Psychological Review

1967, Vol. 74, No. 3, 183-200

SELF-PERCEPTION:
AN ALTERNATIVE INTERPRETATION OF COGNITIVE
DISSONANCE PHENOMENA 1

DARYL J. BEM
Carnegie Institute of Technology

A theory of self-perception is proposed to provide an alternative in-

terpretation for several of the major phenomena embraced by Fest-
inger's theory of cognitive dissonance and to explicate some of the
secondary patterns of data that have appeared in dissonance experi-
ments. It is suggested that the attitude statements which comprise the
major dependent variables in dissonance experiments may be regarded
as interpersonal judgments in which the observer and the observed
happen to be the same individual and that it is unnecessary to postu-
late an aversive motivational drive toward consistency to account for
the attitude change phenomena observed. Supporting experiments are
presented, and metatheoretical contrasts between the "radical" be-
havioral approach utilized and the phenomenological approach typi-
fied by dissonance theory are discussed.

If a person holds two cognitions that ducted within the framework of dis-
are inconsistent with one another, he sonance theory; and, in the 5 years
will experience the pressure of an since the appearance of their book,
aversive motivational state called cog- every major social-psychological jour-
nitive dissonance, a pressure which he nal has averaged at least one article
will seek to remove, among other ways, per issue probing some prediction "de-
by altering one of the two "dissonant" rived" from the basic propositions of
cognitions. This proposition is the dissonance theory. In popularity, even
heart of Festinger's (1957) theory of the empirical law of effect now appears
cognitive dissonance, a theory which to be running a poor second.
has received more widespread attention The theory has also had its critics.
from personality and social psycholo- Reservations about various aspects of
gists in the past 10 years than any other the theory have ranged from mild
contemporary statement about human (e.g., Asch, 1958; Bruner, 1957; Kelly,
behavior. Only 5 years after its in- 1962; Osgood, 1960; Zajonc, 1960) to
troduction, Brehm and Cohen (1962) severe (Chapanis & Chapanis, 1964),
could review over 50 studies con- and alternative interpretations have
been offered to account for the results
research was supported in part by of particular studies (e.g., Chapanis &
Ford Foundation Grant 1400SS to Carnegie Chapanis, 1964; Janis & Gilmore, 1965 ;
Institute of Technology and in part by the Lott, 1963; Rosenberg, 1965). No
Center for Research on Language and Lan-
guage Behavior, University of Michigan, theoretical alternative to dissonance
with funds from the Bureau of Higher Edu- theory has been proposed, however,
cation Research, United States Office of Edu- which attempts both to embrace its
cation. The author is grateful to George major phenomena and to account for
R. Madaras and Kenneth M. Peterson for
aid in conducting the research and to Sandra some of the secondary patterns of re-
L. Bern for critical comments on the manu- sults which have appeared in the sup-
script. porting experiments but which were
183
184 DARYL J. BEM

not predicted by the theory. This ar- ies is, with very few exceptions, a sub-
ticle proposes such an alternative. ject's (S's) self-descriptive statement
Like many theories in psychology, of an attitude or belief. Indeed, this
the theory of cognitive dissonance at- is the dependent variable in nearly all
tempts to account for observed func- of contemporary social psychology.
tional relations between current stim- But how are such self-descriptive be-
uli and responses by postulating some haviors acquired ? What are their con-
hypothetical process within the organ- trolling variables? It is to these ques-
ism, in this case, an inferred process of tions that the analysis turns first.
the arousal and reduction of disso-
nance. Like many other contemporary SELF-PERCEPTION : A SPECIAL CASE
personality and social psychological OF INTERPERSONAL PERCEPTION
theories, dissonance theory is further
characterized by an emphasis on the Self-perception, an individual's abil-
individual's current phenomenology; ity to respond differentially to his own
the explanatory account in the theory behavior and its controlling variables,
itself is ahistorical. is a product of social interaction
In contrast, the alternative formu- (Mead, 1934; Ryle, 1949; Skinner,
lation to be presented here eschews any 1957). Verbal statements that are
reference to hypothetical internal pro- self-descriptive are among the most
cesses and seeks, rather, to account for common responses comprising self-per-
observed functional relations between ception, and the techniques employed
current stimuli and responses in terms by the community to teach its members
of the individual's past training history. to make such statements would not
Such an approach has been called "rad- seem to differ fundamentally from the
ical" behaviorism (see Scriven, 1956), methods used to teach interpersonal
a position most often associated with perception in general. The community,
the name of B. F. Skinner. In ana- however, does face severe limitations
lyzing a complex behavioral phenome- in training the individual to make
non, the radical behaviorist attempts to statements describing internal events
establish it as a special case of some to which only he has direct access.
previously substantiated functional re- Skinner (1953, 1957) has analyzed
lation discovered in the experimental the limited resources available to the
analysis of simpler behaviors. His community for training its members
functional analysis is thus based on em- thus to "know themselves," and he has
pirical generalization and, accordingly, described the inescapable inadequacies
is frankly inductive not only in its ex- of the resulting knowledge.
perimental execution, but in its formal Skinner suggests that some self-de-
presentation. scriptions of internal stimuli can be
A functional analysis characteristi- learned through metaphor or stimulus
cally begins by inquiring into the onto- generalization. The child, for example,
genetic origins of the observed depend- can easily learn to describe "butterflies
ent variable and attempts to ascertain in the stomach" without explicit dis-
the controlling or independent vari- crimination training. More often, how-
ables of which that behavior is a func- ever, a socializing community must
tion. The present analysis of disso- teach the descriptive responses more
nance phenomena proceeds in the same directly. In training a child to de-
way by noting first that the dependent scribe pain, for example, the com-
variable in cognitive dissonance stud- munity, at some point, must teach him
 SELF-PERCEPTION 185

