Jones N Harris 1967
Jones N Harris 1967
Jones N Harris 1967
EDWARD
E. JONES
AND VICTOR
A. HARRIS
Duke University1
Three experiments were conducted within the framework of correspondent inference theory. I n each of the experiments the
subjects were instructed to estimate the "trueJ' attitude of a
target person after having either read or listened t o a speech by
him expressing opinions on a controversial topic. Independent
variables included position of speech (pro, anti, or equivocal),
choice of position vs. assignment of position, and reference group
of target person. The major hypothesis (which was confirmed with
varying strength in all three experiments) was that choice would
make a greater difference when there was a low prior probability
of someone taking the position expressed in the speech. Other
findings of interest were: (1) a tendency to attribute attitude in
line with behavior, even in no-choice conditions; (2) increased
inter-individual variability in conditions where low probability
opinions were expressed in a constraining context; (3) that this
variability was partly a function of the subjectsJ own attitudes on
the issue; (4) that equivocation in no-choice conditions leads t o
the attribution that the equivocator opposes the assigned position.
The main conclusion suggested is that perceivers do take account
of prior probabilities and situational constraints when attributing
~rivate attitude, but perhaps do not weight these factors as
:avily as would be expected by a rational analysis.
0 1%7 by Academic
Press Inc.
'
"on his own hook." Information gain is lowest when the person expresses
highly conventional opinions under conditions where it would be extremely difficult for him to express any other opinions.
An alternative way of stating the relationship between perceived choice,
prior probability, and correspondence conceives of the first two as orthogonal independent variables and the latter as the dependent variable. The
crucial hypothesis to be tested by the experiments to be presented below
is: When a person expresses a modal (high probability) opinion, attribution of underlying attitude will not vary as a function of perceived choice;
when an unexpected or unpopular opinion is expressed, correspondent
attribution will vary directly with the amount of choice perceived. We
are predicting, then, a particular kind of statistical interaction between
the amount of perceived choice and the prior probability of an opinion's
being expressed, in determining the correspondence of resulting inferences
about opinion-related attitudes.
This hypothesis is hardly paradoxical, though it does propose that
persons perform a kind of implicit information theory calculus in making
sense out of the behavior of others. The information contained in a
statement (act) goes up as the prior probability of the statement goes
down, assuming that the speaker was not forced, bribed, or otherwise
constrained to make the statement. But does the average perceiver in fact
follow this prescription? Relevant experimental evidence is sparse. Steiner
and Field (1960) conducted an experiment in which an accomplice expressed pro-segregation opinions in a group discussion. I n some cases he
chose this role with apparent freedom, in other cases he was assigned the
role by the experimenter. Other group members were more confident in
their assignment of pro-segregation beliefs to the accomplice in the choice
condition. They also liked the accomplice less, probably because the
subjects were Northern college students who themselves were against
segregation. In effect the Steiner and Field (1960) study tests half of the
proposed hypothesis, whereas the present experiments attempt to show
that choice is not an important variable when the prior probability of the
behavior being observed is high. A number of other studies (e.g., Jones,
1965; Bem, 1965; Thibaut and Riecken, 1955; Jones, Davis, and Gergen,
1961) are indirectly relevant in assessing the role of perceived choice in
correspondent attribution. They do not, however, provide a direct test of
the theory.
Such a test requires a design in which a stimulus person states either
;In expected or an unexpected opinion under conditions of high- versus
lorn-perceived choice. Three closely related experiments were conducted
within this general design; the subjects' primary task in each was to
estimate the true attitude of the person making the statement.
J O N E S AND HARRIS
EXPERIMENT I
Method
I n the first experiment subjects were asked t o read a short essay on "Castro's
Cuba" and to record their estimates of the essayist's true attitude toward Castro.
The essay itself was either pro-Castro or anti-Castro; i t had either been written
under conditions of free choice or by assignment from a course instructor. The two
variables of choice and behavior direction thus comprised a two-by-two factorial
design. This basic design was repeated in each of the three experiments, though
there were variations from experiment to experiment in the manner in which the
cross-cutting variables were manipulated.
Subjects
Thirty-six male and 15 female students served in the experiment in one group
session. They were volunteers from the introductory psychology course a t Duke
University who received course credit for their participation.
