 49.86 vs. 2 16.49). The BIC difference of 33.37 was strong evidence in favor of the
Intertwined Process Model. Figure 2 presents the path coefﬁcients for the Intertwined
Table7 presents theﬁt indices for the drinking data, all of which favoredthe Intertwined
Process Model over its competitors. That model had a larger AGFI (.93 vs. .76), a smaller
RMSEA (0.077 vs. 180), and a smaller BIC (2 28.71 vs. 12.71). The BIC difference of 41.42
Table 6 Fit Indices and Model comparison for the Flossing Data
Models df AGFI
Dual Process Model 9 .87 .110 2 16.49
Intertwined Process Model 13 .94 .006 2 49.86 33.37
AGFI ¼ Adjusted Goodness-of-ﬁt Index.
RMSEA ¼ Root Mean Square Error Approximation.
BIC ¼ Bayesian Information Criterion.
Figure 2 The obtained intertwined process model for the ﬂossing data.
Figure 3 The obtained intertwined process model for the alcohol data.
158 J. P. Dillard & L. Shen
also indicated the superiority of the model over the alternatives. However, analysis of
the drinking data yielded non-signiﬁcant paths for the interaction term onto reactance
and for attitude onto intention. The interaction term was removed, ﬁrst, and the
remaining paths re-estimated. When the modiﬁcation indices suggested the need for
elimination of the attitude-intention link and the addition of a direct path from reactance
to intention, we altered the model accordingly then re-estimated the parameters. Figure 3
presents the path coefﬁcients for the ﬁnal Intertwined Process Model.
Research Questions 1 and 2: The Nature of Reactance
The ﬁrst two research questions inquired as to the nature of reactance. RQ1 asked: Is
reactance best conceptualized as cognition, affect, or both? In fact, both data sets
showed that cognition and affect mediated the effects of threat and proneness on
attitude (see Tables 6 and 7). Thus, the two single process models were rejected.
RQ2 posed the question: If reactance has both cognitive and affective components,
how are the two components combined? Here too, the results were consistent across
the two data sets. The Intertwined Process Model was superior to the Dual Process
Model, in terms of ﬁt to the data, on both absolute (i.e., AGFI, RMSEA, BIC) and
relative (BIC difference) indices (see Tables 6 and 7).
Hypotheses 1 and 2: Antecedents of Reactance
H1 predicted a positive association between threat to freedom and reactance. The path
from threat to reactance was .31, p , :05; in the ﬂossing data and .51, p , :05; in the
drinking data. Therefore, Hypothesis 1 received consistent support across the two data sets.
H2 anticipated a positive correlation between reactance proneness and reactance.
The path from proneness to reactance was .41, p , :05; in the ﬂossing data and .29,
p , :05; in the drinking data. Therefore, Hypothesis 2 was supported in both studies
(see Figures 2 and 3).
Research Question 3: Interaction
RQ3 asked whether threat to freedom interacts with reactance proneness. The product
term that represented this interaction was signiﬁcant in the ﬂossing data (.20,
p , :05), but not in the drinking data (.05,ns). Thus, the two studies did not provide a
consistent answer to RQ3 (see Figures 2 and 3). Regression analysis was used to
explore the functional form of the interaction in the ﬂossing data. The data were
broken on threat, and the composite measure of reactance, using standardized
measures of anger and negative cognition, was regressed on reactance proneness. In
the high threat condition, the analysis yielded r ¼ :40; p , :001; whereas in the low
threat condition, the coefﬁcient was r ¼ :27; p , :001: That the interaction term was
signiﬁcant indicated that the two coefﬁcients were themselves signiﬁcantly different.
These results showed that reactance was strongest when both threat and reactance
proneness were high, and weakest when both were low. Thus, the interaction was of a
form that was compatible with the theory.
Our primary concern in this research was to illuminate the nature of reactance and the
operation in the persuasion process. However, because of the important ramiﬁcations
of those theoretical questions for applied undertakings, we chose to test our thinking
in the context provided by two health issues: ﬂossing and drinking. In the remainder of
this paper, we ﬁrst examine the conceptual issues, then turn our attention to questions
The Nature of Reactance
Brehm and Brehm (1981) argued that reactance could not be measured: The only
means available for assessing the presence of reactance was to infer it from its effects.