the correct response at the critical time extensive cross-cultural generality as

when the appropriate private stimuli well (Osgood et al., 1957). These
are impinging upon him. But the findings, too, are consistent with the
community itself must necessarily view that an individual is unable to
identify the "critical time" on the basis make more than a small number of in-
of observable stimuli or responses and dependent discriminations among stim-
implicitly assume that the private stim- uli that have never been publicly avail-
uli are, in fact, accompanying these able to a socializing community, and it
public events. is suggested that the many subtle dis-
This analysis suggests that many of criminations which individuals do make
the self-descriptive statements that ap- when describing their attitudes are
pear to be exclusively under the dis- based, rather, on the kinds of cues that
criminative control of private stimuli are potentially available to an outside
may, in fact, still be partially controlled observer. In particular, it is sug-
by the same accompanying public gested that self-descriptive attitude
events used by the training community statements can be based on the indi-
to infer the individual's inner states. vidual's observations of his own overt
Private stimuli may play a smaller role behavior and the external stimulus con-
than the individual himself suspects. ditions under which it occurs. A num-
For example, by manipulating the ex- ber of recent experimental studies
ternal cues of the situation, Schachter provide support for this proposition.
and Singer (1962) were able to evoke Several studies have shown that an
self-descriptions of emotional states as individual's belief and attitude state-
disparate as euphoria and anger from ments can be manipulated by inducing
5s in whom operationally identical him to role-play, deliver a persuasive
states of physiological arousal had been communication, or engage in any be-
induced. It appears that these 5s util- havior that would characteristically
ized internal stimuli only to make the imply his endorsement of a particular
gross discrimination that they were set of beliefs (Brehm & Cohen, 1962;
emotional, but that the more subtle King & Janis, 1956; Scott, 1957,
discrimination of which emotion they 1959). A recent experimental analysis
were experiencing was under the con- of these phenomena of "self-persuasion"
trol of external cues. demonstrates that an individual bases
A similar division of control be- his subsequent beliefs and attitudes on
tween internal and external stimuli ap- such self-observed behaviors to the
pears to operate in the domain of atti- extent that these behaviors are emitted
tude statements. Osgood, Suci, and under circumstances that have in the
Tannenbaum (1957) theorize that a past set the occasion for telling the
pattern of internal responses elicited truth (Bern, 1965, 1966). For ex-
by a word or an object comprises the ample, in one of three studies reported
connotative or "emotional" meaning of in Bern (1965), 5s were first trained
the stimulus for an individual, includ- to tell the truth in the presence of a
ing his attitude toward it. Using the colored light and to tell lies in the pres-
Semantic Differential technique, these ence of another. Later in the experi-
investigators report that an individual's mental session, 5s were required to
verbal descriptions of these hypothe- state attitudes with which they dis-
sized internal responses can be factor agreed; one of the two colored lights
analyzed into a very small number of was illuminated as each attitude state-
factors, factors which appear to have ment was made. It was found that 5s
186 DARYL J. BEM
subsequently endorsed the attitude I'm always eating it," it seems un-
statements they had uttered in the necessary to invoke a fount of privi-
presence of the "truth light" signifi- leged self-knowledge to account for the
cantly more than attitude statements reply. In such a case the reply is
they had made in the presence of the functionally equivalent to one his wife
"lie light"; the lights, in short, de- might give for him: "I guess he does,
termined the degree to which 5"s be- he is always eating it." Only to the
lieved what they had heard themselves extent that "brown bread" elicits
say. Furthermore, no 5" could report strongly conditioned internal responses
any awareness of the control exerted might he have additional evidence, not
by his statements or the lights over his currently available to his wife, on
subsequent attitudes. which to base his self-descriptive at-
In another study, the same technique titude statement.
was employed to demonstrate that an The present analysis of dissonance
individual can be induced to believe in phenomena, then, will rest upon the
"false confessions" he has made if single empirical generalization that an
there are external cues present that individual's belief and attitude state-
characteristically set the occasion for ments and the beliefs and attitudes that
telling the truth (Bern, 1966). These an outside observer would attribute to
several studies have also illustrated that him are often functionally similar in that
the control over an individual's beliefs both sets of statements are partial "in-
and attitudes exerted by his overt be- ferences" from the same evidence: the
havior is vitiated to the extent that cues public behaviors and accompanying
are present implying that the behavior stimulus cues upon which the social-
is deceitful or, more generally, is being izing community has relied in training
emitted for immediate specific rein- him to make such self-descriptive state-
forcement. For example, just as a ments in the first place.
communicator is more persuasive to
others if he is known to be receiving PHENOMENA OF DISSONANCE THEORY
no payment for his communication, so
too, it is found that he is more likely The major phenomena of dissonance
to believe himself under such circum- theory have been classified into three
stances (Bern, 1965). The effective- main categories (Brehm & Cohen,
ness of self-persuasion can thus be al- 1962, p. 21) : (a) forced-compliance
tered by many of the techniques typi- studies; (b) free-choice studies; and
cally used to manipulate the credibility (c) exposure-to-information studies.
of any persuasive communicator. Within each category, this discussion
The major implication of these find- will treat the major functional relation
ings is that, to the extent that internal predicted and subject the data from a
stimuli are not controlling, an indi- single dissonance experiment to de-
vidual's attitude statements may be tailed analysis. Two sets of secondary
viewed as inferences from observations findings will also be discussed.
of his own overt behavior and its ac- Because the literature of dissonance
companying stimulus variables. As theory has now become so large that it
such, his statements are functionally would be impossible to discuss all the
similar to those that any outside ob- experimental paradigms that have been
server could make about him. When employed, the specific experiments se-
the answer to the question, "Do you lected for detailed analysis had to
like brown bread?" is "I guess I do, satisfy certain criteria. First, when-
 SELF-PERCEPTION 187

ever possible, they had to be illustra- THE FORCED-COMPLIANCE STUDIES

tive of several others in the same cate-
gory so that the applicability of the The most frequently cited evidence
self-perception analysis to studies not for dissonance theory comes from an
explicitly discussed would be apparent. experimental procedure known as the
Second, for each experiment there had forced-compliance paradigm. In these
to be at least one other study in the experiments, an individual is induced
literature that had successfully repli- to engage in some behavior that would
cated the same conceptual phenomenon imply his endorsement of a particular
employing different experimental pro- set of beliefs or attitudes. Following
cedures if possible. These first two his behavior, his "actual" attitude or
criteria together attempt to ensure that belief is assessed to see if it is a func-
the analysis avoids "explaining" phe- tion of the behavior in which he has
nomena that are artifactual, while at the engaged and of the manipulated stimu-
same time not requiring that the par- lus conditions under which it was
ticular study under analysis be invul- evoked. The best known and most
nerable to methodological criticism. widely quoted study of this type was
Finally, the studies selected are those conducted by Festinger and Carlsmith
which are best known and most widely (1959). In their experiment, 60 un-
reprinted or cited. dergraduates were randomly assigned
There will be no discussion of stud- to one of three experimental condi-
ies that simply use the vocabulary of tions. In the $1 condition, 5" was first
dissonance theory but which explore required to perform long repetitive
functional relations that are not deri- laboratory tasks in an individual ex-
vations from the major propositions of perimental session. He was then hired
the theory (e.g., studies of postdecision by the experimenter as an "assistant"
regret; Festinger, 1964). There will and paid $1 to tell a waiting fellow
also be no additional discussion of student (a stooge) that the tasks were
phenomena which, although derivable enjoyable and interestng. In the $20
from dissonance theory propositions, condition, each S was hired for $20 to
are already considered by the disso- do the same thing. Control 5"s simply
nance theorists themselves to be as engaged in the repetitive tasks. After
parsimoniously accounted for by the experiment, each 5" indicated how
straightforward empirical generaliza- much he had enjoyed the tasks. The
tions concerning the interpersonal judg- results show that .$> paid $1 evaluated
mental skills of 5"s (e.g., attitude the tasks as significantly more enjoy-
change phenomena produced by per- able than did 5"s who had been paid $20.
suasive communication). (See Brehm The $20 Ss did not express attitudes
significantly different from those ex-
& Cohen, 1962, pp. 105-111.) Indeed, pressed by the control 6"s.
the primary purpose of the present Dissonance theory interprets these
analysis is to extend this same kind of findings by noting that all 5s initially
empirical generalization to the very hold the cognition that the tasks are
phenomena that the dissonance the- dull and boring. In addition, however,
orists claim to be "entirely closed to the experimental SB have the cognition
the judgmental interpretation and that they have expressed favorable at-
rather unequivocally explainable by the titudes toward the tasks to a fellow
dissonance formulation [Brehm & student. These two cognitions are dis-
Cohen, 1962, p. 111]." sonant for 5s in the $1 condition be-
 188 DARYL J. BEM

