Journal oj Consulting Psychology
1967, Vol. 31, No. 3, 248-252
INFERENCE OF ATTITUDES FROM NONVERBAL
COMMUNICATION IN TWO CHANNELS x
ALBERT MEHRABIAN ANDSUSAN R. FERRIS
University of California, Los Angeles
In the present study 3 degrees of attitude (i.e., positive, neutral, and negative)
in facial expression were each combined with 3 degrees of attitude communicated
vocally. The vocal communications of attitude were superimposed on
a neutral word. In preparing the 2-component communications, the components
were selected so that the degree of positive attitude communicated facially
was equivalent to that communicated vocally—that is, the independent effects
of the 2 components were comparable. It was found that attitudes inferred
from combined facial-vocal communications are a linear function of the
attitudes communicated in each component, with the facial component
receiving approximately 3/2 the weight received by the vocal component.
Implications of the findings for more general attitude-communication problems
are discussed.
While there are many studies of nonverbal
attitude or feeling communication in single
channels (e.g., reviews by Davitz, 1964 or
Mahl & Schulze, 1964), investigators are only
beginning to explore simultaneously transmitted
feelings or attitudes in two or more
channels. Gates' (1927) investigation of
single-channel decoding of facial and vocal
stimuli is relevant to the present study. She
found that children are more accurate in
their judgments of facial compared to vocal
expressions of feeling. Unfortunately, her
method only allows a tentative conclusion
that discrimination of feeling on the basis of
facial cues is easier than discrimination of
feeling on the basis of vocal cues. There is,
however, some corroboration of Gates' findings
in a study by Levitt (1964). Communicators
were filmed as they attempted to
communicate six emotions facially and vocally,
using neutral verbal materials. The decoding
of facial and vocal stimuli in combination
was only as accurate as the decoding of
facial stimuli alone, and both conditions were
more accurate than the decoding of vocal
stimuli alone. This finding can be interpreted
to indicate that in a two-channel facial-vocal
1
This study was supported in part by United
States Public Health Service Grant 1 R03-MH-
12629-01 and in part by University of California
Grant No. 2189. The authors wish to thank Dee
Nieto, Nina Spitzer, and Barbara Ullman, who acted
as models, and Elaine Blackman and Katherine Nelson,
who, along with one of the authors, acted as
speakers in preparing the experimental materials.
communication of emotion, the facial channel
contributes more to the decoding of the total
message than the vocal channel.
There is one other study in which the
characteristics of two-channel communications
of emotion have been explored. Williams and
Sundene (1965) used the semantic differential
method (Osgood, Suci, & Tannenfoaum, 1957)
to obtain judgments of the same emotions
communicated facially, vocally, and in facialvocal
combinations. All three modes of communication
of emotion were found to be
recognized in terms of the three factors of
general evaluation, social control, and activity.
It should be noted that none of the foregoing
studies investigated two-channel communications
in which the emotion communicated
in the facial expression was inconsistent
with that communicated vocally. Despite the
paucity of experimental studies of decoding
of multichannel communications of feeling
or attitude by any particular population
(e.g., children or adults), there is some theoretical
consideration of the effects of such
communications. Bateson, Jackson, Haley,
and Weakland (1956) proposed a "double
bind" theory of schizophrenia. They consider
the maladaptive responses of schizophrenics
to be a consequence of their being the frequent
recipients of inconsistent attitude communications.
The double-bind communication
can be defined as typically consisting of two
or more inconsistent attitude messages which
are assumed to elicit incompatible responses
248
NONVERBAL COMMUNICATION IN Two CHANNELS 249
from the addressee. For example, a mother
asks her son to come over and kiss her while
she nonverbally communicates disinterest in
what he is requested to do. The child is
assumed to be left with the difficult task of
responding to either the verbal or the nonverbal
component, with the knowledge that
response to the former will elicit a rebuff and
response to the latter will elicit indignation.
The recipients of frequent double-bind messages
are assumed to learn to respond with
their own double-bind messages. In the example
considered, the child may respond with,
"I can't come because my leg hurts," or "I
can't come because Trap is holding me," the
hurt leg and Trap being figments of his
imagination.
While double-bind communications are assumed
to lead to the development of maladaptive
patterns of interpersonal functioning,
Haley (1963) also conceptualized most
psychotherapeutic processes as being interpretable
within a beneficial double-bind paradigm.
Haley's thesis is that applications of
the beneficial double bind serve the function
of successfully eliminating the secondary gain
which is associated with a symptomatic behavior
and therefore eliminating the behavior.
