University of Southern California Dissertations and Theses

A Multidimensional Scaling Of Mood Expressions

(USC Thesis Other)

A Multidimensional Scaling Of Mood Expressions

PDF

Download a page range

Download transcript

Copy asset link

Request this asset

Transcript (if available)

Content This dissertation has been
microfilmed exactly as received 70-8517
BRADLEY, Paul Arthur, 1943-
A MULTIDIMENSIONAL SCALING OF MOOD
EXPRESSIONS.
University of Southern California, Ph.D., 1969
Psychology, general
University Microfilms, Inc., Ann Arbor, Michigan
A MULTIDIMENSIONAL SCALING
OF MOOD EXPRESSIONS
by
PAUL ARTHUR BRADLEY
A Dissertation Presented to the
FACULTY OF THE GRADUATE SCHOOL
UNIVERSITY OF SOUTHERN CALIFORNIA
In Partial Fulfillment of the
Requirements for the Degree
DOCTOR OF PHILOSOPHY
(Psychology)
August 1969
UNIVERSITY O F SO U T H E R N CALIFORNIA
T H E G RAD U ATE S C H O O L
U N IV ERSITY PARK
LO S A N G E L E S, C A L IF O R N IA 9 0 0 0 7
This dissertation, written by
........
under the direction of h..X S.. Dissertation Com
mittee, and approved by all its members, has
been presented to and accepted by The Gradu
ate School, in partial fulfillment of require
ments of the degree of
D O C T O R O F P H I L O S O P H Y
\ J Dean
D a te AUGUST.1961
ACKNOWLEDGMENTS
I would like to single out two individuals in par
ticular for their special contributions in the making of
this dissertation. To my chairman, Dr. Norman Cliff, I
will always be grateful for his enthusiastic support and
guidance of this research, without which the completion of
this dissertation would have been infinitely more diffi
cult.
A driving force over the past four years, and
especially this last one, was my wife. She has not only
lent moral support and encouragement, but she has also
actively participated in various aspects of the project.
Thanks are also extended to the Aptitudes Research Project
for their permission to use materials from several tests
and to project personnel for their cooperation in the
selection of the mood expressions.
This investigation was partially supported by
Biomedical Sciences Support Grant FR-07012-02 from the
General Research Support Branch, Division of Research
Resources, Bureau of Health Professions Education and
Manpower Training, National Institutes of Health. Comput
ing assistance was obtained from the Health Sciences Com
puting Facility, UCLA, sponsored by NIH Grant FR-3.
TABLE OF CONTENTS
ACKNOWLEDGMENTS .................................. ii
LIST OF TABLES.................................. v
LIST OF ILLUSTRATIONS............................ vi
Chapter
I. GENERAL BACKGROUND ........................ 1
The Concept of Mood
II. DEVELOPMENT OF PROBLEM .................... 9
Contributions of Woodworth and Schlosberg
Verification of Schlosberg's Model .
Factor Analyses of Check Lists
Theoretical Categories of Emotion
Postural Expressions
Summary
III. METHODOLOGICAL CONSIDERATIONS ............... 33
Multidimensional Scaling
Individual Differences
Summary
IV. METHOD.............................. 50
Construction of Booklets
Subjects
Administration
Data Analysis
V. RESULTS..................................... 64
Repeated Measurements
Individual Differences
Multidimensional Scaling
iii
VI. DISCUSSION................................ 103
Repeated Measurements
Individual Differences
Multidimensional Scaling
Conclusions
VII. SUMMARY.............................. 127
REFERENCES...................................... 130
APPENDIX.................................... 138
iv
LIST OF TABLES
Table Page
1. Analysis of Variance for Repeated
Measurements ............................ 66
2. Correlations among Groups (all items--
above diagonal; repeated items--below
diagonal; reliabilities--diagonal)
and Descriptive Statistics .............. 71
3. Stress Values ................... ...... 74
4. Eigenvalues................................. 77
5. Rotated Coordinate Matrices ................. 79
6. Correlations between AVE Dimensions and
Corresponding Group Dimensions .......... 92
7. Regression Coefficients for Dimensions
in Groups when Predicting Corresponding
Dimensions in AVE........................ 93
8. Stress Values: Within Mode................ 98
9. Rotated Coordinate Matrices: Within Mode . . 99
v
LIST OF ILLUSTRATIONS
Figure Page
1. Mood Expressions Used in this Study........ 53
2. Schematic Representation of the Three-
Dimensional Configuration of Mood
Clusters: Anger - Top of Triangle;
Fear-Anxiety - Lower Left Corner of
Triangle; Sadness - Lower Right Corner
of Triangle; Happiness - Point............ 95
vi
CHAPTER I
GENERAL BACKGROUND
The Concept of Mood
In past, present, and, undoubtedly, future
research there has been, is, and will be a concern with
identifying basic psychological concepts which can be
used to describe characteristics of human behavior. In
the affective domain such concepts have been referred to
as moods, emotions, feelings, or affects.
While the actual words used by various people to
define these terms are usually different, there is common
agreement that they refer to temporary, unstable, organis-
mic states that are recognized in both the self and others
through cognition. While this statement distinguishes
mood from the personality domain, it does not imply that
they are unrelated. One piece of evidence for their
relatedness is that the same adjectives frequently appear
in both mood and personality inventories. For instance,
while most people experience feelings of anxiety or ten
sion at certain moments, some people are characterized as
nervous or anxious, implying that these states are less
temporary in such persons and tend to be descriptive of
more stable behavioral characteristics. Similar state
ments could be invoked for most terms used to describe
moods and emotions.
A common bond between mood states and personality
traits is their motivational characteristics, and their
common status as intervening variables. They interact in
a complex and unknown way with each other and with the
situation to determine the occurrence of a particular
behavioral response. Knowing another's emotional state,
like knowing certain personality traits, leads to infer
ences of how that person reacts in a given situation and
also what behavior will not occur. Although there are
these and other similarities between mood and personality,
a fundamental difference is that mood and emotional states
are temporary reactions, but personality traits are stable
behavioral tendencies.
The emphasis here is on some cognitive aspects of
mood, as is most recent research in this domain; neverthe
less, there are physiological aspects that have received
attention and are an integral part of emotion. The cur
rent concern with the role of cognition in mood has devel
oped from a lack of evidence that emotional states are
discriminated from one another on the basis of patterns
of physiological reactions. Awareness of one's own mood
is partly dependent on cues received from the visceral
3
system, but very little of this information is available
to the perceiver when inferring another's mood state.
The cues used in mood perception include facial expres
sions, verbal expressions, vocal inflection, postural
reactions, behavioral cues, and situational cues. Some
of these cues also play an important role in cognizing
one's own mood. Concern with these issues has tended to
relegate to a subordinate position those aspects of mood
concerned with arousal level and the seat of emotions,
which assumed primary attention in Lindsley's theory of
emotions and in the Canon-Bard controversy (see
Schlosberg, 1954).