Our results, instead, suggest that it is possible to use a combination of self-report
cognitive and emotional measures to create a more or less direct index of reactance.
This conclusion rests on the ability of the combined variables to mediate the effects of
threat and reactance proneness on attitude and behavioral intention. However, there
are limitations to these conclusions that arise from the design of our experiment. First,
among them is the fact that our project demonstrated mediation for only two
independent variables and for only two message topics. Additional tests would
strengthen our certainty about the generalizability of the results.
The data clearly favored the Intertwined Process Model, which claimed that reactance
is best understood as an intermingling of negative cognition and anger. Moreover,
because the factor loadings for anger and cognition on reactance are similar in
magnitude in both data sets, it appears that each contributes about equally to the
motivation to restore freedom. But our claim that affect and cognition constitute a
psychological alloy is also limited by aspects of our research design. Consider that
cognition and affect are phenomena capable of rapid change. For example, an
appropriate time frame for the retrieval and integration of some elements of thought
may be on the order of 100 ms or less (van Turennout, Hagoort, & Brown, 1998). In
contrast, our data aggregated thoughts and feelings over the time that respondents took
to read the recommendation components of the two messages (i.e., the last page of six-
page messages). Informal observation indicated that this was a period of 10–15 s. Thus,
the data do support the idea that cognitive and affective responses that characterize
Table 7 Fit Indices and Model Comparison for the Drinking Data
Models df AGFI
Dual Process Model 9 .77 .161 5.65
Intertwined Model 11 .90 .089 2 28.14 33.79
Final Intertwined Process Model 9 .93 .077 2 28.71
AGFI ¼ Adjusted Goodness-of-ﬁt Index.
RMSEA ¼ Root Mean Square Error Approximation.
BIC ¼ Bayesian Information Criterion.
160 J. P. Dillard & L. Shen
reactance are empirically inseparable, but that claim is limited to the 10 –15 s period
assessed in this study. Although this in no way compromises the conclusion, it does
suggest that researchers attempting to model cognition and affect as indicators of
reactance over an entire message may not be able to replicate our results.
Antecedents of Reactance
A great deal of research has been conducted on reactance theory since the appearance
of Brehm’s (1966) treatise. Several studies have claimed to examine the effects of threat
and reactance proneness on reactance (e.g., Doob & Zabrack, 1971; Heller et al., 1973;
Kohn & Barnes, 1977; Worchel & Brehm, 1970). In reality, however, those projects
have tested the impact of threat and proneness on some outcome variable such as
attitude and behavior. Though cast as a mediator, reactance went unmeasured.
We believe that our data are the ﬁrst to demonstrate an effect for these two variables on
reactance itself, that is, the combination of negative cognition and anger. In that
respect, the results make a unique contribution to literature on reactance theory.
We anticipated that message variables that enhanced the strength of the threat to
freedom would show a positive association and this proved to be the case. Although
our ﬁndings are consistent with previous ﬁndings (e.g., Benoit, 1998; Bensley & Wu,
1991; Doob & Zabrack, 1971; Heller et al., 1973; Kohn & Barnes, 1977; Worchel
& Brehm, 1970), one might critique our messages on two grounds. One is that we
knowingly confounded intent to persuade with language intensity. Another is that we
may have unwittingly manipulated other message variables. For example, one reviewer
noted that both low-threat messages mention speciﬁc groups (i.e., “The majority of
dentists view that evidence as strong and clear” and “Most people agree that over-
consumption of alcohol is a serious campus problem”). From our perspective, neither
of these issues poses a concern because the aim of the study was not to generate
knowledge about the effects of speciﬁc message features on reactance. Instead, we
sought develop strong inductions and, importantly, to create messages that resembled
those used in previous research. This latter point was an over-riding concern. In fact,
the credibility of the tests of mediation hinged on independent and dependent
variables that were derived from the reactance literature in obvious ways. This was true
for the message variables as well as the measure of trait reactance. The second
hypothesis suggested that individuals who are prone to reactance would experience
a higher level of reactance as a function of the messages than persons who were not so
inclined. On this point too, the data supported our expectations, and in so doing, lend
further credence to claims of mediation.