cause their overt behavior does not individual to be favorable toward the
"follow from" their cognition about the tasks. If one now places the hypo-
task, nor does it follow from the small thetical observer and the communicator
compensation they are receiving. To into the same skin, the findings ob-
reduce the resulting dissonance pres- tained by Festinger and Carlsmith are
sure, they change their cognition about the result. There is no aversive mo-
the task so that it is consistent with tivational pressure postulated; the de-
their overt behavior: they become more pendent variable is viewed simply as a
favorable toward the tasks. The 6"s in self-judgment based on the available
the $20 condition, however, experience evidence, evidence that includes the
little or no dissonance because engag- apparent controlling variables of the
ing in such behavior "follows from" observed behavior.
the large compensation they are receiv- If this analysis of the findings is cor-
ing. Hence, their final attitude ratings rect, then it should be possible to repli-
do not differ from those of the control cate the inverse functional relation be-
group. tween amount of compensation and the
In contrast with this explanation, the final attitude statement by actually
present analysis views these results as letting an outside observer try to infer
a case of self-perception. Consider the the attitude of an 5" in the original
viewpoint of an outside observer who study. Conceptually, this replicates
hears the individual making favorable the Festinger-Carlsmith experiment
statements about the tasks to a fellow with the single exception that the ob-
student, and who further knows that server and the observed are no longer
the individual was paid $1 ($20) to do the same individual.
so. This hypothetical observer is then
asked to state the actual attitude of the AN INTERPERSONAL REPLICATION OF
individual he has heard. An outside THE FESTINGER-CARLSMITH
observer would almost certainly judge EXPERIMENT
a $20 communicator to be "manding" Seventy-five college undergraduates par-
reinforcement (Skinner, 1957); that ticipated in an experiment designed to "de-
is, his behavior appears to be under the termine how accurately people can judge
control of the reinforcement contingen- another person." Twenty-five 5s each
cies of the money and not at all under served in a $1, a $20, or a control condition.
All 5s listened to a tape recording which
the discriminative control of the tasks described a college sophomore named Bob
he appears to be describing. The $20 Downing, who had participated in an ex-
communicator is not credible in that periment involving two motor tasks. The
his statements cannot be used as a tasks were described in detail, but nonevalu-
atively; the alleged purpose of the experi-
guide for inferring his actual attitudes. ment was also described. At this point,
Hence, the observer could conclude that the control 5s were asked to evaluate Bob's
the individual found such repetitive attitudes toward the tasks. The experimental
tasks dull and boring in spite of what 5s were further told that Bob had ac-
he had said. Although the behavior of cepted an offer of $1 ($20) to go into the
waiting room, tell the next 5 that the tasks
a $1 communicator also has some mand were fun, and to be prepared to do this
properties, an outside observer would again in the future if they needed him. The
be more likely to judge him to be ex- 5s then listened to a brief conversation
pressing his actual attitudes and, hence, which they were told was an actual record-
ing of Bob and the girl who was in the
would infer the communicator's atti- waiting room. Bob was heard to argue
tude from the content of the communi- rather imaginatively that the tasks were fun
cation itself. He would thus judge this and enjoyable, while the girl responded very
 SELF-PERCEPTION 189
little except for the comments that Festin- TABLE 1
ger and Carlsmith's stooge was instructed ATTITUDE RATINGS AND INTERPERSONAL
to make. The recorded conversation was ESTIMATES OF ATTITUDE RATINGS TOWARD
identical for both experimental conditions in THE TASKS FOR EACH CONDITION
order to remain true to the original study in
which no differences in persuasiveness were Experimental condition
found between the $1 and the $20 com-
munications. In sum, the situation attempted Study
to duplicate on tape the sitution actually ex- SI $20
Control compensa- compensation
perienced by Festinger and Carlsmith's 6"s. tion
All 5s estimated Bob's responses to the
same set of questions employed in the origi- Festinger-
nal study. The key question required 5s to Carlsmith -0.45 + 1.35 -0.05
rate the tasks (or for Bob's attitude toward Interpersonal
them) on a scale from —5 to +5, where replication -1.56 +0.52 -1.96
—5 means that the tasks were extremely dull
and boring, +5 means they were extremely Note.—For the Festinger-Carlsmith study, N = 20
interesting and enjoyable, and 0 means they in each condition; for the Interpersonal replication
study, N = 25 in each condition.
were neutral, neither interesting nor un-
interesting.
personal observers to duplicate those
Results of Ss in the original experiment, it is
seen that the two sets of ratings are
Table 1 shows the mean ratings for quite comparable on the 10-point scales.
the key question given by 5s in all Since the above replication was con-
three conditions of both the original ducted, Jones (1966) has reported a
experiment and the present replication. study in which 5s' attitudes and ob-
The results show that in both studies servers' judgments were compared di-
the $1 and control conditions are on rectly in the same experiment. Again,
different sides of the neutral point and the observers' judgments not only rep-
are significantly different from one an- licated the inverse functional relation
other at the .02 level of significance displayed by the attitude statements of
(t = 2.48 in the original study; t — 5s themselves, but the actual scale posi-
2.60 in the replication).2 In both tions of observers and 5s were again
studies, the $1 condition produced sig- similar.
nificantly more favorable ratings to- These successful replications of the
ward the tasks than did the $20 con- functional relation reported by Festin-
dition (t = 2.22, p < .03 in the original ger and Carlsmith provide support for
study; t - 3.52, p < .001 in the repli- the self-perception analysis. The origi-
cation). In neither study is the $20 nal 5s may be viewed as simply making
condition significantly different from self-judgments based on the same kinds
the control condition; and, finally, in of public evidence that the community
neither study were there any significant originally employed in training them to
differences between conditions on the infer the attitudes of any communi-
other questions asked of 5s about the cator, themselves included. It is not
experiment. Thus, the inverse rela- necessary to postulate an aversive moti-
tion between amount of compensation vational drive toward consistency.
and the final attitude rating is clearly These interpersonal replications are
replicated; and, even though the pres- illustrative of others which have been
ent analysis does not require the atti- reported elsewhere (Bern, 1965). It
tude judgments themselves of the inter- has been shown that the present analy-
2
All significance levels in this article are sis applies as well to forced-compliance
based on two-tailed tests. experiments which utilize compensa-
190 DARYL J. BEM