Nevertheless, there are very few conceptualiza
tions of mood or emotion that fail to mention the physi
ological aspect and its role in the total organismic
reaction. Both Plutchik (1960) and Tomkins (1962, 1963),
who have dealt with the theoretical issues of moods and
emotions, include in their definitions the internal func
tions of the muscular, glandular, and nervous systems, and
how they are organized in patterns of responses. As impor
tant as these bodily responses are, they are subordinate
to behavioral reactions in the identification and expres
sion of mood. Schacter and Singer (1962) hypothesize that
both types of activity are necessary conditions for cog
nizing emotional states, but neither alone is sufficient.
They suggest that cognitive factors are the major deter
minants of emotional states, serving as a steering func
tion for labeling the state of physiological arousal.
Up to this point the terms mood and emotion have
been used interchangeably, and will be throughout the
report. Along with these two, the terms feeling and
affect have also denoted similar concepts. Very frequent
ly all four terms appear together and very little effort
is made to distinguish among them. There are different
predilections for emphasizing one of them, but not to
such an extent that clear distinctions between them are
drawn. The concept of mood is most frequently applied in
factor-analytic studies, where it has been said that mood
and emotion are overlapping categories, but mood states
are likely to be more persistent and involve less intense
physiological arousal than emotional states (Brown, 1965;
McNair § Lorr, 1964). Nowlis (1965) takes mood as his
object of study, but its distinction from affect and feel
ing is not made, as evidenced by the following remarks:
" . . . search for . . . instruments for monitoring the
current affective status of an individual [p. 352]";
"The MACL is based on . . . the tendency . . . to . . .
complete the sentence 'I feel ______ [p. 353].'" In the
oretical developments and in studies involving facial
expressions, the prevalent term is emotion. This
preference is not exclusive and is usually not accompanied
with its distinction from mood, feeling, or affect. For
example, Plutchik (1966) talks about the need for a theory
of emotions to provide some structure when affective
states are measured by asking persons to report what moods
they are feeling.
Some differences in connotations arise when these
terms are used to describe persons. For instance, a moody
person suggests something different than an emotional per
son or an affectionate person. Feeling is used in other
contexts to refer to attitudes or thought processes.
When these terms are restricted in their reference to cur
rent and temporary state of being, however, they are
interchangeable and there are no difficulties in communi
cation of their meaning. While the particular words that
may be used in defining and delimiting the concepts differ
from each other, the same general impression is expressed.
That impression is that moods, emotions, feelings, and
affects involve physiological responses, skeletal
responses, facial expressions, and behavioral reactions,
and the feedback or observation of these patterns give
rise to the perception and cognition of the current func
tioning and orientation of the organism.
The apparent consensus of opinion with regard to
what is the basic nature of a mood or an emotion is not
paralleled by agreement as to what are the basic moods or
emotions. This situation is not unique to the affective
domain, however, as a similar state of affairs exists in
the intellectual and personality domains. This disagree
ment can be traced to two sources: theoretical versus
experimental considerations, and strategies in the identi
fication of basic dimensions. The nature of the disagree
ment is common to both these sources and is exemplified
by two research techniques.
One strategy or approach has involved the appli
cation of correlational and factor-analytic techniques to
measures of individual differences in self-reported mood
or emotional states. The measuring device in such studies
is usually an adjective check list. The results have led
to the identification of five to twelve dimensions of
mood. These dimensions are similar in nature to catego
ries of primary emotions that have been postulated on a
theoretical basis. Another strategy has involved the
application of either multidimensional scaling techniques
to ratings of perceived similarity of facial expressions,
or factor-analytic techniques to semantic differential
ratings of facial expressions. In contrast to the results
of between-person studies such as adjective check-list
studies and to some theoretical conceptualizations, this
strategy has usually produced evidence for only two to
four dimensions. Furthermore, the nature of these dimen
sions does not at all correspond to the nature of the
previous categories or dimensions.
It was this gap in the number and nature of pri
mary concepts in the affective domain that led to the
research reported here. On the one hand, the factor-
analytic studies produced dimensions that represented
specific categories of mood, while, on the other hand, the
multidimensional scaling studies produced dimensions that
represented general characteristics of mood. There was a
possibility, however, that the above differences had
arisen as a consequence of the two strategies dealing with
different types of stimuli. That is, the factor-analytic
studies were based on mood adjectives, while the multi
dimensional scaling studies were based on facial expres
sions. The former type of stimuli is a verbal mode of
mood expression.
One goal of this research was to determine if a
multidimensional scaling analysis of a new set of mood
expressions would result in a set of dimensions that rep
resented general characteristics of mood. A second goal
was to determine if different modes of expression would
have any effect on ratings of mood similarity.
The strategy adopted was to select four mood
categories that had been isolated by the between-persons
approach and cross each mood category with four modes of
expression. The data obtained were similarity ratings of
pairs of mood expressions, and these ratings were sub
jected to the within-person analysis of multidimensional
scaling. Prior to the multidimensional scaling analysis,
however, the ratings of perceived similarity of mood were
analyzed for individual differences. The goal of this
analysis was to determine if there might be different
spatial representations of the mood expressions for sep
arate groups of persons.
The crucial information is obtained from the spa
tial representation of the mood expressions, which would
indicate if the stimuli cluster together on the basis of
mood similarity, mode similarity, both, or neither, and
if the dimensions are interpretable as mood factors, mode
factors, both, or neither. The major concern is with the
number of dimensions and the configuration of the mood
expressions in the spatial representation.
CHAPTER II
DEVELOPMENT OF PROBLEM
The study of moods and emotions has a very long
history, extending back to the middle of the nineteenth
century. This early research was concerned with identify
ing emotions in facial expressions, where the observer's
task was to choose or provide a label that described the
emotional content of the facial expression. Two names
associated with these studies are Piderit and Darwin (see
Engen, 1965), who illustrated emotions with line-drawn
faces and photographs, respectively. The results were not
very successful, as there was a considerable degree of
confusion between supposedly different emotions. While
these early and following studies of accuracy in identify
ing mood states were not themselves concerned with discov
ering basic dimensions of emotions, it was the confusion
of labels that led to the first scale for judging facial
expressions.
Contributions of Woodworth
and Schlosberg
Woodworth (1938) analyzed the data from labeling
studies and, as a result, found that emotional expressions
10
could be ordered in a six-category scale. These catego
ries were love, mirth, happiness; surprise; fear, suffer
ing; anger, determination; disgustt and contempt. Wood-
worth proposed that these categories could be arranged in
a six-step scale, with love, mirth, happiness at one end
and contempt at the other end. The basis for this order
ing was that errors of identification could be better
accounted for, as adjacent categories were more likely to
be confused than categories further apart.
Schlosberg (1941) asked observers to sort the
Frois-Wittman (Hulin § Katz, 1935) pictures into Wood
worth's six categories plus a seventh category for expres
sions not easily classified in the other six. The results
indicated that there was a modal response for each pic
ture, with dispersion into adjacent categories. One
exception to this generality led Schlosberg to propose
that the scale was circular rather than linear: contempt
was often confused with love, mirth, happiness.