The above assumptions about the changeinducing
properties of inconsistent communications
require clarification through investigation
of the ways in which multichannel attitude
communications are decoded. Since the
quantification of degree of consistency or inconsistency
between communications in two
channels is only possible if the two communications
can be scaled along a common
dimension, the general evaluation dimension
obtained in the Williams and Sundene (1965)
study seems appropriate. Mehrabian and
Wiener (1967) pursued these considerations
by investigating the decoding of two-channel
vocal-verbal communications in which three
degrees of attitude (i.e., positive, neutral, and
negative) in the vocal component were each
combined with three degrees of attitude in the
verbal component (i.e., meanings of words).
They found that when vocal communication
of attitude is inconsistent with verbal communication
of attitude, normal addressees respond
to the two-channel communication by
subordinating the verbal component to the
vocal component. If, for example, the word
"scram" is said in a tone of voice which
is independently judged as communicating
positive attitude towards the addressee, the
consensual interpretation of the total communication
is positive.
The present study was designed to investigate
the decoding of inconsistent and consistent
communications of attitude in facial
and vocal channels. Three degrees of attitude
(i.e., positive, neutral, and negative) communicated
in facial expressions were each
combined with three degrees of attitude communicated
vocally. In accordance with Gates'
and Levitt's findings, it was expected that the
decoding of a consistent facial-vocal communication
would yield a judgment equivalent
to that obtained from the decoding of
the facial component only—that is, the facial
component would be dominant.Furthermore,
since Mehrabian and Wiener's (1967) study
indicated that the dominant component in a
two-channel communication determines the
meaning of inconsistent communications, it
was expected that the decoding of an inconsistent
facial-vocal communication would
yield a judgment equivalent to that obtained
from the decoding of the facial component
only. It was therefore hypothesized that judgments
of attitude, on the basis of consistent
and inconsistent pairings of facial and vocal
attitude communications, would yield a main
effect due to variations in the facial component
and no effect due to variations in the
vocal component or its interaction with the
facial component.
METHOD
Subjects. A group of 25 subjects (Ss) was used
in the preliminary selection of a neutral word. A
second group of 17 Ss was used to assess the independent
effects of facial and vocalcommunications.
A third group of 20 Ss was used to obtain the
combined effects of facial-vocal communications.
All 62 Ss were female University of California
undergraduates who participated in the study in
partial fulfillment of introductory psychology course
requirements.
Materials. For the selection of a neutral attitudecommunicating
word, 25 Ss were asked to judge the
attitude of a speaker towards her addressee when
saying each of 15 words. The 15 words were each
presented in written form and Ss recorded their
judgments of attitude on a 9-point scale designated
"like very much," +4 and "dislike very much," —4
250 ALBERT MEHRABIAN AND SUSAN R. FERRIS
at its poles. The word "maybe" was selected as an
appropriate neutral verbal carrier of vocal communications,
since it was rated .28 on the attitude scale
with a standard deviation of .72.
For the selection of vocal communications of three
degrees of attitude, three female speakers were
instructed to vary their tone of voice while saying
the word "maybe," so as to communicate like, neutrality,
and dislike towards an imagined addressee.
Each speaker was instructed to say the word
"maybe" twice in the same way while her communications
were being recorded on magnetic tape.
The 18 items, consisting of two instances each of
positive, neutral, and negative vocal attitude communications
obtained from the three speakers, were
presented to 17 female 5s with the following written
instructions:
The purpose of this study is to find out how well
people can judge the feelings of others. In this
part, you will hear a recording on which the word
"maybe" is spoken in different tones of voice. You
are to imagine that the speaker is saying this word
to another person, the addressee. For each tone,
indicate on the scale what you think the speaker's
attitude is towards the addressee.
A modified form of the semantic differential instructions
(Osgood, Suci, & Tannenbaum, 1957, pp. 80-
85) was used to direct Ss' use of an attitude scale
designated "like," +3 and "dislike," —3 at its poles.
The 18 items were presented in a different random
order to each 5. The positioning of positive and
negative poles of the scale was counter-balanced and
was random in the 18-item sequence.
The facial communications of three degrees of
attitude were selected in a similar manner. Photographs
of three female models were taken as they
attempted to use facial expressions to communicate
like, neutrality, and dislike towards another person.
The photographs were 34 X 4i inch black and white
TABLE 1
INDEPENDENT EFFECTS or VOCAL AND FACIALCOMMUNICATIONS:
DEGREE OF POSITIVE ATTITUDE
INFERRED FROM THREE KINDS OF VOCAL
ATTITUDE COMMUNICATION BY Two
SPEAKERS AND THREE KINDS OF
FACIAL ATTITUDE COMMUNICATION
BY Two MODELS
Communi-
cator
Speaker 1
Speaker 2
Model 1
Model 2
Inferred Attitude Scores Corresponding
to Communications Considered
Positive
M SD
2.41 0.79
2.35 0.70
2.12 0.70
2.3S 0.61
Neutral
M SD
0.06 0.82
-1.12 1.11
0.35 0.61
-0.24 1.15
Negative
M SD
-2.29 0.77
-2.18 0.73
-2.65 1.00
-2.53 0.62
prints of head only against a neutral background.