tions much smaller than $20, to studies were actually written was not sig-
which manipulate variables other than nificantly different from the mean of
the amount of compensation, and to attitude ratings obtained after the es-
studies which evoke different behaviors says were written. But the variance
from 5". Alternative dependent vari- across 5s was much greater in the
ables have also been considered. For latter case. That is, actually writing
example, Brehm and Cohen show that the essays increases and decreases the
6"s rating of how hungry he is can be initial effect of volunteering. In ad-
manipulated by inducing him to volun- dition, there was a negative relation-
teer to go without food for different ship between the number of arguments
amounts of compensation (1962, pp. S wrote and the degree to which his
132-137), and a successful interper- final attitude statement agreed with the
sonal replication of that experiment position advocated in the essay. On the
again supports the present self-percep- other hand, Cohen, Brehm, and Flem-
tion analysis of these forced-compliance ing (1958) report a positive relation-
phenomena (Bern, 1965). ship between "original arguments" and
The merits of alternative formula- amount of attitude change, but this re-
tions to an established theory are often lationship appeared in only one of the
sought in their ability to explicate func- experimental conditions. Unpublished
tional relations about which the origi- data from the Festinger-Carlsmith ex-
nal theory must remain mute. Accord- periment show a negative correlation in
ingly, the analysis now turns to a one condition between attitude ratings
pattern of related findings which have and "number and variety" of argu-
not been adequately accommodated by ments and a positive correlation in the
dissonance theory: the observed rela- other (reported by Brehm & Cohen,
tionships between the amount of be- 1962, p. 119). Finally, when 5s them-
havior evoked from S in a forced-com- selves rate the quality of their persua-
pliance setting and his final attitude sive communications, the confusion is
statements. further compounded. Brehm and
A number of forced-compliance ex- Cohen conclude that "the role of dis-
periments have demonstrated that the crepant verbal behavior in the arousal
differential effects of the stimulus and reduction of dissonance remains
manipulations on attitude statements unclear [p. 121]." How might the
can be obtaned even before any of the self-perception analysis treat these
behavior to which the individual has effects ?
committed himself is actually emitted If an outside observer begins with
(Brehm & Cohen, 1962, pp. 115-116). the discrimination that a communicator
That is, the behavior of volunteering to is credible, then the more arguments
emit the behavior is sufficient to con- put forth, the more persuasive the
trol the individual's subsequent self- speaker might well become, if nothing
judgment of attitude. (The self-per- intervenes to change the observer's
ception interpretation of this effect has judgment of the communicator's cred-
also been confirmed by an interpersonal ibility. If, however, the observer dis-
replication, Bern, 1965.) In fact, in criminates the communicator as mand-
an experiment in which 5s volunteered ing reinforcement, then it seems likely
to write essays against their initial that the more insistent the speaker be-
opinions, Rabbie, Brehm, and Cohen comes in pushing his point of view, the
(1959) report that the mean of atti- more it appears to the observer that
tude ratings obtained before the essays he "doth protest too much," and the less
 SELF-PERCEPTION 191

likely it is that the speaker's statements tical for both conditions. All Ss heard
will be taken to express his "actual" the speaker present a fairly imaginative
and lengthy set of reasons as to why he
attitudes. had enjoyed the tasks. For the present ex-
Now consider the self-observer. If tension, a second communication was de-
5"s in the dissonance experiments begin signed, which was somewhat shorter and
with the discrimination that they are contained comparatively unimaginative ar-
not manding (Ss in the low compensa- guments. The replication was then rerun on
an additional SO 5s assigned either to a $1
tion conditions, for example), then the or a $20 condition. The 5s were again
more arguments they put forth, the asked to estimate the actual attitude of the
more self-persuasive they might be- speaker. Thus, except for the length and
come. For any given S, however, variety of arguments in the communication,
this replication is identical with the earlier
presenting a communication counter to one. The total design, then, contains four
his initial position might itself provide experimental groups: $l-long communica-
him with the cues that he is manding tion, $l-short communication, $20-long com-
and hence destroy the initial effect of munication, and $20-short communication.
volunteering under nonmand condi- If the present analysis is correct, then
within the $1 condition, where the communi-
tions; he will become less self-persua- cator is more likely to be perceived as credi-
sive as he continues. This analysis, ble, the long communication should lead
then, leads one to expect the increased interpersonal observers to infer that the
variability in postessay as compared communicator enjoyed the tasks more than
to pre-essay measures of attitude. It the short communication would. Within the
$20 condition, however, the long communi-
is equally clear, however, that to con- cation should be less persuasive than the
firm this analysis, the hypothesized short one; the longer the speaker carries
discrimination of credibility must be on, the harder he appears to be trying to
brought under experimental control earn his $20. He "doth protest too much."
Thus, an interaction effect is predicted be-
rather than being left under the con- tween the two variables of communication
trol of the unique past histories of in- length and amount of compensation. It will
dividual 5"^. To do this, the Fest- be noted that this is equivalent to saying that
inger-Carlsmith experiment is again the "dissonance" effect, the inverse func-
used as an illustrative example. tional relation between compensation and
attitude change, is itself a function of
communication length. The shorter the com-
AN EXTENDED INTERPERSONAL REPLI- munication, the smaller the inverse relation-
CATION OF THE FESTINGER-CARL- ship should become, perhaps even reversing
SMITH EXPERIMENT itself at very short communication lengths.