Further analysis by Schlosberg (1952) suggested
that the circular scale could better be conceived as a
surface, like the color surface, with two dimensions:
Unpleasantness-Pleasantness (UP) and Attention-Rejection
(AR). The meaning of the AR dimension was not easily
explicated, but it was thought that Attention was exempli
fied by expressions maximally open to stimulation and
11
Rejection was exemplified by expressions that seemed to be
shutting out any stimulation. To support this two-
dimensional model, Schlosberg (1952) obtained ratings of
the Frois-Wittman pictures on these two dimensions, and
attempted to predict Woodworth categories from them. The
success of this venture is evident from the correlations
between predicted and actual scale values. These correla
tions ranged from .92 to .96, and they supported the
notion that the two-dimensional surface provided a better
model of the data than the circular scale. In addition to
the above prediction, it was also clear that expressions
located in the middle of the surface were sorted into
Woodworth categories with less accuracy and with more
scattering.
A third dimension was added to Schlosberg1s model
(Schlosberg, 1954) to account for some expressions that
were not discriminated by the two-dimensional model. This
dimension was Sleep-Tension (ST), and was analogous to the
black-white axis in the tri-dimensional color model. It
was also likened to intensity of emotion or degree of
arousal. To have better expressions of this added dimen
sion, a new series of facial expressions was developed,
the so-called Lightfoot faces (Engen, Levy, § Schlosberg,
1957, 1958). These faces have been sorted into Woodworth
categories (Levy 8 Schlosberg, 1960), and the average
12
error of prediction of their scale values from their
ratings on UP and AR, as given by Engen, Levy, and
Schlosberg (1958), was .50 scale steps on a six-point
scale.
No further dimensions have been added to this
three-dimensional model, and research by Schlosberg and
his associates has more or less ceased. A fairly exten
sive review of this model is given by Engen (1965). One
recognized drawback to the model is that the dimensions
are imposed on the observers providing the ratings, and
that experimental methods which do not make this imposi
tion may result in the appearance of more dimensions,
fewer dimensions, or different dimensions. There have
been two approaches to the verification of the three-
dimensional model that have not imposed the Schlosberg
dimensions on the observers: multidimensional scaling
studies of ratings of similarity of facial expressions,
and factor-analytic studies of semantic differential
ratings of facial expressions.
Verification of Schlosberg's
EfodiT----- -------
Multidimensional scaling. Abelson and Sermat
(1962) selected 13 representative photographs from the
Lightfoot series, and secured similarity ratings for the
78 pairs of facial expressions. In an attempt to
13
determine if these distances were accounted for by
Schlosberg's model, a multiple-regression analysis was
performed, where the ratings of similarity were predicted
from differences in scale values on each of the three
Schlosberg dimensions. It was found that differences on
the Unpleasant-Pleasant axis, alone, correlated .71 with
the similarity ratings. Adding Sleep-Tension increased
this to .85, and with all three dimensions the correlation
was .87. From this analysis it seemed that the ST dimen
sion was more important than AR in the prediction of the
similarity ratings. In the multidimensional scaling
analysis, the first two dimensions accounted for 73% of
the total spread in a five-dimensional space. The coordi
nates of the Lightfoot faces on each of these two dimen
sions were correlated with their corresponding values on
UP, AR, and ST. The correlation between axis I and UP
was .95, between II and ST: .92, and between II and AR:
.88. Thus, the first dimension is clearly Unpleasant-
Pleasant, but the second dimension is nearly equally
equivalent to both Sleep-Tension and Attention-Rejection.
A suggested compromise for the second dimension was acti
vation or arousal, which it was felt was more akin to
Sleep-Tension than Attention-Rejection.
Gladstones (1962) used 10 Lightfoot photos and
retained three dimensions that accounted for 49%, 34%, and
17%, respectively, of the variance attributable to the
first three dimensions. The rank order correlations of
coordinates on these dimensions with ratings on the
Schlosberg dimensions were obtained, with results similar
to those of Abelson and Serraat (1962). Dimensions I and
II correlated over .90 with UP and ST, respectively,
while the correlation between II and AR was .82, and
between III and AR was .53. Again it appears that AR is
not as salient as ST in predicting similarity ratings or
euclidean coordinates. Gladstones interpreted II as
Sleep-Tension and III as Expressionless-Mobile, but this
latter interpretation was difficult to make.
The results of a third multidimensional scaling
analysis are reported by Cliff and Young (1968). Thirteen
Lightfoot faces were included in this study, some of which
are common to the faces in the above two studies, and two
dimensions were interpreted, although there was evidence
that one more dimension could be reliably demonstrated.
The first dimension was again interpreted as Unpleasant-
Pleasant, and projections on the second dimension were
found to correlate (Spearman Rho) .89 or more with ratings
of intensity of the expressed emotion. However, even
higher correlations were obtained when the intensity
ratings were compared to distances from a hypothetical
origin in the two-dimensional space. This origin was
15
located at a point that has been occupied by a facial
expression representing light sleep.
In all three of these studies it has been quite
clear that the three Schlosberg dimensions do not all
emerge. Rather, there appears to be considerable redun
dancy between Sleep-Tension and Attention-Rejection, and
ST seems to be the more salient of the two. This overlap
is also exhibited in the correlations of .82 between the
ratings of the Lightfoot faces on these two scales (Engen,
1965). In addition, the configuration of the faces in the
two-dimensional space has been very consistent in the
above three studies and Shepard's (1962b) reanalysis of
the Abelson and Sermat (1962) data with his nonmetric
multidimensional scaling program.
Semantic differential. Another approach to veri
fying Schlosberg's model has been through the factor
analysis of correlations between semantic differential
scales, where the concepts have been facial expressions.
Frijda and Philipszoon (1963) appear to be the first to
take this approach. In their study twelve subjects rated
30 facial expressions (not from either the Lightfoot or
Frois-Wittman series) on 22 bipolar scales. A centroid
factor analysis was applied to the rank order correlations
between scales, and four factors were obtained. Three of
these were interpreted as corresponding to Schlosberg's
16
dimensions, and the fourth was tentatively labeled social
submission versus social condescension (as exemplified in
the semantic differential scale submissive-authoritarian).
Kauranne (1964) obtained ratings of 30 Frois-
Wittman faces on 10 semantic differential scales and com
puted the distances between the profiles of the faces.
These D2 values were scaled so as to range from 0.0 to
1.0, factor analyzed by the centroid method, and three
dimensions were rotated to oblique structure. The dimen
sions were interpreted as anger, pleasure, and contempt,
with the interpretation being aided by labels associated
with the faces that were univocal on each factor. While
these factors do not correspond to Schlosberg dimensions,
they do correspond to locations in the space spanned by
UP and AR. Furthermore, the appropriate analysis of the
D2 values should have been a multidimensional scaling
analysis rather than a factor analysis. If they were
analyzed in this way it would be quite likely that only
two dimensions would appear, and these dimensions might
have corresponded to UP and AR.
In yet a third study, Mordkoff (1967) selected 24
of the Frois-Wittman photos and obtained ratings on 48
semantic differential scales. Intercorrelations among the
48 scales were factor analyzed by the principal components
method, and four factors were rotated according to the
17
normal Varimax criterion. The largest factor was not
readily interpreted as corresponding to any of the
Schlosberg dimensions, but appeared to be an amalgamation
of Osgood's (Osgood, Succi, § Tannenbaum, 1957) potency,
oriented activity, and evaluative dimensions of meaning.