RQ3 asked about the existence and the form of the interaction between threat to
freedom and trait reactance proneness. The data showed a signiﬁcant interaction in
the ﬂossing data that was of a form compatible with the theory. However, the
interaction effect was not replicated in the alcohol data. Taking these two ﬁndings
together, we can only conclude that reactance proneness may interact with threat for
some topics and not for others. To appreciate the nature of this conclusion, it is
important to bear in mind that trait reactance proneness represents a propensity to
experience a state, not the state itself. Even individuals who are high in trait proneness
are not constantly in state of reactance. Rather, they respond more strongly to the same
stimuli than do persons low in trait proneness. Our data show that this tendency is
additive for one topic and multiplicative for another, thereby suggesting that some
topics are especially potent with regard to their ability to elicit state reactance from
individuals high in the propensity to experience reactance.
One implication of these ﬁndings is that message designers should avoid high threat
messages altogether. While we would endorse that reasoning, it is important to
remember that threat is not a property of messages, but rather a judgment made by
members of the target audience. Thus, advice to message designers to cast messages as
informative, rather than persuasive, and to steer clear of dogmatic language is sound,
but limited. A course of action that might produce more valuable knowledge would be
to work toward a theoretically grounded means of classifying health topics that are
likely to interact with strength of threat versus those that are not. Reactance theory
and research directs our attention to variables such as importance of the
behavior (Bensley & Wu, 1991) or prior intention (Buller et al., 2000). But other
concepts may need to be considered too. The two topics examined in this research
differed on the extent to which the behavior is public versus private as well as the
direction of the behavior change that was called for (i.e., more ﬂossing vs. less
Implications for Assessing Reactance Induced by Persuasive Messages
Although we stand by the conclusion that reactance should be viewed as a cognitive and
affective amalgam, it does not follow that all anger or all negative cognitions result from
perceived threats to freedom. In fact, from a theoretical standpoint, we prefer to view
reactance simply as the output of more general and frequently interdependent
psychological systems (i.e., cognition and emotion). In practice, the tight interweaving
of cognition and affect assumes less signiﬁcance. As the results relate to the evaluation of
persuasive health messages, we see two important conclusions: (a) Reactance can be
measured using well-known and commonly applied self-report techniques such as those
used in this study, and (b) both anger and negative cognitions should be assessed. For
purposes of formative or summative message evaluation, it matters little whether or not a
composite is created from them. As long as both are assessed the evaluation will be
comprehensive. In fact, as we noted earlier, efforts to treat affect and cognition as
indicators of a latent construct may fail in research designs that focus on whole messages
rather than message components. Evaluation researchers should take heart from the fact
that they already have in hand the tools to assess reactance as well as other affects
Reactance Theory and Message Design
Further inquiry is warranted into the question of how messages can be designed
that will defuse, or at least not exacerbate, innate tendencies toward reactance.
162 J. P. Dillard & L. Shen
As this research proceeds, it will be important to acknowledge the limits of
the relatively narrow view of communication that is characteristic of inquiry in
the reactance tradition. As was done in the current investigation, studies of
reactance typically characterize persuasive messages uni-dimensionally in terms of
degree of threat. However, empirical research on the perception of inﬂuence
messages clearly reveals that individuals utilize at least three distinct dimensions
(see Dillard, Wilson, Tusing, and Kinney, 1997, for a summary). Explicitness is the
degree to which the language of the message makes plain the source’s intent.
Roughly parallel to forcefulness or authoritarianism, dominance captures the
extent to which a message reveals that the source believes he or she can control
the message recipient. Finally, reason is present in any given message when
justiﬁcations are offered in support of the claim that audience members should
adopt the position advocated by the source. Although our knowledge base
concerning the effects of these three dimensions is not large, such data as do exist
suggest that dominance causes anger (Dillard, Kinney, & Cruz, 1996), whereas
reason giving softens perceptions of intrusiveness (Dillard et al., 1997) and is,
therefore, less likely to provoke anger/reactance. The effect of explicitness is likely
to be highly context dependent, but the available research shows that it generally
enhances persuasion (O’Keefe, 1997) and may produce positive emotional
responses (Dillard et al., 1996). In light of the distinctiveness and variability of the
impact of these aspects of message perception, it is essential that message design
research move beyond a unidimensional treatment of threat.