Festinger and Carlsmith found that Results

within the $1 condition, the greater the
number and variety of arguments Tables 2 and 3 display the results
stated by 6" about the tasks, the more and their analysis, respectively. It will
favorable his final evaluation was of be recalled that scores can range from
them. Within the $20 condition, how- —5 to +5, the higher the score, the
ever, the greater the number and va- more favorable the communicator is
riety of arguments, the less favorable judged to be toward the tasks.
his final rating. The following study It is seen that the interpersonal in-
thus seeks to replicate this pattern terpretation of self-perception is sup-
of results with interpersonal observers. ported by these results. By employing
Method attitude estimates of outside observers,
In the earlier replication, the persuasive the study has replicated Festinger and
communication heard by Ss was iden- Carlsmith's positive correlation between
192 DARYL J. BEM

TABLE 2 onciliation of the conflicting predictions

INTERPERSONAL ESTIMATES OF ATTITUDE made by a reinforcement theory (e.g.,
RATINGS TOWARD THE TASKS Scott, 1957, 1959), "incentive" theories
(Janis & Gilmore, 1965), and disso-
Experimental Long com- Short com- nance theory within a single theoretical
condition munication munication
framework (cf. Bern, 1965).
$1 compensation +0.52 -1.04
$20 compensation -1.96 -0.64 NONVERBAL DEPENDENT VARIABLES
Note.—N = 25 in each cell. Both dissonance theory and the pres-
ent analysis can be characterized as
number of arguments and attitude dealing with cognitions, attitudes, be-
change within the $1 condition and the liefs, or self-descriptive statements as
negative correlation between these two the dependent variables. There is,
variables within the $20 condition. however, a long history of evidence
The main effect of compensation seen demonstrating that an individual's be-
in Tables 2 and 3 is, of course, the liefs and attitudes can function as ante-
primary "dissonance" effect reported cedent or independent variables as well,
earlier. As also noted earlier, however, exercising partial control not only over
the "dissonance" effect is itself a func- his instrumental and consummatory be-
tion of communication length, and the haviors, but over many of his physio-
main effect is due entirely to the in- logical responses as well. Accordingly,
verse relation appearing in the condi- the same "dissonance" manipulations
tions employing the long communica- that can produce attitude change should
tion, the communication designed to be expected to produce changes in some
duplicate these found in the original of these nonverbal behaviors too, an
Festinger-Carlsmith experiment. A expectation that has now been well
nonsignificant reversal actually appears confirmed (e.g., Brehm, Back, & Bog-
when very short communications are donoff, 1964; Brock & Grant, 1963;
employed. It may be that communica- Cohen, Greenbaum, & Mansson, 1963;
tion length is thus one of the confound- Freedman, 1965; Zimbardo, Cohen,
ing parameters responsible for the con- Weisenberg, Dworkin, & Firestone,
flicting findings, including reversals, 1966). It should be clear, however,
reported in forced-compliance experi- that, although either theory could
ments, a point noted by Aronson claim to anticipate these "noncognitive"
(1966). If this is so, then the present effects, neither formulation contains
conceptualization might provide a rec- the theoretical machinery for "explain-
ing" them in any nontrivial sense; this
TABLE 3 is especially true for the striking physi-
ological effects (e.g., Zimbardo et al.,
SUMMARY OF ANALYSIS OF VARIANCE OF
INTERPERSONAL ESTIMATES OF ATTITUDE 1966). In addition, Weick (in press)
RATINGS TOWARD THE TASKS has persuasively argued that the ap-
parent motivational effects of disso-
Source of Sum of Mean F nance manipulations on the intensity
variation squares df square
Long versus Short 0.360 1 0.360 0.05 of nonverbal instrumental behavior
$1 versus $20 27.040 1 27.040 4.07* often fail to show the predicted atti-
Interaction 51.840 1 51.840 7.80**
Within cells 637.920 96 6.645 tude change that should accompany
such effects and that these effects are
*p <.05
** p < .01 more parsimoniously accounted for by
 SELF-PERCEPTION 193
propositions drawn from frustration ratings obtained a week before the ex-
theory and cue-utilization theory. periment. The main displacement ef-
In sum, it is suggested that the inter- fect appeared as predicted: Chosen
personal model of self-perception pro- toys were displaced in the more favor-
vides a viable alternative to the theory able direction; rejected toys were gen-
of cognitive dissonance in accounting erally displaced in the unfavorable di-
for the attitudinal phenomena that have rection. In addition, however, the
emerged from the forced-compliance displacement effect was larger when
experiments. the choice was made from the larger
number of alternatives. This is so,
THE FREE-CHOICE STUDIES according to dissonance theory, be-
cause "the greater the number of al-
In the second major category of data ternatives from which one must choose,
on dissonance theory, an S is per- the more one must give up and conse-
mitted to make a selection from a set quently the greater the magnitude of
of objects or courses of action. The dissonance [p. 373]." Similarly, the dis-
dependent variable is his subsequent placement effect was larger when the
attitude rating of the chosen and re- choice was made from dissimilar rather
jected alternatives. Dissonance the- than similar alternatives because "what
ory reasons that any unfavorable as- one has to give up relative to what one
pects of the chosen alternative and any gains increases [p. 373]," again in-
favorable aspects of the rejected al- creasing the magnitude of the disso-
ternatives provide cognitions that are nance experienced.
dissonant with the cognition that the To interpret these findings within the
individual has chosen as he did. To framework of self-perception, consider
reduce the resulting dissonance pres- an observer trying to estimate a child's
sure, the individual exaggerates the ratings of toys; the observer has not
favorable features of the chosen alter- seen the child engage in any behavior
native and plays down its unfavorable with the toys. Now compare this ob-
aspects. This leads him to enhance his server with one who has just seen the
rating of the chosen alternative. Simi- child select one of the toys as a gift for
lar reasoning predicts that he will lower himself. This comparison parallels, re-
his rating of the rejected alternatives. spectively, the prechoice and the post-
These predictions are confirmed in a choice ratings made by the children
number of studies. (See Brehm & themselves. It seems likely that the
Cohen, 1962, p. 303; see also Festinger, latter observer would displace the esti-
1964). mated ratings of the chosen and re-
A number of secondary predictions jected alternatives further from one
concerning parameters of the choice another simply because he has some be-
have also been confirmed. In an ex- havioral evidence upon which to base
periment by Brehm and Cohen (1959), differential ratings of these toys. This
school children were permitted to se- is the effect displayed in the children's
lect a toy from either two or four final ratings.
alternatives. Some children chose The positive relation between the
from qualitatively similar toys; others number of alternatives and the dis-
chose from qualitatively dissimilar al- placement effect can be similarly ana-
ternatives. The children's postchoice lyzed. If an observer had seen the
ratings of the toys on a set of rating selected toy "win out" over more com-
scales were then compared to initial peting alternatives, it seems reasonable
194 DARYL J. BEM