Two other dimensions, however, were interpreted as cor
responding to UP and AR, while the fourth factor was
labeled oriented vs. disoriented activity. The large
first dimension was labeled forceful-submissive, and
Mordkoff suggests that such a dimension was omitted in
Schlosberg's model because the Woodworth categories did
not seem capable of generating such a dimension. Mordkoff
further suggests that the one Schlosberg dimension not
obtained, Sleep-Tension, was also the dimension postulated
from considerations other than studies of facial expres
sions.
One further application of semantic differential
scales to facial expressions also resulted in four dimen
sions of meaning (Sweeney, Tinling, Eby, § Schmale, 1968).
In this study the stimuli were 13 photographs of both face
and body obtained from subjects (some of whom were under
light hypnosis) who were given a cue word and asked to
express the emotion facially and posturally. The inter
correlations among 46 bipolar scales were factor analyzed
by the principal components method, and the factors
18
rotated to normal Varimax criterion. The largest factor
was clearly an evaluative dimension, and would therefore
seem to correspond to Unpleasant-Pleasant. The second
largest factor was labeled activity-potency, and might
possibly correspond to the overlap of Attention-Rejection
and Sleep-Tension. The third and fourth factors were
much smaller than the first two, and did not seem to be
analogous to any Schlosberg dimension.
In a non-semantic differential study, Osgood
(1966) factor analyzed confusion data for 40 emotional
labels and obtained at least three dimensions that were
interpretable. Two of these corresponded to Unpleasant-
Pleasant and Sleep-Tension, and while the third was
labeled Control, it did resemble Attention-Rejection in
that it contrasted loathing with interest and horror.
Wundt's tri-dimensions of feelings include the same two
dimensions in common with the Schlosberg model as Osgood's
results: Unpleasant-Pleasant and Tension-Relaxation (see
Mordkoff, 1967). The third dimension is Excitement-
Acquiescence.
It is seen that there is a good deal of consis
tency in finding the Unpleasant-Pleasant and Sleep-Tension
dimensions across the multidimensional scaling and seman
tic differential studies, as well as Osgood's study and
Wundt's model. It is worth noting that the two studies
19
which failed to support the Sleep-Tension dimension
(Kauranne, 1964; Mordkoff, 1967) were the only ones that
used the Frois-Wittman series of facial expressions.
Schlosberg's early work was based on this series, but when
the Sleep-Tension axis was postulated the Lightfoot series
was produced in order to obtain better representation on
this dimension. On the other hand, the Lightfoot series
has been criticized for not covering the full range of
Attention-Rejection, and this may, in part, account for
this dimension not appearing in studies based on the
Lightfoot series. A host of third dimensions have been
interpreted, partially excepting the multidimensional
scaling studies, and some of the labels attached to these
dimensions sound alike, but it is not at all easy to
reconcile the similarities and differences. A hindrance
is that in the semantic differential studies (Mordkoff,
1967; Sweeney, et al., 1968, in particular) there were
very few common scales.
While there is this supporting evidence for at
least two of the Schlosberg dimensions, there are some
attenuating circumstances in the studies cited. One is
that most of the studies have employed a set of facial
expressions on which the model was originally based. Only
Osgood (1966) and Sweeney, et al. (1968) developed a new
series of facial expressions. One could be tempted to
20
say that a two-dimensional model of emotions applies only
to the two actors, or further, to middle-aged, Caucasian
males and females. Secondly, in the multidimensional
scaling studies, the analyses were based on no more than
13 photographs. It is possible that this sample is inad
equate to cover the entire range of emotional expressions.
Another limiting factor is that in the semantic differen
tial studies the factors are derived from correlations
among adjective scales, not facial expressions, and as a
consequence there was no spatial representation of the
faces. And, finally, these dimensions of emotional
expression are limited to just one mode--facial expres
sions, and are based on perception of mood in others.
Clearly, the most important cue in perceiving mood is the
face, but there are other manifestations of emotions. As
important as it is to have an understanding of the impor
tant dimensions in mood perception, it is equally impor
tant to know if these dimensions apply in reporting one's
own moods.
Despite these limitations, as minor or severe as
they may be, Schlosberg's model has withstood attacks upon
it and has maintained certain constancies. There have
been no overwhelmingly contradictory findings in the studr
ies of facial expressions, and because of this, one could
safely conclude that Unpleasant-Pleasant, Sleep-Tension,
21
and possibly Attention-Rejection are salient dimensions of
mood. However, there are studies and theories of moods
and emotions that are not compatible with Schlosberg's
model. These studies are factor analyses of mood adjec
tive check lists, where the data are self-reports of mood
states, and the theories are attributable to Plutchik
(I960, 1962, 1965) and Tomkins (1962, 1963; Tomkins §
McCarter, 1964). There are two basic differences between
Schlosberg1s model and the results of the factor-analytic
studies and the theories of emotion. One is that more
dimensions of mood are found or postulated (the range is
from five to twelve), and secondly, these dimensions are
interpreted in terms of categories or typologies of mood.
That is, the names of the dimensions include anger, fear,
depression, joy, surprise, anxiety, and the like.
Factor Analyses of Check Lists
There are four published studies that report the
results of one or more factor analyses. Although the
experimental conditions, size and type of sample, pool
of adjectives, and type of factor analysis are somewhat
different in these studies, there are several dimensions
that appear to be invariant. The most comprehensive
research has been conducted by Nowlis (1965) , who devel
oped a check list of 130 words. This list was adminis
tered to a sample of approximately 450 college males at
22
the beginning and end of six experimental sessions that
attempted to induce different mood states. Five sets of
data were selected to be analyzed by the centroid method
of extraction and oblique rotation of axes. A total of
12 factors were found over the five analyses, but only
eight appeared in four or more solutions. These factors
were: Aggression, Surgency, Concentration, Fatigue,
Social Affection, Sadness, Anxiety, and Skepticism. The
less frequently found factors were Egotism, Elation, Non
chalance, and Vigor. Replications with a shorter form of
the check list supported some of the consistent dimen
sions. The influence of these studies and the develop
ment of the Mood Adjective Check List (MACL) is wide
spread, as Nowlis (1965) reviews 40 published studies that
used various versions of the MACL.
Borgatta (1961) obtained two sets of responses to
a 40-item version of the MACL, and rotated seven orthog
onal factors. Of these seven, only six were stable and
well defined across the two analyses (comparable Nowlis
factors are given in parentheses): Lonely (Depression),
Warmhearted (Social Affection), Tired (Fatigue), Thought
ful (Concentration), Defiant (Aggression), and Startled
(Anxiety). The latter two factors were small, with only
two adjectives defining each scale. The seventh factor
was Egotism, a factor found in only one of the Nowlis
analyses.
23
McNair and Lorr (1964) analyzed the responses of
psychiatric outpatients to a 55-item check list. Three
sets of data were collected, and five oblique factors were
invariant across the three analyses. These factors were:
Tension-Anxiety, Anger-Hostility (Aggression), Depression-
Rejection, Vigor-Activity, and Fatigue-Inertia. In one of
the analyses two additional factors were found: Friendli
ness (a weak factor) and Confusion (sort of opposite of
Concentration). Contrary to the Nowlis studies, where the
correlations among the oblique factors were low, there
were several correlations in- a moderate range of .40 to
.60. It was suggested, however, that the nature of the
sample may have inflated the correlations, as data from
normals resulted in lower correlations between the fac
tors .