Eighteen items, consisting of two photographs for
each degree of attitude communicated by each of
the three models, were presented to the same 17 5s
who judged the vocal communications. The instructions
for recording judgments of the facial communications
of attitude were analogous to those
used with the vocal communications. The Ss were
randomly assigned so that eight of them judged
facial communications prior to judging vocal communications
and nine made their judgments in
reverse order.
On the basis of 5s' judgments of the vocal and
facial communications, three vocal communications
(i.e., positive, neutral, and negative), obtained from
each of two speakers, and three facial communications,
obtained from each of two models, were
selected. As the data in Table 1 indicate, the facial
attitude communications of a given value (e.g.,
positive) were selected to match the vocal attitude
communications of the same value. Standard deviations
of judgments as well as their means were
approximately matched. A 3 Attitude X 4 Communicator
X 17 Subject analysis of variance of the
data summarized in Table 1 indicated a significant
effect due to the Attitude factor (F =333.47,
£*/= 2/32; MS, = 1.12, p < .001), no significant effect
due to the Communicator factor (F < 1, df =
3/48; MS, — .37,p > .25), and no significant Attitude
X Communicator effect (F=1.0S, d/ = 6/96; MS, =
.70, p>.25). Thus, the independent effects of all
vocal communications of attitude are comparable
to the independent effects of all facial communications
of attitude within each of the three levels of
attitude investigated.
Design. The three vocal attitude communications
of each speaker were each paired with the three
facial attitude communications of each model. Therefore,
there were 36 experimental conditions, consisting
of 3 Vocal Attitude X 3 Facial Attitude X
2 Speaker X 2 Model interactions. All 36 conditions
were administered to each of 20 5s, thus yielding a
3 X 3 X 2 X 2 factorial design with repeated measures
on all factors.
Procedure. The experiment was individually administered
to each 5. The written instructions presented
to the 5s were:
The purpose of this study is to find out how
well people can judge the feelings of others. You
will be shown photographs of different facial
expressions and at the same time you will hear a
recording of the word "maybe" spoken in different
tones of voice. You are to imagine that the person
you see and hear (A) is looking at and talking
to another person (B). For each presentation,
indicate on the scale what you think A's attitude
is towards B.
A second form of instructions was identical to
those presented above with the exception that references
to facial expressions and seeing, and tones
of voice and hearing, were made in reverse order.
The 5s were randomly assigned so that half received
NONVERBAL COMMUNICATION IN Two CHANNELS 251
the first form and half received the second form
of instructions. A modified form of the semantic
differential instructions was again used to direct Ss'
use of an attitude scale designated "like," +3 and
"dislike," —3 at its poles.
The 36 two-channel communications were presented
in a different random order to each of 20
5s. In each experimental condition, the vocal and
facial components of the communication were presented
simultaneously, so that 5s heard the vocal
component only while seeing the facial component
and vice versa.
RESULTS
Each S recorded 36 responses, corresponding
to all possible combinations of three
facial communications of each of two models
with three vocal communications of each of
two speakers. The responses, which had a
possible range of +3 to —3, were analyzed in
a 3 X 3 X 2 X 2 factorial design with repeated
measures on all factors. The analysis
indicated a significant effect due to Facial Attitude
(F = 233.14, df = 2/38; MSe = 2.37,
p < .001) and a significant effect due to Vocal
Attitude (F = 77.49, df = 2/38; MSe = 3.33,
p < .001). None of the other main or interaction
effects attained the .05 level of significance.
The Facial Attitude X Vocal Attitude
interactions with MSe = 2.21 are summarized
in Table 2. The Newman-Keuls method
(Winer, 1962) yielded significant differences
at the .01 level for all comparisons within
each level of both factors.
The Facial Attitude factor accounted for
41.4% of the total sum of squares, whereas
the Vocal Attitude factor accounted for 19.3%
of the total sum of squares. Furthermore,
the effects of the facial and vocal components
were strongly linear. The linear trend accounted
for 97% of the effect due to the facial
component and 99% of the effect due to the
vocal component. Moreover, the combined effect
of the facial and vocal components was
a weighted sum of their independent effects,
since there was no significant interaction between
them. The following regression equation
summarizes the relative contributions of
facial and vocal components to interpretations
of combined facial-vocal attitude
communications:
AT = 1.50 AF + 1.03Ay
TABLE 2
EFFECTS OF TWO-CHANNEL FACIAL-VOCAL COMMUNICATIONS
: DEGREE OFINFERRED POSITIVE ATTITUDE
CORRESPONDING TO THE FACIAL ATTITUDE
X VOCAL ATTITUDE INTERACTIONS
Vocal component
Positive
Neutral
Negative
Facial component
Positive Neutral
2.45
1.33
0.20
1.31
0.50
-1.07
Negative
-0.91
-1.62
-2.47
AT represents attitude inferred on a —3 to
+3 scale from the two-channel communications.