that he might increase the estimated dents were given a sheet of paper with the
displacement between the "exceptional" following information: "In a psychology ex-
periment, an 11-year-old boy was asked to
toy and the group of rejected alterna- rate how well he liked toys that are typi-
tives. Finally, the fact that the dis- cally popular with this age group. He was
placement effect is larger when the then permitted to select one of these toys
alternatives are dissimilar would ap- to keep for himself. We are interested in
pear to be an instance of simple stimu- how well college students can estimate his
ratings." Each sheet also informed S which
lus generalization. That is, to the ex- toy the child had chosen and from which al-
tent that the chosen and rejected al- ternatives he was permitted to choose. He
ternatives are similar to one another, then made his estimates of the child's ratings.
they will be rated closer together on The 5s were randomly assigned to one of
four conditions corresponding to the com-
a scale by any rater, outside observer, binations of number of alternatives (two or
or the child himself. four) and similarity of alternatives (similar
In sum, if one regards the children or dissimilar).
as observers of their own choice be-
havior and their subsequent ratings as Results
inferences from that behavior, the Table 4 lists the toys employed, the
dissonance findings appear to follow. control group means, and the displace-
The following demonstration illustrates ments from those means of the corre-
the point. sponding experimental group means for
the chosen and rejected alternatives in
AN INTERPERSONAL REPLICATION OF each condition. Scores can range from
THE TOY STUDY 0 to 5, where a higher number indi-
Twenty-four college students served as cates greater liking for the toy; a posi-
control 6"s by estimating how an 11-year-old tive displacement indicates increased
boy might rate several different toys. These
toys were selected from the list reported liking for the toy. To facilitate com-
by Brehm and Cohen (1959) and were rated parisons among conditions, the toy
on the same rating scales. The toys to be rated as most popular by the control
rated in the subsequent experiment were group, the swimming snorkel, was em-
then selected on the basis of these ratings
according to the same criteria of selection ployed as the "chosen" toy in all ex-
employed by the original investigators. perimental conditions. In addition, it
For the experiment itself, 96 college stu- will be noted that it was possible to

TABLE 4
MEAN DISPLACEMENT IN TOY RATINGS FROM CONTROL GROUP MEANS TOR
CHOSEN AND REJECTED ALTERNATIVES IN EACH CONDITION

Similar alternatives Dissimilar alternatives

Experimental condition
Toy Con- Displace- Toy Con- Displace-
trol ment trol ment

Two alterna- Chosen Swimming snorkel 3.45 +.35 Swimming snorkel 3.45 +.22
tives Rejected Swimming mask 3.44 -.39 Archery set 2.79 -.42
Four alterna- Chosen Swimming snorkel 3.45 + .69 Swimming snorkel 3.45 +.99
tives Rejected Swimming mask Archery set
Swimming fins 2.54 -.01 Bowling game 2.58 -.26
Life jacket Ship model

Note.—N = 24 in the control and each experimental condition.

 SELF-PERCEPTION 195

match closely the combined mean rat- sis is confirmed insofar as it is possible
ings of the rejected alternatives in the to test it with these data.
four-alternative conditions; unfortu- Number of alternatives. The pres-
nately this could not be done for the ent experiment is attempting to repli-
two-alternative conditions. cate the positive relation found be-
It is seen in Table 4 that the chosen tween the displacement and the
alternative was rated higher and the number of alternatives. Again, the
rejected alternatives were rated lower displacements of the rejected alterna-
than the corresponding control group tives in the two-alternative conditions
means in every condition. In both of cannot be legitimately incorporated into
the two-alternative conditions, the total the comparison. The present analysis,
displacement effect is significant at the therefore, is confined to the upward
.01 level 0 = 3.66 and 2.81 for the displacement of the chosen alternative.
similar and dissimilar conditions, re- For the four-alternative conditions, the
spectively) ; for both of the four-al- mean upward displacement of the
ternative conditions, it is significant at swimming snorkel is seen to be .84; for
the .001 level (t = 5.26 and 9.18, re- the two-alternative conditions, it is .28.
spectively). Some of the downward This difference is significant at the .01
displacement of the rejected alterna- level (t = 3.29). The dissonance find-
tives in the two-alternative conditions ings are again replicated by interper-
can be attributed to regression effects sonal observers.
since the initial means of these alterna- Although it would have been de-
tives are above the grand mean, but sirable to test the predictions uncon-
this problem has been avoided in the founded by the noncomparability of
four-alternative conditions by combin- the two-alternative base lines, this
ing the ratings of the three rejected would have required departing from
alternatives; in this case the predicted the toys employed in the original ex-
displacement effect is opposite in direc- periment. More importantly, however,
tion to that due to regression, as is the this would still not have yielded a more
upward displacement of the chosen al- direct comparison between the repli-
ternative in all four conditions. Thus, cation and the original experiment be-
the main displacement effect is clearly cause the results reported by Brehm
replicated by interpersonal judgments. and Cohen are themselves confounded
Similarity oj alternatives. From by uncorrected regression effects. The
simple considerations of stimulus gen- present replication actually provides
eralization, it was predicted that the clearer evidence for the predicted ef-
displacement effect should be greater fects than the original study.
in the dissimilar than in the similar It is suggested that this same kind of
conditions. Because of the differential analysis can be applied to the other
effects of regression, mentioned above, studies in this category of dissonance
however, the analysis must be confined experiments. Once again, it seems un-
to the four-alternative conditions where necessary to invoke a motivational con-
it was possible to match the control struct to account for the data.
group means of the rejected alterna-
tives. Within this condition, the mean EXPOSURE-TO-INFORMATION STUDIES
total displacement is .70 in the similar
condition and 1.25 in the dissimilar The third category of dissonance
condition, a difference significance at studies includes two general para-
the .05 level (t = 2.22). The hypothe- digms : experiments in which an indi-
196 DARYL J. BEM