Finally, Lorr, Datson, and Smith (1967) sought to
replicate the five factors found by McNair and Lorr
(1964), and found three additional dimensions: Cheerful
ness, Thoughtful, and Composed-Relaxed. They hypothesized
that the correlations among the factors would yield a cir-
cumplex matrix, with the mood dimensions arranged in the
following order: Cheerful, Active-Energetic, Angry-
Hostile,' Tense-Anxious, Thoughtful, Dejected, Tired-Inert,
and Composed-Relaxed. While a circular.ordering of moods
is reminiscent of Woodworth's scale, an equitable
24
comparison is somewhat futile as there are only two cate
gories that are definitely comparable: Love, Mirth,
Happiness with Cheerfulness and Anger, Determination
with Angry-Hostile.
Data were collected from normals on a 62-item
adjective check list, and eight factors were extracted by
the multiple group method. Necessarily these eight fac
tors were the same as the eight hypothesized. The pri
mary interest was in the correlations among the factors.
The matrix was partially like a circumplex, but there was
a gap between Thoughtful and Relaxed-Composed, and adja
cent mood categories were low to moderately correlated
with each other. It appeared that if the matrix of cor
relations among the factors were factor analyzed, more
than two second-order factors would be needed to reproduce
the correlation matrix.
Theoretical Categories
of Emotion
A circular ordering of moods similar to that
hypothesized by Lorr, Datson, and Smith (1967) is also
postulated by Plutchik (1960, 1965). In his system,
emotions can be summarized by eight prototype patterns
of reaction that represent four bipolar tendencies toward
action. The adjectives that describe each of these reac
tions are: joy vs. sadness, acceptance vs. disgust,
25
surprise vs. expectancy, and fear vs. anger. The ordering
of these primary emotions on the circular scale is joy,
acceptance, surprise, fear, sadness, disgust, expectancy,
and anger. There are some contrasts between this model,
the Woodworth scale, and the ordering of moods proposed by
Lorr, et al. (1967). In this circle, joy and anger are
adjacent categories, whereas in Lorr, et al., they are
separated by one category and the correlation between
them is -.09. Insofar as Love, Mirth, Happiness is com
parable to joy, Anger, Determination in Schlosberg's
circular ordering of the Woodworth scale is opposite to
Love (Schlosberg, 1941). In that model, Contempt is adja
cent to Love, whereas in Plutchik's scheme they are sepa
rated by two categories. In Plutchik, fear and anger are
bipolar tendencies, but in Schlosberg they are adjacent.
There are several moods in Lorr, et al., that are not
common to either Plutchik or Woodworth and, hence, prevent
further comparison.
Additional aspects of Plutchik's model include a
dimension of intensity that would be placed as a vertical
axis over the circular base, and the derivation of addi
tional emotional reactions by combining the primaries in
various ways. Primary dyads are formed by adjacent emo
tions, whereby, for example, joy and acceptance give love.
Secondary dyads are formed by combining two primary
26
emotions that are separated by one category. For example,
disgust and anger yield contempt, which happens to be a
primary emotion in Woodworth's scale and in Tomkins theory
(1962, 1963). Tertiary dyads are combinations of bipolar
emotions, and the mixing of these near opposite emotions
is taken as an indication of clinical emotions.
Tomkins' theory postulates eight primary affects,
three positive and five negative. The positive affects
are interest-excitement, enjoyment-joy, and surprise-
startle; and the negative affects are distress-anguish,
fear-terror, shame-humiliation, contempt-disgust, and
anger-rage. For Tomkins, affect is primarily facial
behavior, and he has characterized each of the affects in
terms of their corresponding facial expression. Further
more, these primary affects are somewhat distinguishable
in terms of a general principle of the density of neural
firing or stimulation. Startle, fear, and interest are
characterized by different slopes of stimulation increase;
anger and distress correspond to prolonged, steady stimu
lation; and joy corresponds to stimulation decrease.
The listing of all moods and emotions that have
been found in factor-analytic studies or that have been
postulated in theories would provide a rich and compre
hensive catalog of affect terms. A somewhat unfortunate
fact is that there are very few mood concepts that are
27
common across both theory and experimental studies. There
are some dimensions that are unique to the factor-analytic
studies, and others that are unique to theory. There are
three moods that are definitely consistent between theory
and practice: joy or cheerfulness, sadness, and anger.
The Concentration factor found in three factor-analytic
studies, the primary affect, interest-excitement, in
Tomkins' theory, and the primary dyad, curiosity, in
Plutchik's theory, may be indicative of a similar mood at
lower levels of intensity.
Among the dimensions that are unique to the
factor-analytic studies, there are some that are represen
tative of physical states: Energetic, Fatigued, and
Relaxed, and one that may be more like a personality trait
--Social Affection. There are several categories that
appear in only the theoretical classifications. Disgusted-
contempt, fearful, surprised, and shame are included in
Plutchik, Tomkins, and Woodworth (except shame), agreeable
and expectant are unique to Plutchik, and distress is
unique to Tomkins. While it is acknowledged that moods
refer to temporary states, it seems that two of the pri
mary affects and emotions suggested by Plutchik or Tomkins
are more like brief reactions that are not likely to be as
prolonged as moods such as anger or sadness. These two
are shame and surprised. It is possible that some of the
28
remaining emotions unique to theory could be accounted for
by other moods. For example, disgust-contempt may best be
classified with anger, and expectant may be related to the
factor of Tension-Anxiety. Tomkins' category of distress-
anguish and Woodworth's suffering may be more like a
physical state of pain rather than a psychological dimen
sion of mood. Plutchik's category of agreeable is remi
niscent of Schlosberg*s dimension of Pleasantness, and
may, therefore, not be indicative of a particular mood or
emotion, but rather a more general characteristic, such
as "being in a good mood."
While initially there appeared to be much hetero
geneity among the various moods and emotions that had
been isolated in factor-analytic studies or had been
included in theoretical classifications, it seems that
the range can be narrowed to a much smaller set if dimen
sions reflecting physical states and categories reflecting
brief reactions are discarded. Included in this smaller
set are Joy, Anger, Sadness, Fear, Tension-Anxiety, Inter
est and Concentration. It could be said that these dimen
sions or categories of mood are not dependent on a par
ticular mode of expression, as were the dimensions in
Schlosberg's model. Some of these dimensions were
obtained through verbal expressions, while in Tomkins'
and Woodworth's classification there was consideration of
29
mood in terms of facial expressions. But there is one
other source from which information about mood and emotion
is obtained: postural cues.
Postural Expressions
The role of body cues in the perception or expres
sion of mood has received scant attention, although it is
recognized as being an important source of information.
Unlike the study of facial expressions and mood adjec
tives, research with body cues has been focused exclusive
ly on accuracy of perception and not at all on dimensions
underlying the mood content of postural cues. The most
devoted investigator in this area is Ekman (1964, 1965b;
Ekman § Friesen, 1967). In one study (1964) judges were
shown pairs of photographs, some of which did not include
the head, and a speech sample to accompany each pair. The
task was to select the photograph that matched the emo
tional content of the written speech sample, and the
results indicated that accuracy was greater when the pho
tographs showed either the head only or the whole body.