Procedure
Each subject was handed a mimeographed pamphlet that contained an essay on
Castro's Cuba, a prior statement manipulating the choice variable, and a final
questionnaire. The experimenter then explained the purpose of the experiment as
"an attempt to determine if people can make valid judgments of another's personality
and attitudes on the basis of very limited information." The subjects were led to
believe that a variety of personal materials written by the same undergraduate
student were distributed among them. They were told, "Some of you have an excerpt
from the person's autobiography, which was originally written to accompany his
college application. Others have in your pamphlet a short essay prepared for a
creative writing course. The essay deals with conflicting values in contemporary
society. . . . The remainder of you have an answer from a political science hour
exam." The subjects were then told to glance a t the material to identify their
condition. They were led to believe that the conditions would be compared to see
which kind of written material produced the most valid impression, as measured by
"a lot of additional information that you do not know about." The experimenter
went on t o state that other target persons would be evaluated by other subjects.
Actually, each subject was in the "political science exam" condition. His task was
to read the examination answer and attempt to judge the true attitude of the
"target person" toward the topic. The experimenter concluded his orienting overview with some brief remarks identifying the author of the materials as a student
a t the University of North Carolina, a resident of the state, and the son of an
automobile salesman. The mimeographed material began with a reproduction of the
exam question. This instructed the target person in one of three ways: (a) "Based
on the past week's discussion and lectures, write a short cogent criticism of Castro'e
Cuba as if you were giving the opening statement in a debate"; (b) l1 . . . short
cogent defense of Castro's Cuba as if . . . "; (c) " . . . short cogent essay either
defending or criticizing Castro's Cuba as if. . . ." This constituted the choice
manipulation, with subjects in conditions where the target person received either
(a) or (b) instructions considered as "no choice" subjects.
The essay that followed was approximately 200 words. I t was either pro- or
anti-Castro, and in the no choice conditions the direction of the essay was always
that called for by the examination question. Although the essay .was typed, there
sere occasional spelling errors, false starts, and cross-outs. The experimenter explained that this would provide information about the "style and approach" of the
target person. The essay itself was neither polished nor crude; it had a C+ quality,
crnbodying a few reasonable and familiar arguments. For example:
pro-Castro e s s a y
"
anti-Castro e s s a y
"Castro c a n a n d d o e s a t t e m p t t o t a k e o v e r o u r n e i g h b o r s a n d c o n v e r t
them to c o m m u n i s t s a t t e l i t e s by u s i n g m e t h o d s of i n f i l t r a t i o n s a b o t a g e
and subversion.
"
Response measures
The essay was followed in the booklet by (a) a 12-item semantic differential scale
for rating various personal qualities of the target person; (b) a 10-item Likert-type
scale for the subject to use in estimating the target person's true attitude toward
Castro; and (c) a second copy of the same scale for the subject t o record his own
sttitutle.
Results
JONES A N D HARRIS
TABLE 1
EXPERIMENT
I: MEANS=AND VARIANCES
FOR ATTRIBUTED
ATI~TUDE
SCORES
Speech direction
Pro-caslro
Choice
g
No Choice
sZ
1V
Lf
s2
Pditr
13
59.62
13.59
10
44.10
147.65
< .0lb
Anti-Castro
13
17.38
8.92
15
22.87
17.55
< .01
Pdiff
< .001
< .0Olb
a Possible range from 10 (extreme anti) to 70 (extreme pro). The average subject's
"own score" was 32.23, s2 = 35.54.
b Degrees of freedom, and therefore probability values, adjusted for unequal population variances.
more indirect and permissive toward differing viewpoints than the antiCastro items, the scale has no true neutral point.
Table 1 presents the means and standard deviations of the subjects'
prediction ratings. A number of things are clear by inspection. The
direction of the speech was of great importance in guiding the prediction
ratings. This was especially true in the choice conditions, as one would
expect, but it was also true in the no choice conditions to a highly
significant extent ( t = 5.32, df 10.76, p < .001). It is also clear that
within each speech condition, correspondence between speech direction
and attributed attitude was greater when the speaker had choice than
when the essay direction was assigned (p < .O1 in both cases). An equally
obvious fact is that the variances are heterogeneous, that of the No
Choice-Pro condition being almost 10 times as large as the next largest
variance. Because of this heterogeneity, all the statistical comparisons
mentioned above were evaluated with the reduced degrees of freedom
called for when t tests are performed under an assumption of unequal
population variances (cf. Walker and Lev, 1953, pp. 157-158).