vidual is involuntarily exposed to in- negative attribute, and he can again

formation that is discrepant with use himself as the reference for a stan-
information he already possesses, and dard of "good" and "bad." It would
studies that examine an individual's appear to follow that the more an >9
willingness to expose himself volun- judges himself as "good," the higher
tarily to dissonant information. an arousal number, relative to his own,
An experiment of the first type in he would assign to another person on
which male ,9s received information whom he has no additional data, pre-
discrepant with their "self-images" cisely the relationship reported by
was conducted by Bramel (1962). Bramel.8
Each S was first led to view himself This straightforward "psychophysi-
favorably or unfavorably on a number cal" interpretation is further supported
of personality characteristics and then by the observed relationship between
given information that implied that he ,9's prediction of his partner's arousal
was sexually aroused by homoerotic and his overall evaluation of the part-
pictures. This information was in the ner which he had made prior to re-
form of numerical readings from a ceiving the experimental manipulation.
meter which was supposedly measuring A simple "halo effect" is evident: ,9s
,9's sexual arousal to the stimulus pic- attribute high readings to partners to-
tures ; actually, all ,9s received the same ward whom they had generally un-
meter readings. The dependent vari- favorable attitudes; they attribute
able of the study was S"s prediction of moderate readings to moderately es-
the meter readings that were obtained teemed partners; and, with one excep-
from his "partner," another male 5" tion, they attribute low readings to
who was participating in the experi- highly esteemed partners. This mono-
ment concurrently. tonic relation is violated at only one
Bramel reasons from dissonance the- point: 5s who have high self-esteem
ory that the arousal information would appear to use their own meter reading
be quite dissonant for ,9s who had been as a lower bound of "goodness" and
led to view themselves favorably. In simply assign a similar level of arousal
order to reduce this dissonance, these to partners whom they also regard
.9s would "project" or attribute a favorably. This finding, too, is con-
higher amount of homosexual arousal sistent with the judgmental interpre-
to their partners than would ,9s in the tation of Bramel's data.
Unfavorable condition, who would find A number of other experiments ex-
the information less dissonant with amine the effects of giving ,9s informa-
their self-image. The results are con- tion that is dissonant with their self-
sistent with this prediction. The data images. Unfortunately, the more re-
show that the higher 5"s measured markable effects (e.g., Aronson &
self-esteem, the more arousal he at- Carlsmith, 1962) are apparently diffi-
tributed to his partner. cult to reproduce, and it is still pos-
This study may be reanalyzed by
sible that the original findings are
considering the nature of the task set
,9. He is asked to make a comparative artifactual. (See McGuire, 1966, for
judgment of another person along a a review.)
numerical scale. His only standard of Throughout this discussion, it has
reference is his own meter reading. 8
Bramel briefly considers a similar in-
In addition, he knows that high meter terpretation, but rejects it in favor of the
readings are to be associated with a dissonance interpretation.
 SELF-PERCEPTION 197

been argued that the data under analy- it is suggested, justify the reification of
sis could be accounted for without a new internal drive that is assumed to
postulating an aversive motivational be an inherent characteristic of behav-
drive. The second kind of exposure- ing organisms. In any case, the as-
to-information studies may be viewed sumption of any motivational process
as a direct test for the existence of to account for the data reviewed in this
such a drive. If cognitive dissonance discussion would seem gratuitous.
is, in fact, an aversive state, then a per-
son should avoid exposure to sources SOME METATHEORETICAL CON-
of dissonant information and seek out SIDERATIONS
nondissonant sources. Compared with
the theoretical chain of reasoning be- In the opening remarks, some con-
hind the other studies discussed, this trasts were noted between the con-
deduction from dissonance theory is by ceptual approach typified by dis-
far the most direct, the easiest to test, sonance theory and the behavioral
and the most crucial for justifying a approach represented here by the func-
motivational construct like dissonance. tional analysis of self-perception. It
The available evidence, however, is was pointed out that the behaviorist's
not supportive. In an extensive review goal is to account for observed rela-
of the relevant studies, most of which tions between current stimuli and re-
were conducted by investigators whose sponses in terms of an individual's past
theoretical orientation would lead them training history and a small number
to look for selective exposure to non- of basic functional relations discovered
dissonant information, Freedman and in the experimental analysis of simpler
Sears (1965) conclude that "clearly behaviors. The behaviorist's functional
experimental evidence does not demon- analysis of complex behaviors like dis-
strate that there is a general psycho- sonance phenomena was thus seen to
logical tendency to avoid nonsupportive be based on empirical generalization, a
and to seek out supportive informa- feature which infuses it with an in-
tion." ductive flavor and spirit.
There is, of course, nothing within In contrast, the dissonance theorists
the behaviorist's functional framework clearly prefer the "deductive" nature
that would rule out a motivational phe- of their theory and explicitly derogate
nomenon. For example, it is not im- the "weakness of an empirical general-
plausible to suppose that punishment is ization as compared with a true theo-
often contingent upon being inconsis- retical explanation [Lawrence & Fest-
tent, illogical, or just plain wrong in inger, 1962, p. 17]." This criticism of
our highly verbal culture. This would the behaviorist's functional analysis,
be particularly true for the college namely, that it has no deductive fer-
students who typically serve as 5"s in tility or predictive power, is often ex-
cognitive dissonance experiments. Ac- pressed. The radical behaviorist, so
cordingly, evidence demonstrating that the criticism goes, will not venture a
it is aversive for such 5"s to maintain specific prediction without knowing
incompatible responses in their verbal the complete reinforcement history of
repertoires might well be forthcoming. the organism. He cannot provide a
Such a phenomenon is appropriately "true theoretical explanation."
labeled motivational, but it would be It is suggested here that a functional
the consequence of a particularly com- analysis appears to have limited pre-
mon cultural practice and would not, dictive power only because it makes
198 DARYL J. BEM