In a later study (1965a) it was hypothesized that head
cues contained information about particular emotions, but
little information about intensity, whereas it was hypoth
esized that body cues provided more information about
intensity than about particular affects. To test the
hypothesis, photographs of whole body, head only, and body
30
only were rated on the Schlosberg dimensions of Unpleasant-
Pleasant and Sleep-Tension. Among the results, it was
found that correlations between head only and body only
were low for both dimensions, that the correlation between
head only and whole body was higher on UP than on ST, and
that the correlation between body only and whole body was
higher on ST than UP, but the size of this correlation
was at best .57.
The hypothesis that head and body cues provided
different information about specificity and intensity of
mood, although somewhat supported by the above results,
was later discarded in favor of a new hypothesis (Ekman
§ Friesen, 1967). In this new hypothesis there is a dis
tinction between body position and body acts, and between
facial expressions and head orientation. Body acts and
facial expressions are said to convey information about
specific moods, while body positions and head orientation
convey information about gross affective state, such as
pleasantness or intensity. This proposition could be
translated into expected results of a multidimensional
scaling analysis of the four types of information: the
dimensions underlying facial expressions and body move
ments would correspond to categories of mood, whereas the
dimensions underlying head orientations and body positions
would correspond to the Schlosberg dimensions. Previous
31
multidimensional scaling studies of facial expressions
have not supported this hypothesis, however.
Ekman's studies have all been based on photographs
of postural cues. It is possible to abstract the infor
mation in these cues through the use of line-drawn stick
figures. Such stimuli have been incorporated in a test
of conformance in role perception (Sarbin § Hardyck,
1955), as well as tests for measuring behavioral abilities
in Guilford's Structure of Intellect model (Guilford,
1967). In terms of the categories of information indi
cated by Ekman and Friesen (1967) , the stick figures
would cover both body position and head orientation, but
would not include facial expression. According to their
hypothesis, if stick figures were subjected to a multi
dimensional scaling analysis, the result should indicate
dimensions that correspond to Schlosberg's dimensions.
Summary
Early research in the affective domain stressed
identification of emotion in facial expressions rather
than identification of basic mood concepts. This situ
ation changed with Woodworth's (1938) six-category scale
and Schlosberg's (1954) three-dimensional model of facial
expressions. Considerable support for two of these dimen
sions (Unpleasant-Pleasant, Sleep-Tension) has accumulated
32
from multidimensional scaling analyses and semantic dif
ferential studies of facial expressions. However, when
mood adjective check-list data have been factor analyzed,
the results have not supported the Schlosberg model.
Rather, the dimensions obtained represent specific cate
gories of mood, and these dimensions have frequently been
postulated in theories of emotion. Three moods that are
consistent between factor analyses and theories are joy
or cheerfulness, sadness, and anger. In addition to
facial expressions and mood adjectives, a third mode of
expressing mood state is postural cues. These cues have
been symbolized in stick figures, and while stick figures
have been used as test material, they have never been
subjected to a dimensional analysis.
CHAPTER III
METHODOLOGICAL CONSIDERATIONS
Multidimensional Scaling
The basic function of the data analysis is to
obtain the spatial representation of the mood expressions
in as small a space as possible. The outcome of a multi
dimensional scaling analysis is a matrix of numbers that
are the coordinates, projections, or loadings of the
stimuli on a set of orthogonal axes. A decision that can
not be made by any computer is choosing the number of
dimensions that are to be retained for interpretation.
It is a problem analogous to deciding on the number of
factors in factor analysis, and its resolution is accom
plished by trading off between statistical criteria and
psychological meaningfulness of the solutions at various
dimensionalities.
Techniques. The actual steps in progressing from
the raw data to the matrix of coordinates have changed in
recent years with new developments in multidimensional
scaling technology. The basic change has been in the
assumptions made about the properties of the raw data.
33
34
In what might be called classical or metric multidimen
sional scaling, it was assumed that the data were a ratio
or interval scale of euclidean distances (Torgerson,
1958). If the data represented a ratio scale, then the
steps in obtaining the factor matrix were straightforward
and did not involve any pre-processing of the data. In
such a case, the symmetric matrix of distances was con
verted to a matrix of scalar products, and the principal
axes of this matrix were obtained from its characteristic
roots and vectors. If it was assumed that the data were
on an interval scale of distance, then it was necessary
to solve for the additive constant before obtaining the
factor matrix (Messick § Abelson, 1956).
Because the actual multidimensional scaling pro
cedures were relatively straightforward, but because they
required strong assumptions of the data, much attention
was devoted to methods of obtaining data that would meet
the requirements of at least an interval scale. These
methods generally involved a Thurstonian-type scaling of
the raw data, especially if the data were ratings of the
similarity between pairs of stimuli (Torgerson, 1958).
The introduction of the so-called nonmetric multidimen
sional scaling techniques, however, have relaxed the
assumptions made about the raw data and, as a consequence,
have relegated some of the pre-processing of the data to
obscurity.
35
Shepard (1962a, 1962b) introduced these methods
with two papers that provided the rationale and an algo
rithm for implementing his rationale. He has been fol
lowed by several others who have developed new algorithms
for carrying out what is basically the same idea as
Shepard’s. That idea is: assume that the data is only
an ordinal scale of distance, but monotonically related
to the "real" distances, and find the monotonic transfor
mation of the raw data that leads to a spatial represen
tation whose distances best reproduce the order of dis
tances in the raw data. This idea contrasts with the
metric method which assumes at least an interval scale and
a linear relationship between the data and the real dis
tances , and which attempts to derive a configuration whose
distances reproduce the values of the raw data.
Of those who followed Shepard, Kruskal developed
a statistical definition of the best solution and an
algorithm that maximized the index of how well the trans
formed distances fit the raw data (1964a, 1964b). This
index was called Stress, and is a monotonic regression
analog to error variance. Torgerson (Torgerson § Mueser,
1962; Young § Torgerson, 1967) sought to avoid the problem
of local minima by first deriving the metric solution and
then applying monotonic transformations. McGee (1966) and
Guttman (1968) have developed other algorithms that differ
in computational technique, but have the same goal as the
Shepard and Kruskal techniques.
The greatest contribution of the nonmetric methods
is that, because of the relaxation of the assumptions made
about the raw data, virtually any numbers which can be
interpreted as being monotonically related to distances
will yield a solution with strong metric properties. That
this is so has been demonstrated with artificial data by
Shepard (1966) and Sherman and Young (1968). It has also
been argued that, while the various nonmetric procedures
differ in actual computational technique, they will yield
the same solution or solutions whose differences are
trivial (Young § Applebaum, 1968). With these methods it
is also possible to obtain a solution without a complete
data matrix. That is, when the number of stimuli is
large, it is possible to omit some of the entries in the
data matrix in order to save on experimental time.
Interpretation. Once the coordinate matrix is
obtained, psychological meaningfulness is usually sought
by examining the pattern of projections on each dimension,
and comparing adjacent stimuli for psychological similar
ity and contrasting disparate stimuli for psychological
dissimilarity. The sum of these impressions is then
translated into a label that describes the essential char
acteristic of the stimuli that accounts for their different
37
projections on the dimension. There are two problems
underlying this activity: the arbitrariness of the loca
tion of axes, and the subjectivity in choosing a label.