Attitude-Behavior Correspondence: Flossing, Binge Drinking, and Beyond
Some very well-established theories cast attitude as the immediate causal precursor of
behavioral intention (Ajzen, 1991; Fishbein & Ajzen, 1975). In this project, that
theoretical sequence was supported in the ﬂossing data, but not in the drinking study.
In fact, although both attitude toward reducing alcohol consumption and intention to
reduce consumption were negatively inﬂuenced by reactance, the two outcome
variables were empirically independent of one another in the alcohol study. We believe
that the ﬁndings for the two speciﬁc topics may be illustrative of two more general
For ﬂossing, attitude and intention correspond well with one another because
individuals approach the topic with only one aim in mind, that is, health maintenance.
They consider the relevant facts, integrate that information to form an attitude, and
then generate a corresponding intention. In other words, they behave in a reasoned,
planful manner. Accordingly, the prospects are bright for communication campaigns
that adopt a rational approach to increasing the frequency of dental ﬂossing.
For those topics about which individuals are concerned only with self-preservation,
the likelihood of success seems high for appeals that detail the costs and beneﬁts of
Persuasive messages concerning alcohol are likely to be processed quite differently
from appeals to increase dental ﬂossing. Whereas message recipients still may be
responsive to arguments to protect their health, they are likely to be simultaneously
inﬂuenced by an awareness of the many social and commercial forces that promote
binge drinking on college campuses. When these forces are strong in the aggregate,
message recipients may believe that actual behavior is beyond their control.
In fact, Wall, Hinson, and McKee (1998) report an inverse relationship between
perceived control over excessive drinking and intention to drink to excess. Under such
circumstances, attitude toward the behavior is functionally irrelevant to the intention
itself (Greenwald, 1989). The result is an empirical disconnect between the two
constructs: The attitude becomes unrelated to action because behavior is the result of
other forces such as perceived norms (Fishbein & Yzer, 2003). In these instances, a
“successful” persuasive message might change attitude without any corresponding
impact on intention. However, as the alcohol data show (see Figure 3), the aftermath
of reactance arousal is less benign. It can exert a direct effect on behavioral intention.
This is a particularly worrisome problem for health campaigns in light of the fact that
those persons at greatest risk are often the same persons for whom the behavior is
most important. An illustration of this point can be found in Bensley and Wu (1991),
a study in which the importance of drinking (indexed by prior behavior) interacted
with strength of threat to produce substantially heightened alcohol consumption
among male, heavy drinkers. In combination with the current project, such ﬁndings
underscore the difﬁculty of conducting any campaign on a general audience without
arousing reactance among members of the target group. Not to be overlooked is the
possibility of social and cultural iatrogenic effects that may have their roots in
reactance (Guttman, 2000). Of course, such pessimistic possibilities do not mean that
every effort at enhancing public health will do harm. Well-constructed messages that
are subjectively evaluated as compelling have the potential to persuade even in the
presence of reactance (Brehm, 1966; Brehm & Brehm, 1981). Surely some
consideration of the balance between good and ill effects is needed at the planning
and evaluation stages of every persuasive campaign.
Two studies were carried out to explore the nature of reactance and its role in the
impact of persuasive health messages. Whereas the theory itself contends that
reactance cannot be measured, this project supported a different conclusion. The
data indicated that reactance might reasonably be conceived of as an amalgam of
anger and negative cognitions. This result is useful to efforts to improve public health
in that reactance can be assessed in formative and evaluative research using well-
known and widely understood self-report methods. The study also suggested that
strength of threat may interact with trait reactance proneness, but that it is unlikely
to do so for all health topics. Finally, the data implied that attitude-behavior
correspondence may be strong for certain types of health behaviors and weak for
others. We speculated that conditions that place the goal of self-preservation in
conﬂict with other motives might help to explain poor correspondence. In all, these
164 J. P. Dillard & L. Shen
results extend our understanding of reactance theory and its application to
persuasive health communication.
 One alternative to this position is to conceptualize reactance in other terms. For example, as one
anonymous reviewer pointed out, one might think of reactance as a cognitive appraisal of
restriction or illegitimacy. If so, self-report would presumably provide a valid means of
measuring reactance. We concur with this reasoning, but note (as the reviewer did as well) that it
would require one to explicate how the appraisal differs from counter-arguing. More to the
point, we think that to treat reactance as a purely cognitive judgment is to rob it of
the motivational properties that Brehm so clearly believed it possesses. Hence, while we admit to
the logical possibility of conceptualizing reactance in different terms, we see the solution as one
that creates substantial distance from the original theory and judge it undesirable for that reason.