The inflated variance in the No Choice-Pro condition is of interest in
its own right, for i t is precisely in this cell that the largest variance might
be expected. From the subjects' perspective, this should be the condition
of greatest ambiguity: the behavior direction is a t variance with population expectations, but the target person was told to behave that way.
What does he really believe? Apparently some subjects put more weight,
on the behavior direction than the context, others discounted the behavior
almost entirely because of the context.
The main hypothesis was that the difference between Choice and No
r-
Choice means would be greater for subjects hearing the pro-Castro speech
than for those hearing the anti-Castro speech. While the mean differences
in the pro-Castro condition were larger, both Choice-No Choice comparisons were significant (as noted above), and the predicted difference
between differences did not approach significance. The hypothesis was
not confirmed.
Returning to the inflated variance in the No Choice-Pro condition,
one might wonder whether the subjects' own attitudes toward Castro
affected their predictions of the target person's true attitudes. There is
some slight evidence that this was so. The correlation between own and
imputed attitude was +.50 in the No Choice-Pro condition; in the other
conditions the comparable correlations ranged from -.I2 to +.05. Because of the small N's involved, none of these correlations differ significantly from zero.
The semantic differential ratings revealed little information of interest.
There were twelve 7-point scales involving such antonyms as bad-good,
worthless-valuable, weak-strong. Scores were totaled for each subject to
provide a measure of favorability of trait attribution. There was a significant tendency (p < .05) for subjects in the No Choice conditions to
feel more positively toward the target person than those in the Choice
conditions. This may be understood as a sympathy reaction to a student
forced to take a particular side on a touchy issue in writing an examination essay.
Perhaps the most striking result of the first experiment was the
tendency to attribute correspondence between behavior and private attitude even when the direction of the essay was assigned. If the subjects
fully understood the conditions under which the essay was written, their
tendency to be affected by the essay content in attributing an attitude to
the target person would seem to reflect incomplete or distorted reasoning
in the No Choice conditions. Perhaps some of the subjects were inattentive
and did not clearly understand the context of choice or no choice in which
the exam essay was written. Perhaps some felt that the assignment to
write an exam essay on Cuba was unlikely and were skeptical of the cover
story. In order to check on these possibilities, a second experiment was
conducted.
The main hypothesis of the first experiment was not confirmed a t least
in part because the choice variation affected attitude attribution more
than was expected in the anti-Castro essay conditions. I n spite of the fact
that subjects in the four conditions had comparable "own" attitudes, as
n~easuredby a simple one-way analysis of variance, those in the ChoiceAnti condition did have the most pro-Castro attitudes. The mean "ol~m
score" in that condition was greater by t test than the means in each of
Method
The second experiment included a modified replication of the four basic conditions
of the first experiment plus eight additional conditions. Before describing the new
conditions, the procedural modifications of the basic replication will be described.
Added Conditions
The results of the first experiment raised a number of questions concerning the
choice manipulation and how i t was perceived by the subjects. I n the added conditions, an attempt was made to increase the salience of the choice manipulation by
requiring the subject himself to write a pro- or anti-Castro essay before reading
the target person's. In condition 5 (the first four replicated the basic design, as noted
above)'subjects were instructed to write a pro-Castro debate opening speech and
then ,were exposed to a No Choice-Anti target person; in condition 6 subjects were
instructed to write an anti-Castro speech and were then exposed to a No ChoiceAnti target person; and in condition 7 subjects were instructed to write a pro-Castro
speech and mere then exposed to a No Choice-Pro target person. These additions
r
THE ATTRIBUTION O F A'ITITUDES
represented a compromise in design, since the major purpose was to make the
subjects in the No Choice conditions aware of how it feels to be assigned a particular
side of the debate, and the remaining combinations of subject- and target personinstructions were not expected t o be equally informative.