explicit the kinds of knowledge about successful in replicating dissonance

the past and present controlling vari- phenomena for the same reason. Dis-
ables that any theorist must have if he sonance theorists and radical behavior-
is to predict behavior accurately. How, ists need the same kinds of knowledge.
for example, do the dissonance theor- Only the behaviorists, however, take as
ists conclude that dissonance is present their explicit obligation the necessity
in a particular case? That is, how do for accounting for both their own and
they decide when one cognition does their 5"s' differential response to such
not "follow from" another? According controlling variables.
to Festinger, "the vagueness in the con- In sum, it is concluded that the
ceptual definition of dissonance-— greater "deductive fertility" of dis-
namely, two elements are dissonant if, sonance theory is largely illusory. In
considered alone, the obverse of one the process of adequately explicating
follows from the other—lies in the the phrase "follows from" in their fun-
words 'follows from'. . . . One ele- damental statement, the dissonance
ment may follow from another because theorists will necessarily have to per-
of logic, because of cultural mores, be- form the explicit functional analysis
cause of things one has experienced they had hoped to finesse. It remains
and learned, and perhaps in other our conviction that the appeal to hypo-
senses too [1957, p. 278]." Five years thetical internal states of the organism
later, Brehm and Cohen note that "the for causal explanations of behavior is
'follows from' relationship can some- often heuristically undesirable. Such
times be determined empirically but is diversion appears only to retard and
limited by our abilities to specify and deflect the thrust of the analysis that is
measure cognitions and the relation- ultimately required.
ships among them . . . the 'follows
from' relationship is not always clear REFERENCES
and specifiable [1962, pp. 11-12]." ARONSON, E. The psychology of insufficient
In actual practice, however, the dis- justification: An analysis of some conflict-
sonance theorists do not experience dif- ing data. In S. Feldman (Ed.), Cognitive
ficulty in inferring the existence of consistency: Motivational antecedents and
behavioral consequents. New York: Aca-
dissonance from their stimulus oper- demic Press, 1966. Pp. 109-133.
ations. But this is so precisely because ARONSON, E., & CAELSMITH, J. M. Per-
in that inference the dissonance the- formance expectancy as a determinant of
orists sneak through the back door the actual performance. Journal of Abnor-
very knowledge they claim to do with- mal and Social Psychology, 1962, 65, 178-
182.
out. It is in that inference that they ASCH, S. E. Review of L. Festinger, A
implicitly make use of the fact that they theory of cognitive dissonance. Contempo-
have been raised by the same social- rary Psychology, 1958, 3, 194-195.
izing community as their 5s. The dis- BEM, D. J. An experimental analysis of
self-persuasion. Journal of Experimental
sonance theorists can thus infer that a Social Psychology, 1965, 1, 199-218.
$1 compensation will produce more BEM, D. J. Inducing belief in false confes-
dissonance than a $20 compensation, sions. Journal of Personality and Social
just as it has been our common history Psychology, 1966, 3, 707-710.
with these same 5s that permits us to BRAMEL, D. A dissonance theory approach
speculate that the difference in com- to defensive projection. Journal of Ab-
normal and Social Psychology, 1962, 64,
pensation represents a difference in 121-129.
the mand properties of the induced be- BREHM, J. W., & COHEN, A. R. Re-evalu-
havior. Interpersonal observers are ation of choice alternatives as a function
 SELF-PERCEPTION 199
of their number and qualitative similarity. sense? Paper read at American Psycho-
Journal of Abnormal and Social Psychol- logical Association, New York, September
ogy, 1959, 58, 373-378. 1966.
BREHM, J. W., & COHEN, A. R. Explora- KELLY, G. Comments on J. Brehm, Motiva-
tions in cognitive dissonance. New York: tional effects of cognitive dissonance. In
Wiley, 1962. M. R. Jones (Ed.), Nebraska symposium
BREHM, M. L., BACK, K. W., & BOGDONOFF, on motivation: 1962. Lincoln: University
M. D. A physiological effect of cognitive of Nebraska Press, 1962. Pp. 78-81.
dissonance under stress and deprivation. KING, B. T., & JANIS, I. L. Comparison of
Journal of Abnormal and Social Psychol- the effectiveness of improvised vs. non-
ogy, 1964, 69, 303-310. improvised role-playing in producing
BROCK, T. C, & GRANT, L. D. Dissonance, opinion change. Human Relations, 1956,
awareness, and motivation. Journal of 9, 177-186.
Abnormal and Social Psychology, 1963, LAWRENCE, D. H., & FESTINGER, L. Deter-
67, S3-60. rents and reinforcement. Stanford: Stan-
BRUNER, J. S. Discussion of L. Festinger, ford University Press, 1962.
The relationship between behavior and LOTT, B. E. Secondary reinforcement and
cognition. In J. S. Bruner (Ed.), Con- effort: Comment on Aronson's "The ef-
temporary approaches to cognition. Cam- fect of effort on the attractiveness of re-
bridge: Harvard University Press, 1957. warded and unrewarded stimuli." Jour-
Pp. 151-156. nal of Abnormal and Social Psychology,
CHAPANIS, N. P., & CHAPANIS, A. Cogni- 1963, 67, 520-522.
tive dissonance: Five years later. Psy-
chological Bulletin, 1964, 61, 1-22. MEAD, G. H. Mind, self, and society. Chi-
COHEN, A. R., BREHM, J. W., & FLEMING, cago: University of Chicago Press, 1934.
W. H. Attitude change and justification McGuiRE, W. J. Attitudes and opinions.
for compliance. Journal of Abnormal and Annual Revieiv of Psychology, 1966, 17,
Social Psychology, 1958, 56, 276-278. 475-514.
COHEN, A. R., GREENBAUM, C. W., & MANS- OSGOOD, C. E. Cognitive dynamics in human
SON, H. H. Commitment to social de- affairs. Public Opinion Quarterly, 1960,
privation and verbal conditioning. Jour- 24, 341-365.
nal of Abnormal and Social Psychology, OSGOOD, C. E., Suci, G. J., & TANNENBAUM,
1963, 67, 410-421. P. H. The measurement of meaning.
FESTINGER, L. A theory of cognitive dis- Urbana: University of Illinois Press,
sonance. Stanford: Stanford University 1957.
Press, 1957. RABBIE, J. M., BREHM, J. W., & COHEN,
FESTINGER, L. Conflict, decision and disso- A. R. Verbalization and reactions to cog-
nance. Stanford: Stanford University nitive dissonance. Journal of Personality,
Press, 1964. 1959, 27, 407-417.
FESTINGER, L., & CARLSMITH, J. M. Cogni- ROSENBERG, M. J. When dissonance fails:
tive consequences of forced compliance. On eliminating evaluation apprehension
Journal of Abnormal and Social Psychol- from attitude measurement. Journal of
ogy, 1959, 58, 203-210. Personality and Social Psychology, 1965,
FREEDMAN, J. L. Long-term behavioral ef- 1, 28-42.
fects of cognitive dissonance. Journal of RYLE, G. The concept of mind. London:
Experimental Social Psychology, 1965, 1, Hutchinson, 1949.
103-120. SCHACHTER, S., & SINGER, J. Cognitive,
FREEDMAN, J. L., & SEARS, D. O. Selective social, and physiological determinants of
exposure. In L. Berkowitz (Ed.), Ad- emotional state. Psychological Review,
vances in experimental social psychology. 1962, 69, 379-399.
Vol. 2. New York: Academic Press, SCOTT, W. A. Attitude change through re-
1965. Pp. 57-97. ward of verbal behavior. Journal of Ab-
JANIS, I. L., & GILMORE, J. B. The influ- normal and Social Psychology, 1957, 55,
ence of incentive conditions on the success 72-75.
of role playing in modifying attitudes. SCOTT, W. A. Attitude change by response
Journal of Personality and Social Psy- reinforcement: Replication and extension.
chology, 1965,1,17-27. Sociometry, 1959, 22, 328-335.
JONES, R. G. Forced compliance dissonance SCEIVEN, M. A study of radical behavior-
predictions: obvious, non-obvious, or non- ism. In H. Feigl & M. Scriven (Eds.),
200 DARYL J. BEM
Minnesota studies in philosophy of science. theory? Organisational Behavior and
Vol. 1. Foundations of science and the Human Performance, in press.
concepts of psychology and psychoanalysis. ZAJONC, R. B. The concepts of balance,
Minneapolis: University of Minnesota congruity, and dissonance. Public Opinion
Press, 1956. Pp. 88-131. Quarterly, 1960, 24, 280-296.
SKINNER, B. F. Science and human be- ZIMBARDO, P. G., COHEN, A. R., WEISEN-
havior. New York: Macmillan, 1953. BERG, M., DWORKIN, L., & FIRESTONE, I.
SKINNER, B. F. Verbal behavior. New Control of pain motivation by cognitive
York: Appleton-Century-Crofts, 1957. dissonance. Science, 1966, 151, 217-219.
WEICK, K. E. Dissonance and task en-
hancement: A problem for compensation (Received July 8, 1966)