 Some readers might reasonably object to our models on the grounds that we did not treat the
perceived threat induction check as a mediator between the induction and mediators of interest
(i.e., affect and cognition). Although we believe that such an analysis is desirable in principle, it
would have precluded tests of the interaction between strength of threat and reactance proneness
(because one cannot form an interaction term between an endogeneous and an exogeneous
variable). Because the interaction was something that we wished to test, we opted for models that
allowed us to do that.
Ajzen, I. (1991). The theory of planned behavior. Organizational Behavior and Human Decision
Processes, 50, 179–211.
Benoit, W. L. (1998). Forewarning and persuasion. In M. Allen, & R. W. Preiss (Eds.), Persuasion:
Advances through meta-analysis (pp. 139–154). Cresskill, NJ: Hampton Press Inc.
Bensley, L. S., & Wu, R. (1991). The role of psychological reactance in drinking following alcohol
prevention messages. Journal of Applied Social Psychology, 21, 1111–1124.
Bollen, K. (1989). Structural equations with latent variables. New York: Wiley.
Brehm, J. W. (1966). A theory of psychological reactance. New York: Academic Press.
Brehm, J. W., Stires, L. K., Sensenig, J., & Shaban, J. (1966). The attractiveness of an eliminated choice
alternative. Journal of Experimental Social Psychology, 2, 301–313.
Brehm, S. S., & Brehm, J. W. (1981). Psychological reactance: A theory of freedom and control.
New York: Academic Press.
Brown, M. W., & Cudeck, R. (1993). Alternative ways of assessing model ﬁt. In K. A. Bollen, & J. S.
Long (Eds.), Testing structural equation models (pp. 136–162). Newbury Park, CA: Sage.
Buller, D. B., Borland, R., & Burgoon, M. (1998). Impact of behavioral intention on effectiveness of
message features: Evidence from the family sun safety project. Human Communication
Research, 24, 422–453.
Buller, D. B., Burgoon, M., Hall, J. R., Levine, N., Taylor, A. N., Beach, B., Buller, M. K., & Melcher, C.
(2000). Long-term effects of language intensity in preventive messages on planned family solar
protection. Health Communication, 12, 262–275.
Burgoon, M., Alvaro, E., Grandpre, J., & Voulodakis, M. (2002). Revisiting the theory of
psychological reactance: Communicating threats to attitudinal freedom. In J. P. Dillard, &
M. W. Pfau (Eds.), The persuasion handbook: Developments in theory and practice
(pp. 213–232). Thousand Oaks, CA: Sage.
Cacioppo, J. T., & Petty, R. E. (1981). Social psychological procedures for cognitive response
assessment: The thought-listing technique. In T. V. Merluzzi, C. R. Glass, & M. Genest (Eds.),
Cognitive assessment (pp. 309–342). New York: Guilford Press.
Chaffee, S. H. (1991). Explication. Newbury Park, CA: Sage.
Derryberry, D., & Tucker, D. M. (1994). Motivating the focus of attention. In P. N. Niedenthal, &
S. Kitayama (Eds.), The heart’s eye: Emotional inﬂuences in perception and attention
(pp. 167–196). San Diego, CA: Academic Press.
Dillard, J. P., Kinney, T. A., & Cruz, M. G. (1996). Inﬂuence, appraisals, and emotions in close
relationships. Communication Monographs, 63, 105–130.
Dillard, J. P., & Meijnders, A. (2002). Persuasion and the structure of affect. In J. P. Dillard, & M. Pfau
(Eds.), The persuasion handbook: Developments in theory and practice (pp. 309–327). Newbury
Park, CA: Sage.
Dillard, J. P., & Peck, E. (2000). Affect and persuasion: Emotional responses to public service
announcements. Communication Research, 27, 461–495.
Dillard, J. P., & Peck, E. (2001). Persuasion and the structure of affect: Dual systems and discrete
emotions as complementary models. Human Communication Research, 27, 38–68.
Dillard, J. P., Plotnick, C. A., Godbold, L. C., Freimuth, V. S., & Edgar, T. (1996). The multiple
affective consequences of AIDS PSAs: Fear appeals do more than scare people. Communication
Research, 23, 44–72.