Five additional conditions were added in which, regardless of what the target
person was instructed to do, the resulting speech was ambivalent. It combined both
pro- and anti-Castro arguments in a balanced presentation. Pre-instructions t o the
subject and to the target person mere varied to generate these five conditions.
The full design of Exp. I1 is presented in Table 2, along with the means and N's for
each of the twelve conditions. The ambivalent speech conditions were included to
explore the attribution of attitude when behavior is out of line with role prescriptions. Presumably someone who is instructed to write a pro-Castro debate opener,
but in fact writes an ambivalent statement, should be seen as strongly anti-Castro.
The obvelse should also be true: an ambivalent statement following anti-Castro
instructions should lead to predictions of pro-Castro attitude.
Administration of Experiment
Results
The subjects' own attitudes were again predominantly anti-Castro, and
this time the mean "own scores" in each condition were almost identical.
There is no cluestion that the subjects were attentive to the choice
manipulation. When asked how much choice the target person perceived
bc had (on a 9-point scale) there was very little overlap between the
Choice and the No Choice distributions ( t = 8.09, p < .001). Whether or
not the subject himself was instructed to write an essay (in the choicesalience conditions) did not affect the attribution of choice-perceived-bytarget-person.
Turning to the major dependent variable, attribution of attitude toward Castro, the mean attribution scores for all 12 conditions are presented in Table 2. Comparing conditions 5 and 6 with condition 1, and
condition 7 with condition 2, it is clear that there were no systematic
effects of the choice-salience manipulation. Since this was the case, all
No Choice-Anti conditions were combined, and all No Choice-Pro conditions were combined, to produce the data presented in Table 3. These
data may be compared directly with those presented in Table 1. (The
tittribution data for the ambiguous speech conditions will be deferred for
later consideration.)
Once again there were striking effects for direction of speech. These
effects were significantly greater when the target person was given a
-"-
lo
J O N E S AND HARRIS
TABLE 2
DESIGNOF EXPERIMENT
I1
Instructions to
target
person
Target
person's
performance
NCA
NCP
C
C
anti
Pro
anti
Pro
21.78
38.57
22.89
57.67
Pro
anti
Pro
NCA
NCA
NCP
anti
anti
Pro
25.89
23.56
43.75
pro
anti
Pro
Pro
C
NCA
NC A
NCP
C
C
amb
amb
amb
arnb
arnb
42.00
39.33
33.00
40.00
42.86
Total A'
Instructions to
subject
A. Basic conditions
1. NCA
2. NCP
3. CA
4. C P
B. Salience-of-choice conditions
5. Pro-NCA
6. Anti-NCA
7. Pro-NCP
C. Salience-ambivalent conditions
8. Pro-NCA-a
9. Anti-NCA-a
10. Pro-NCP-a
11. Pro-C-a
12. GC-a
.Y
Note. C stands for free choice, NC for assignment by the debate captain, A for antiCastro speech, P for pro-Castro speech, and a for ambivalent speech.
choice, but the pro-anti difference was highly significant even when the
debate side was assigned. Once again the variances were heterogeneous,
and since the greatest variability recurred in'the No Choice-Pro condition
this replicated the pattern of variances in Experiment I. The mean values
seem to provide strong support for the main hypothesis that degree of
TABLE 3
EXPERIMENT
11: MEANSOAND VARIANCES
FOR ATTRIBUTED
ATTITUDE SCORES
Speech direction
Pro-Castro
Choice
s2
No Choice
2
82
'pdirf
9
57.67
21.00
15
41.33
134.81
< ,001"
Anti-Cast,ro
9
22.89
34.86
27
23.74
:50.12
n.s.
Pdifr
< ,001
< .O m h
a Possible range from 10 (extreme anti) to 70 (extreme pro). The average subject's
"own score" was 31.67, s2 = 82.25.