AF represents attitude communicated
in the facial component alone and is assigned
values of +1, 0, and —1 for positive, neutral,
and negative attitude, respectively. Similarly,
Av represents attitude communicated in the
vocal component alone. The .95 confidence
interval for the coefficient of AF is 1.32 to
1.68, while that of the coefficient of Av is
.79 to 1.28. The absence of overlap between
the two intervals indicated that the effect due
to the facial component was significantly
greater than that due to the vocal component.
DISCUSSION
The hypothesis of the present study was
only partially supported. A main effect due
to variations in the facial component and
no effect due to variations in the vocal component
or its interaction with the facial
component had been expected. The results of
the study indicate that the facial and vocal
components do not interact and that the facial
component has a stronger effect than the
vocal component. However, in contrast to the
hypothesis, the effect due to the vocal component
is also significant. Thus, the results of
the study can be summarized as follows:
Attitudes inferred from two-channel facialvocal
attitude communications are a linear
function of the attitude communicated in
each component, with the facial component
receiving approximately 3/2 the weight
received by the vocal component.
The above results were obtained from a
sample of normal adult female 5s who were
communicators and addressees. However, it is
252 ALBERT MEHEABIAN AND SUSAN R. FERRIS
likely that the linear model for two-channel
communications of attitude obtained for
female communicator-addressee combinations
has broader applicability. For instance, the
model (with, perhaps, slightly different relative
weights for the facial and vocal components)
may be applicable to same- and
different-sex communicator-addressee pairs of
various ages.
One interesting implication of the linear
model with positive coefficients relates to
redundant multichannel communications of
attitude. The model indicates that the effect
of redundancy (i.e., consistent attitude communication
in two or more channels) is to
intensify the attitude communicated in any
one of the component channels. Thus, pushing
a child away while turning away from
him is assumed to communicate a more
negative attitude toward the child than only
pushing him away or only turning away from
him. Similarly, holding and kissing a child
is assumed to communicate a more positive
attitude towards the child than only holding
or only kissing the child.
A final comment is required to integrate
the implications of the findings of the present
study with the findings of the Mehrabian and
Wiener study (1967). It is suggested that
the combined effect of simultaneous verbal,
vocal, and facial attitude communications is
a weighted sum of their independent effects—
with the coefficients of .07, .38, and .55,
respectively. Analytic procedures outlined by
Anderson (1962, 1964) can presently be employed
to test this proposed weighted-sum
model for any single decoder. In view of
these extrapolations of experimental findings
from the decoding of multichannel inconsistent
communications, the assumptions
underlying the effects of double-bind communications
can be questioned. Further
experimentation with schizophrenics or children
as addressees is needed to clarify
the pathology-inducing or behavior-modifying
effects of inconsistent communications of
attitude.
REFERENCES
ANDERSON, N. H. Application of an additive model
to impression formation. Science, 1962, 138, 817-
818.
ANDERSON, N. H. Note on weighted sum and linear
operator models. Psychnomic Science, 1964, 1,
189-190.
BATESON, G., JACKSON, D., HALEY, J., & WEAKLAND,
J. Toward a theory of schizophrenia. Behavioral
Science, 1956, 1, 2S1-264.
DAVITZ, J. R. (Ed.) The communication of emotional
meaning. New York: McGraw-Hill, 1964.
GATES, G. S. The role of the auditory element in
the interpretation of emotion. Psychological Bulletin,
1927, 24, 175. (Abstract)
HALEY, J. Strategies of psychotherapy. New York:
Grune & Stratton, 1963.
LEVITT, E. A. The relationship between abilities to
express emotional meanings vocally and facially.
In J. R. Davitz (Ed.), The communication of
emotional meaning. New York: McGraw-Hill,
1964. Pp. 87-100.
MAHL, G. R., & SCHULZE, C. Psychological research
in the extra-linguistic area. In T. A. Sedeok, A. S.
Hayes, & M. C. Bateson (Eds.), Approaches to
semiotics. The Hague: Mouton, 1964. Pp. 51-124.
MEHRABIAN, A., & WIENER, M. Decoding of inconsistent
communications. Journal of Personality and
Social Psychology, 1967, 6, 108-114.
OSGOOD, C. E., Suci, G. J., & TANNENBAUM, P. H.
The measurement of meaning. Urbana: University
of Illinois, 1957.
WILLIAMS, F., & SUNDENE, B. Dimensions of recognition:
Visual vs. vocal expression of emotion.
Audio Visual Communications Review, 1965, 13,
44-52.
WINER, B. J. Statistical principles in experimental
design. New York: McGraw-Hill, 1962.
(Received December 14, 1966)