Dillard, J. P., Wilson, S. R., Tusing, K. J., & Kinney, T. A. (1997). Politeness judgments in personal
relationships. Journal of Language and Social Psychology, 16, 297–325.
Doob, A. N., & Zabrack, M. (1971). The effect of freedom-threatening instructions and monetary
inducement on compliance. Canadian Journal of Behavioral Science, 3, 408–412.
Dowd, E. T., Hughs, S. L., Brockbank, L., & Halpain, D. (1988). Compliance-based and deﬁance-
based intervention strategies and psychological reactance in the treatment of free and unfree
behavior. Journal of Counseling Psychology, 35, 370–376.
Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, and behavior: An introduction to theory and
research. Reading, MA: Addison-Wesley.
Fishbein, M., & Yzer, M. C. (2003). Using theory to design effective health behavior interventions.
Communication Theory, 13, 164–183.
Foxcraft, D. R., Lister-Sharp, D., & Lowe, G. (1997). Alcohol misuse prevention for young people:
A systematic review reveals methodological concerns and lack of reliable evidence of
effectiveness. Addiction, 92, 531–537.
Greenwald, A. G. (1989). Why are attitudes important? In A. R. Pratkanis, S. J. Breckler, & A. G.
Greenwald (Eds.), Attitude structure and function (pp. 1–11). Hillsdale, NJ: Lawrence Erlbaum
Guttman, N. (2000). Public health communication interventions: Values and ethical dilemmas.
Thousand Oaks, CA: Sage.
Guttman, N., Kegler, M., & McLeroy, K. R. (1996). Health promotion paradoxes, antinomies, and
conundrums. Health Education Research, Theory, and Practice, 11, i–xiii.
Hammock, T., & Brehm, J. W. (1966). The attractiveness of choice alternatives when freedom to
choose is eliminated by a social agent. Journal of Personality, 34, 546–554.
Heller, J., Pallak, M., & Picek, J. (1973). The interactive effects of intent and threat on boomerang
attitude change. Journal of Personality and Social Psychology, 26, 273–279.
Hong, S. M. (1992). Hong’s Psychological Reactance Scale: A further factor analytic validation.
Psychological Reports, 70, 512–514.
Hong, S. M., & Faedda, S. (1996). Reﬁnement of the Hong Psychological Reactance Scale.
Educational and Psychological Measurement, 56, 173–182.
Hong, S. M., & Page, S. (1989). A psychological reactance scale: Development, factor structure and
reliability. Psychological Reports, 64, 1323–1326.
166 J. P. Dillard & L. Shen
Hornik, R. (2002). Public health communication: Making sense of the evidence. In R. Hornik (Ed.),
Public health communication (pp. 1–19). Mahwah, NJ: Lawrence Erlbaum Associates, Inc.
reskog, K., & So
rbom, D. (1996). Lisrel 8: User’s reference guide. Chicago: Scientiﬁc Software
Kelly, A. E., & Nauta, M. M. (1997). Reactance and thought suppression. Personality and Social
Psychology Bulletin, 23, 1123–1132.
Kim, K. S., & Hunter, J. E. (1993a). Attitude–behavior relations: A meta-analysis of attitudinal
relevance and topic. Journal of Communication, 43, 101–142.
Kim, K. S., & Hunter, J. E. (1993b). Relationships among attitudes, behavioral intentions, and
behavior. Communication Research, 20, 331–364.
Kohn, P. M., & Barnes, G. E. (1977). Subject variables and reactance to persuasive communications
about drugs. European Journal of Social Psychology, 7, 97–109.
Kuhl, J. (1986). Motivation and information processing: A new look at decision making, dynamic
change, and action control. In R. M. Sorrentino, & E. T. Higgins (Eds.), Handbook of
motivation and cognition: Foundations of social behavior (pp. 404–434). New York: John Wiley.
Lang, P. J. (2000). Emotion and motivation: Attention, perception, and action. Journal of Sports and
Exercise Psychology, 22, S122–S140.
Leventhal, H. (1970). Findings and theory in the study of fear communications. In L. Berkowitz
(Ed.), Advances in experimental social psychology, (Vol. 5, pp. 119–186). New York: Academic
Merz, J. (1983). A questionnaire for the measurement of psychological reactance [in German].