Degrees of freedom (and therefore probability values) adjusted for unequal population variances.
choice makes a greater difference when the behavior has a low prior
probability. An interaction t test, adjusting the degrees of freedom to
compensate for heterogeneity of variance, is significant ( t = 3.37, df =
36.1, p < .01). This test compares the means in the following way:
(57.6741.33) > (23.74-22.89), cf. Table 3, and thus appropriately ignores the direction of the differences in each column. Otherwise the
interaction test would merely show that the means are further apart in
the first row than in the second row, or t h a t the differences between
speeches is greater when choice is allowed. As we have already noted
above, this form of the interaction is also significant, linking degree of
correspondence of inference directly t o degree of choice.
Eflccts of making an ambivalent speech. When a target person is directed to make an anti-Castro presentation and equivocates in his
argumentation, he should be seen as relatively pro-Castro. The same
ambivalent speech under pro-Castro directions should result in the attribution of an anti-Castro attitude. The results of conditions 8, 9, and
10 bear out this hypothesis (see Table 2). When conditions 8 and 9 are
combined (since both require the target person to give an anti-Castro
speech) and contrasted with condition 10, the difference is significant
(t = 2.07, df = 21, p < .05). The target person who gave an ambivalent
speech under free choice conditions was seen t o be rather in favor of
Castro, as one would expect. I n spite of the differences noted, the effects
on attitude attribution of having the target person violate instructions are
far from overwhelming. The target person who gave an ambivalent speech
under anti-Castro instructions was seen as no more in favor of Castro
than the one who gave a pro-Castro speech under pro-Castro instruction+.
The target person who responded ambivalently under pro-Castro instructions was seen as much more in favor of Castro than the target person
mho slavishly followed anti-Castro directions. This seems to be further
evidence that the average subject in these experiments attaches insufficient
weight to the constraining force of authoritative directions to behave in a
certain way.
Own and attributed attitude. Table 4 presents the pattern of correlations between the subjects' own attitudes toward Castro and the attitudes
imputed to the target person. There are several points of interest in this
table, though conclusions drawn from correlations with such small N's are
obviously risky. First of all, there is a dramatic replication of the
correlation between own attitude and imputed attitude in the No ChoicePro condition. Apparently we may venture the conclusion that when a
subject attempts to predict attitude in a situation with conflicting cues
(where the behavior tells him one thing and the context tells him another), he tends to fall back on his own attitudes as a guide for his
c~timate.Note, Iiowerer, that this was only true in the no-salience
J O N E S AND HARRIS
TABLE 4
EXPERIMENT
11: RELATIONSHIP
BETWEEN OWNATTITUDE
TOWARD CASTRO
AND IMPUTED
ATTITUDE
(Product Moment Correlations)
Speech direction
Pro-Castro Anti-Castro
Ambivalent
Choice
Instructions
No ChoiceNo Salience
No ChoiceSalience
N
T
IV
1'
'i
.93**
8
- .51
.37
18
.33
Pro
Anti
16
- .43*
.64
condition. When the prediction task was preceded by the task of writing
a pro-Castro speech under directions, the correlation vanishes. Having to
write a speech against one's own position seems to reduce the significance
of that position when it comes to imputing the attitude of a target person
operating under the same prescription. Perhaps the subjects in the No
Choice-Salience condition were alerted to concentrate more on the speech
itself, being sensitive to any nuances or signs of sincerity in the arguments
presented.
The other intriguing feature of the correlational data is the tendency
for own position to be negatively related to imputed position when an
ambiguous speech was given, unless the instructions were to give a
pro-Castro speech. I n order to account for this we present the following
post hoc speculation. The target person in the Pro-Ambivalent condition
does not follow instructions slavishly and is therefore seen as quite antiCastro on the average. Perhaps this is what the anti-Castro subject thinks
he might do in the same circumstances, thus causing him to assimilate
Ghe target person's attitude to his own. The target person who writes an
ambivalent speech when told to write an anti-Castro speech, or told to
choose one side or the other, reveals himself as moderately pro-Castro.
Here there seems to be a contrast effect: the more anti-Castro the subject,
the more he judges the speech (and therefore the debater) to be proCastro. These speculations are based on rather complex assumptions, but
the reasoning is compatible with the judgmental theory of Sherif and
Hovland (1961) further elaborated by Berkowitz (1960). This theory
proposes that opinion positions close to one's own will be judged as closer
than they are; the extremity of more distant positions will be exaggerated.