Neither problem is avoidable, and there is no easy solu
tion. The former issue relates to the rotation problem,
and any of the orthogonal rotation procedures that were
developed in the realm of factor analysis are equally
applicable in multidimensional scaling. However, the
notion of simple structure and the possibility of oblique
rotation are generally not suited to multidimensional
scaling, and very often it is the principal axes that are
interpreted.
As an aid in overcoming these problems, the rela
tionship between the projections of the stimuli on the
dimensions and outside variables is frequently examined.
Such was the case in several of the multidimensional
scaling studies of facial expressions, where correlations
between coordinates and scale values on the Schlosberg
dimensions were obtained. The fruitfulness of this tactic
would depend on the judicious choice of marker variables
through thoughtful examination of the stimuli.
The interpretation of dimensions isnot, however,
the only possible method of finding psychological meaning
in the configuration. It is very possible that the most
meaning is found in the shape or structure of the
38
configuration. That is, there may be clusters in the
space that are formed as a result of the stimuli in them
having a common property, and each separate cluster is
identified by a different class property. Torgerson
(1965) has described three kinds of structure that might
be found. One is a class structure as noted above, where
it is better to describe the classes rather than the
dimensions. A second is a class structure like the first,
but the stimuli can belong to more than one class. This
might be called a permeable class structure, and suggests
the possibility that relations among classes may have psy
chological meaningfulness. The third kind of structure
is one where there are classes, as in the first two, but
also dimensions that account for quantitative aspects of
the stimuli. In all of these structures, the search for
meaning in terms of dimensions may be quite unrewarding
and may, at best, be of secondary importance.
Implications. With the mood expressions that are
multidimensionally scaled here, there was a strong likeli
hood of obtaining a configuration that is predominantly
interpretable as a class structure. That is so because
the stimuli were selected to represent a small set of
mood qualities, and there were several expressions of each
of those mood qualities. This makeup of the sample of
mood stimuli may indirectly impose a judgmental context
39
whereby subjects are making ratings on the basis of
whether or not the two stimuli in each pair express the
same mood rather than making ratings of the similarity of
moods. The emphasis in the directions that the subjects
received was on the latter type of rating, and there was
no encouragement to make binary-type ratings. If it hap
pened that most ratings fell at the two ends of the scale,
the nonmetric methods would tend to collapse all those
stimuli that are rated as very similar into one point,
and the number of dimensions obtained would be one less
than the number of points or classes.
While there may be a prejudice in favor of obtain
ing a class structure, there are two aspects of the multi
dimensional scaling that would have important implications
for the relationships among mood expressions. One of
these is concerned with whether or not, if there is class
structure, the various modes of expression form separate
clusters within each mood class. A second concern would
be with the configuration of the mood clusters; that is,
what are the relative distances between the mood clusters?
Not to be forgotten is that the obtained configuration may
correspond to the third kind of structure described above,
and that the dimensions could be interpreted in terms of
quantitative components of mood. These dimensions may, in
fact, be comparable to the dimensions in Schlosberg's
model.
40
A relevant side issue in this study is the rela
tionship of the configuration of moods to the concept of
impression formation. This concept is usually applied in
the domain of person perception, and recent studies have
applied multidimensional scaling techniques to the problem
of trait inference (Rosenberg, 1968; Walters 8 Jackson,
1966). Person perception is generally conceived of as
pertaining to the personality domain, and attention has
been focused on impression formation as a result of review
articles that pointed out some of the methodological prob
lems in measuring accuracy of perception (Bruner 8
Tagiuri, 1954; Gage 8 Cronbach, 1955). It was suggested
in these reviews that one of the most important variables
affecting accuracy of perception might be the observer's
organization of personality variables. This organization
has been referred to as an implicit or lay personality
theory, and there is no reason to suspect that there are
not implicit theories of mood or emotion.
The analog is not without foundation, as it has
been noted that accuracy in mood perception is far from
perfect in experimental research (Brown, 1965; Davitz,
1964; and Ekman, 1965a). A contribution that a multidi
mensional scaling analysis of mood expressions can make
is in revealing the configuration underlying the various
moods and modes of expression. Inspection of the
41
configuration should indicate what moods are likely to be
confused with others and should indicate if there are com
binations of mood and mode that facilitate or hinder dis
crimination of different emotions. It seems quite reason
able to assume that discrimination of moods is related to
ratings of dissimilarity and distance between points in a
euclidean space.
Individual Differences
With regard to data that represent the ratings of
similarity between mood expressions, there is a possibil
ity that these ratings reflect different organizations of
the stimuli in a perceptual space. This is the problem
of individual differences, and recognizes that there may
be more than one implicit mood theory. If the data for
all persons were pooled into one set of data that repre
sented the mean ratings, then any individual differences
would be masked. Similarly, if the ratings for each per
son were analyzed separately, then any similarities
between persons would have to be determined from inspec-
i
tion of each solution. This approach may be time consum
ing if the sample size is at all more than ten or twenty,
and the solution for each person may be capitalizing on
unique or error variance. Consequently, methods have been
developed that seek to provide a compromise between these
two extremes of pooling data.
42
Techniques. The most widely used method for deal
ing with individual differences in multidimensional scal
ing is the "points of view" analysis (Tucker f j Messick,
1963). In this procedure, the complete matrix of raw data
is decomposed into the product of two orthonormal matrices
of characteristic vectors and a diagonal matrix of char
acteristic roots. One orthonormal matrix gives the load
ings of the stimulus pairs on factors, and the other gives
the loadings of persons on the factors. A "point of view"
is usually found by looking for clusters of persons in the
subject factor space and then deriving a set of data for
that "point of view" by averaging the ratings of the sub
jects in the cluster, or by defining an "idealized indivi
dual" in terms of factor loadings and then reproducing a
set of similarity judgments that corresponds to this
"idealized individual."
In addition to the problem of deciding how many
factors to look at in the person factor space, there have
also been problems in defining what is meant by a "point
of view." Ross (1966) suggests that, rather than a "point
of view" being a summarization of the judgments of persons,
the ratings given by each subject are a combination of
"points of view." This point of view about "points of
view" opens up the possibility that there may be negative
weights in the linear combination equation, and poses the
43
difficulty of interpreting a negative weight. A "point of
view" has sometimes been taken as one of the columns in
the orthonormal matrix of stimulus pairs. The loadings of
the stimulus pairs on a factor are treated as if they were
the estimates of distances between points for a "point of
view." Cliff (1968) has shown that such an interpretation
holds only for the first factor, and that this "point of
view" can be taken as an "average point of view." For
Cliff, a "point of view" is given by the spatial represen
tation of the stimuli obtained from the multidimensional
scaling of one set of data that represents the judgments
of several people. Both Cliff (1968) and Ross (1966) con
cur that there is a preference for analyzing data that are
derived from actual judgments and that can be shown to be
representative of real people. This implies that the
main function of the Tucker and Messick (1963) procedure
is to identify the clusters of people with very similar
judgments, and that the judgments can be summarized by
simply averaging the ratings of the people in the cluster.