Diagnostica, 29, 75–82.
Nabi, R. L. (2002). Discrete emotions and persuasion. In J. P. Dillard, & M. Pfau (Eds.), The
persuasion handbook: Developments in theory and practice (pp. 289–308). Newbury Park, CA:
O’Keefe, D. J. (1997). Argumentative candor and persuasive success: A meta-analysis of the
persuasive effects of implicit and explicit message conclusions. In J. F. Klumpp (Ed.),
Argument in a time of change: Deﬁnitions, frameworks, and critiques (pp. 63–69). Annandale,
VA: National Communication Association.
Petty, R. E., & Cacioppo, J. T. (1986). Communication and persuasion: Central and peripheral routes to
attitude change. New York: Springer-Verlag.
Raftery, A. E. (1995). Bayesian model selection in social research. In P. V. Marsden (Ed.), Sociological
methodology (pp. 111–163). Cambridge, UK: Basil Blackwell.
Ringold, D. J. (2002). Boomerang effect in response to public health interventions: Some unintended
consequences in the alcoholic beverage market. Journal of Consumer Policy, 25, 27–63.
Rogers, R. W. (1983). Cognitive and physiological 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 Press.
Schwarz, N., Frey, D., & Kumpf, M. (1980). Interactive effects of writing and reading a persuasive
essay on attitude change and selective exposure. Journal of Experimental Social Psychology, 16,
Seltzer, L. F. (1983). Inﬂuencing the “shape” of resistance: An experimental exploration of
paradoxical directives and psychological reactance. Basic and Applied Social Psychology, 4,
Shaver, P., Schwartz, J., Kirson, D., & O’Connor, C. (1987). Emotion knowledge: Further
explorations of a prototype approach. Journal of Personality and Social Psychology, 52,
Shen, L., & Dillard, J. P. (in press). Psychometric properties of the Hong Psychological Reactance
Scale. Journal of Personality Assessment.
Smith, M. J. (1977). The effects of threats to attitudinal freedom as a function of message quality and
initial receiver attitude. Communication Monographs, 44, 196–206.
Stephenson, M. T. (2003). Examining adolescents’ responses to antimarijuana PSAs. Human
Communication Research, 29, 343–369.
Stewart, D. W., & Martin, I. M. (1994). Intended and unintended consequences of warning messages:
A review and synthesis of empirical research. Journal of Public Policy and Marketing, 13, 1–19.
Van Turennout, M., Hagoort, P., & Brown, C. M. (1998). Brain activity during speaking: From syntax
to phonology in 40 milliseconds. Science, 280, 572–574.
Wall, A. -M., Hinson, R. E., & McKee, S. A. (1998). Alcohol outcome expectancies, attitudes toward
drinking and the theory of planned behavior. Journal of Studies on Alcohol, 59, 409–419.
Wallack, L. M. (1981). Mass media campaigns: The odds against ﬁnding behavior change. Health
Education Quarterly, 8, 209–260.
White, G. L., & Zimbardo, P. G. (1980). The effects of threat of surveillance and actual surveillance on
expressed opinion toward marijuana. Journal of Social Psychology, 111, 49–61.
Wicklund, R. A. (1974). Freedom and reactance. Potomac, MD: Lawrence Erlbaum Associates, Inc.
Wilde, G. J. S. (1993). Effects of mass media communications on health and safety habits:
An overview of issues and evidence. Addiction, 88, 983–996.
Witte, K. (1994). Fear control and danger control: A test of the extended parallel process model
(EPPM). Communication Monographs, 61, 113–134.
Worchel, S. (1974). The effect of three types of arbitrary thwarting on the instigation to aggression.
Journal of Personality, 42, 300–318.
Worchel, S., & Andreoli, V. (1974). Attribution of causality as a means of restoring behavioral
freedom. Journal of Personality and Social Psychology, 29, 237–245.
Worchel, S., Andreoli, V. A., & Archer, R. (1976). When is a favor a threat to freedom: The effects of
attribution and importance of freedom on reciprocity. Journal of Personality, 44, 294–310.
Worchel, S., & Brehm, J. (1970). Effect of threats to attitudinal freedom as a function of agreement
with the communicator. Journal of Personality and Social Psychology, 14, 18–22.
168 J. P. Dillard & L. Shen