The sema.ntic differential results were essentially negative. The tend-
THE ATTRIBUTION
OF ATTITUDES
13
14
.TONES
AND
HARRIS
15
The entire tape recording, including the speech itself, was a carefully scripted
playlet in which the target pelson presented himself as a junior political science
major from either Sanderwrille, Georgia or from New Brunswick, New Jersey. The
target person's accent was accordingly very southern or vely northeastern.' At the
cnd of the speech, which either favored or opposed segregation, the experimenter
(on the tape) asked the target person to sign a voucher for the $2.50, explained that
the tapes would be used as stimulus materials in an attitude perception project a t
Duke, and secured his permission to use the recording for this purpose without ever,
of course, revealing his identity.
The qeeches. The scripted speeches were as comparable as possible except for
their direction and conclusion. They were constructed so that the same "facts" were
discussetl from different perspectives. For example:
It is a stardard argument of the "dogooders" and others that the reason for
the Negro's weaknesses is due not to
innate factors but rather to environmental conditions. . . . This argument
. . . is completely fallacious.
Negroes, while contributing far less
than their share to the development of
America, have contributed far more than
their share to crime statistics and welfare ease lists.
Etc.
16
subsets B and C that the average subject used in the same manner as the A items.
For all subjects combined, the attribution of segregationist attitude on A correlated
with attribution in the same direction on B, r = .90, df = 120; A correlated with C,
r = .85;B with C, r = .81 (all p-values <.001).
An additional reason for constructing the scale in three varts was the vossibilitv
of testing a rather subtle hypothesis concerning behavioral departures from reference
group norms. How do subjects perceive a Southerner who chooses t o construct an
integrationist speech, or a Northerner who chooses to act like a segregationist?
Perhaps a more coherent attitude structure is assigned to such a "maverick" because
his position differs from the expectations of his community of origin and does not
simply reflect reference group norms. The Northerner who espouses the cause of
integration may be judged to be a "knee-jerk liberal" who passively reflects reference
group norms without fighting through to a coherent attitudinal integration. The same
kind of conclusion ("knee-jerk racist"?) may also be applied to the Southern
segregationist. These considerations suggest the following hypothesis: (1) the
"maverick" who chooses t o differ with the assumed position of his community
reference group will be perceived t o have a more correspondent attitude than one
who chooses to stand with his reference group; thus the Southern integrationist
will be seen as more pro-integration than the Northern integrationist and the
Northern segregationist will be seen as more in favor of segregation than his
Southern counterpart. (2) There will be a higher correlation among the subsets of
the attitude scale for "mavericks" than for "knee-jerk liberalsJ' or "knee-jerk
racists." I n other words, if the "maverick" is seen as anti-segregation he will also be
seen as basically liberal; if he is seen as pro-segregation he will also be seen as
basically conservative.
Postexperimental questionnaire. After the subject recorded his at.titude predictions
("please try t o predict how the target person would honestly respond t o the
following statements"), he was asked to indicate his own attitude on the same
15-item scale. Finally, each subject answered seven questions in the form of 9-point
scales, each designed to check on some aspect of the subject's perceptions of the
experiment.
Results
Once again, the subjects were very attentive to the choice manipulation.
When asked how much choice the target person perceived he had (on a
9-point scale), there was very little overlap between the Choice and the
No Choice distributions ( t = 10.63, p < ,001).
Attribution of Attitude toward Segregation
As noted above, the attitude scale for measuring attribution was constructed of three subsets of items varying in degree of remoteness from
the arguments specifically mentioned in the speech. As might have been
expected from the high overall correlations among these subsets (see
above), attribution to each subset was highly comparable within experimental conditions. This can be verified both by close inspection of Table 5
and by the fact that significance levels for all statistical tests are roughly
the same whether they refer to subset A, B, C, or the combined total.
17
*$f)
AbQ*b~k
18
JONES
A N D HARRIS
TABLE 5
EXPERIMENT
111: MEANS-A N D VARIANCES
FOR ATTRIBUTED
ATTITUDESCORES
No Choice
x
a2
Anti-segregation
Tot,al
Total
N = 14
= 11
Choice
Anti-segregation
Total
Total
a2
s2
N = 16
No Choice
N = 15
= 15
A' = 13
ing speculation may be found in the actual attitude scale scores of subjects from the South. The home states of 100 subjects could be readily
ascertained from the Duke student directory. Fifty-seven of these were
from southern states, while 43 were from the North. The average scale
score of northern subjects ( 2= 80.42) was significantly higher than
= 70.07; tdiff= 3.49, p < .001). The
that of the average Southerner (.8
southern subject was, not surprisingly, more in favor of segregation. Of
greater interest for our present argument, however, is the greater variability among Southerners ( s Z = 260.42) than Northerners ( s 2 = 166.01)
-significant at the .06 level. Thus, being a college student from a southern
town is not a very informative objective indicator of attitude toward
segregation. The attribution results suggest that the subjects were aware
19
of attitudinal variability among southern college students (from a neighboring university) and were especially responsive to the evidence that the
target person had chosen his speech. The No Choice Southerner is an
ambiguous stimulus object, a conclusion that is supported by the fact
that subjects were more variable in attributing 'attitudes to this target
person than to others. It may be noted in Table 4 that the two largest
variances (total score columns) appear in the two No Choice Sout,herner
conditions.
O w n and attributed attitude. Although subjects were assigned to conditions through their own initiative in signing up for particular times,
there were some rather peculiar variations among subjects exposed to the
lionthern target person. Those in the Choice Pro and No Choice Anti
conditions were significantly more against segregation than those in the
No Choice Pro and Choice Anti conditions. No such pattern emerged in
subjects exposed to the northern target person. However, among the latter
subjects, those hearing the anti-segregation speech ended up significantly
Inore opposed to segregation than those hearing the pro-segregation
speech. Since the subjects filled out the own-attitude questionnaire after
hearing the speech and after making their attribution ratings (as was the
case in the two preceding experiments as well) i t is impossible t o estimate
the extent to which they may have been influenced by the speech. The
pattern of "own attitude" scores is reproduced in Table 6 but we demur
from any attempt to interpret the differences between conditions.
.Also presented in Table 6 are the correlations between own and attributed attitude for each experimental condition. Ea.ch correlation is
positive, suggesting a general tendency to assimilate the target person's
attitude to one's own. This tendency reaches significance in only three
conditions, all involving the northern target person. I n the two previous
espcriment~s,the highest correlation between own a.nd attributed attitude
occurred in the condition where no choice was combined with a pro-Castro
speech (low prior probability behavior). The present correlational results
neither replica,te closely nor disconfirm the proposition that the correlation is highest when ambiguous or conflicting information is presented.
The correlation is high for subjects in the Northern-No Choice-Pro condition, but it is also high in two other Northern conditions. The correlations
smong subjects exposed to the Southern speaker are low and non~ignificant.
Perception o j the " ~ ~ m v w i c k ' s attitude
"
st~zc.ctw.e.We proposed in introducing this experiment that correspondence of attribution would be
especially high in the case of the "maverick" who chooses to differ with
the assumed position of his community reference group. A second hypothesis, developed from the same reasoning, was that the intercorrela-
20
TABLE 6
EXPERIMENT
111: SUBJECTS'
OWNATTITUDES
TOWARD SEGREGATION
(x), AND
CORRELATIONS
BETWEEN OWNAND A'ITRIBUTED
ATTITUDE(T),BY CONDITIONS
A. Northern target person
Speech direction
Choice
N o Choice
69.70
64.91
.09
.60*
20
11
76.70
78.36
.49*
.60*
14
.24
.07
15
13
1,
Choice
No Choice
86.50
69.60
.06
.38
16
15
71.53
84.15
-
-~
Note. Significant comparisons across conditions. Northern target person: Anti versus
Pro, F = 6.67, p < .05; Southern target person: interaction P = 13.78, p < ,001;
Northern versus Southern:F = 3.87, p < .lo.
* p < .05.
tions among item subsets would be higher for "mavericks" than for those
who choose the expected speech direction, the "knee-jerk liberals" and
"knee-jerk racists." We have already noted that there is no support for the
first of these hypotheses. The Southerner who chooses to give a prointegration speech is not seen as more in favor of integration than the
Northerner who chooses pro-integration; the Northerner who chooses prosegregation is not seen as more of a segregationist than his Soutlierll
counterpart. Perhaps the reasoning behind this hypothesis confused extremity with certitude. The pro-integration Southerner may be more
confident that he is right, without being more extreme in his dedication to
integration. There is no evidence in the present study that would support
this alternative, however.
The second hypothesis does receive some support. Table 7 presents the
intercorrelstions between item subsets for each experimental condition.
These correlations are generally positive and, as we have already noted,
the three over-all correlations are very high. Those correlations especially
involved in the second hypothesis are italicized in the table, and the
probabilities of the observed differences are indicated. The correlations between subsets for the "mavericks" (southern-choice-A&, and
TABLE 7
EXPERIMENT
111: CORRELATIONS
BETWEEN ITEMSUBSETS,
BY CONDITIONS
ATTRIBUTED
A ~ T U DSCORES,
E
Condition
Ar
A-B
B-C
-4-C
NNCA
SNCA
NNCP
SNCP
SCA'
NCA
SCP
NCP
Tot,al
"Mavericks."
* p <.05
* * p <.01
*** 7) < ,001
J O S E S A N D I-IARHIS
ISSUECASTRO'S CUM
EXPERIMENT I
I EXPERIMENT I[
ISSUE SEGREGATION
PRO
ANTI
P
2
G C I C N C
1a
I
I
NORTHERNER
ALTI
CNC
I SOUTHERNER
EXPERIMENTm
23
this describes the results without really explaining them. We have already
wondered (in the introduction to Exp. 111) whether the amount of choice
perceived in the no choice conditions is enough to have significant effects.
Obviously, the target persons did have the ultimate option to refuse their
~nstructorin Exp. I, their debate captain in Exp. 11, or the dormitory
visitor in Exp. 111. Nevertheless, it seems fair to assume that each subject himself would agree to express false opinions under comparable
circumstances of authoritative assignment, and the subjects' postexperimental ratings of choice indicated their awareness of strong external
constraints on the target person's behavior.
An important area of choice does remain, however, even in the no
choice condition. This is the choice between various ways of expressing
the directed opinion. The arguments advanced in each essay or speech
mere not specified in detail by the constraining authority; an unknown
degree of freedom to select and organize arguments remained even in the
third experiment. I n planning the present experiments, we assumed that
this ~ninimalambiguity was necessary to bring out the specific interaction
that was the test of our major hypothesis. If the target person had been
niercly handed a speech to read under very strong external constraints,
liis co~npliancewould have conveyed little or no information to the
subjects. Under these circumstances we would have expected no differences between No Choice-Pro and No Choice-Anti conditions. Were the
constraints to extend to every detail and facet of an observed per[ormance, the prediction that attribution is uninfluenced by performance
\\.auld be a trivial one. The present experiments show that when the
niajor decisions about the direction and form of behavior are made for
tlic target person, his performance is still a powerful source of variation
In the attribution results. Short of some extreme degree of specification,
hehavior does engulf the field and it is difficult for the perceiver to assign
appropriate weights to the situational context.
We are led to conclude that correspondence in attributing underlying
i~ttitudesto account for expressed opinions is high when the opinions are
unexpected and expressed in a context of free choice. However, the content and direction of the opinions exert a clear inference on attribution
cven when choice is drastically reduced. I n a context that permits the
target person some very minimal degree of spontaneity, the perceiver
qeems to view his performance as more informative than a rational
analysis of act and context mould suggest. This bias may have important
implications for interpersonal relations, and we might propose a hypothesis
for further research that distortion, in the form of assigning too much
significance to performance, increases as the objective constraints on a
target person's actions increase.
24
REFERENCES
BEM, D. J. An experimental analysis of self-persuasion. Journal o f Experimental
Social Psychology, 1965, 1, 19S218.
B E R K ~ ~ I TL.Z ,The judgmental process in personality functioning. Psychologicnl
Review, 1960, 67, 130-142.
BERLYNE,
D. E. Structure and direction i n thinking. New York: Wiley, 1965.
HEIDER,F. Social perception and phenomenal causality. Psychological Review, 1944,
51, 358-374.