Beyond these problems of identifying "points of
view," Gollob (1968) has suggested that real individual
differences are found if the raw data matrix of ratings is
converted to a matrix of interaction effects by removing
row, column, and overall mean effects. His FANOVA model
shows that variance on the successive factors is
44
contaminated by row, column, and overall mean effects if
it is the raw data matrix that is factor analyzed. There
have also been two proposals for handling individual dif
ferences that avoid the necessity of finding clusters of
people or "points of view." Carroll and Chang (1969)
extend the Eckart-Young decomposition of a two-way matrix
to the decomposition of a three- or N-way matrix. For
multidimensional scaling data, the three-way matrix would
correspond to the distance matrices of the individuals.
The analysis yields two matrices: one gives the coordi
nates of the stimuli on a set of dimensions and the other
gives the weights for each subject on each of the dimen
sions in the stimulus space. Individual differences are
noted by there being a different set of weights for each
person, and similarity between persons would be indicated
by similar weights. At present, this analysis only allows
for individual differences that take the form of either
different number of stimulus dimensions or differential
stretching or shrinking of the stimulus dimensions. It
does not allow for the possibility of a stimulus dimension
for a particular person being a combination of two or more
stimulus dimensions for another person. This may be a
remote possibility, so there is not so much a drawback by
it not being available. McGee (1968) proposes a way of
treating individual differences in nonmetric multidimen
sional scaling. It is similar to the Carroll and Chang
45
(1969) method in that it does not seek clusters of people,
but it is different in that it does not yield the subject
by dimensions weight matrix. Rather, McGee considers
four ways of treating the data for each person, and
these ways are the four possible combinations of two
choices: one choice is whether or not the same monotonic
transformation is applied to each set of data, and the
second is whether or not the final coordinates are to be
the same for each person. This approach leaves it up to
the investigator to find similarities and differences
among the persons by examining the spatial representations
found for each person.
Outside the area of multidimensional scaling,
there are several methods for dealing with individual dif
ferences that fall into the category of cluster analysis.
These techniques typically are concerned with clustering
people together on the basis of profile similarity, where
a profile is the set of scores for a person on a set of
variables. These methods are commonly applied to scores
on scales of the MMPI or on scales of the Wechsler Intel
ligence Scales, with a goal of finding clusters of people
that might be related to clinical groups. There are some
differences in this type of approach to individual differ
ences and that in multidimensional scaling; One is that
it is possible to find some meaning in the clusters by
46
examining the variables on which the individuals are being
clustered. That is, the sub-scales of the MMPI or the
Wechsler scale are inherently meaningful and descriptive,
whereas the variables in a profile of similarity ratings
are only pairs of stimuli and do not have the same salien-
cy for interpretation. A second difference is that the
cluster analysis procedures are ends in themselves, but
in multidimensional scaling the search for clusters is
only a means to an end. That is, it is an intermediate
step that is taken as a safeguard against overlooking
different organizations of the stimuli that would be con
cealed by the multidimensional scaling of an average set
of ratings.
The relevance of cluster analysis procedures in
multidimensional scaling is that some of the problems
that have been pointed out in cluster analysis are per
tinent to the treatment of individual differences in
rating data. These problems as well as several methods
of cluster analysis are presented in Lorr and Lyerly
(1967). One of the major problems is deciding how to
define similarity between profiles. With regard to this
problem, there have generally been two types of profile
similarity: one that indicates pattern similarity and
another that indicates level similarity. Pattern similar
ity is usually given by a correlation, covariance, or
47
even just a cross product. The Tucker and Messick (1963)
procedure takes one of these measures as the index of pro
file similarity. These measures do not, however, take
into account differences in level of profiles, and it may
be that this information should not be overlooked. A mea
sure of relation between profiles that is based on level
has been proposed by Osgood and Succi (1952). Their index
is the distance between profiles, and a routine for clus
tering individuals on the basis of this measure has been
given by Ward (1963; Veldman, 1967). While the Ward pro
cedure (1963) is not the only one available, it does have
intuitive appeal for treating rating data, in that it will
seek to cluster together subjects whose ratings are most
like one another, and it yields an index of how homoge
neous the ratings in a cluster are.
Interpretation. Beyond the problems inherent in
just finding individual differences or clusters, there is
the additional task of interpreting a "point of view."
As mentioned earlier, Cliff (1968) looks for meaning of
individual differences in the multidimensional scaling
solutions for the various sets of data. For others,
though,, interpretation of individual differences is sought
by relating dimensions in the person factor matrix to
outside variables. For example, Pedersen (1965) found
that differences in perceived personality trait
48
relationships were related to authoritarianism, but
Walters and Jackson (1966) did not find any personality
or cognitive style correlates of individual differences in
trait inferences. Messick and Kogan (1966), in a multi
dimensional scaling of role constructs, found some corre
lates of individual differences with personality and cog
nitive style variables, but the size of the relationships
were generally small. These results are fairly typical
of attempts to determine the nature of individual differ
ences in multidimensional scaling of personality vari
ables. It appears that there has not been any attempt to
relate individual differences in the perception of mood
similarity to marker variables.
The major function here of the individual differ
ence analysis is not to find correlates in the personality
domain, or correlates with cognitive styles. Rather, the
primary purpose is to ensure that the ratings of each per
son are represented in the sets of data that are submitted
to a multidimensional scaling analysis. Stating this in
another way, a limited number of vectors of judgments will
be sought such that they represent a best fit of the data,
even if that number is only one. If it is more than one,
then interpretation of the differences will be based on
the results of the multidimensional scaling.
49
Summary
Multidimensional scaling techniques have pro
gressed from the metric procedures to the so-called non
metric procedures. The advantage of nonmetric multidi
mensional scaling is that very robust results are obtained
even though weak assumptions are made about the relation
ship between the raw data and the true euclidean dis
tances. The problem of deciding how many dimensions to
retain is seen as a compromise between statistical cri
teria and psychological meaningfulness. Psychological
meaningfulness may be found in either interpretable dimen
sions or in clusters of stimuli that have a common prop
erty. When similarity ratings are replicated over per
sons, there is a possibility that these individuals have
different spatial representations of the stimuli. To
determine if individual differences do exist, the raw
data can be analyzed in terms of both pattern similarity
and level similarity.
CHAPTER IV
METHOD
Construction of Booklets
Selection of stimuli. The mood expressions were
selected so as to fit into a completely crossed paradigm
of four types of mood and four types of mode of expres
sion. For each of the 16 combinations of a mood and mode
type there were to be three stimuli. The availability of
appropriate mood expressions led to the prior determina
tion of the four types of mode: mood adjectives, stick
figures, line-drawn faces, and photographed faces. The
determination of the four mood types depended on searching
the source materials and selecting four mood types that
were well represented in these sources.
As a starting point, ten lists of mood adjectives
were drawn up as a guide in finding appropriate stimuli
for the other modes of expression. Each list corresponded
to a mood dimension that had been found in the factor-
analytic studies of Nowlis (1965), Borgatta (1961), McNair
and Lorr (1964), and Lorr, Datson, and Smith (1967). The
members of each list corresponded to adjectives that had
50
51
defined the factor in one or more of the above studies.
The source materials from which the other stimuli would be
culled were tests that had been constructed or adapted for
use by the Aptitudes Research Project, University of
Southern California.
There were three steps in selecting the moods and
the stimuli that would be included in the final sample.
First, taking each list of adjectives, the above source
materials were searched for visual stimuli that seemed to
express the mood state indicated by the adjectives, and
these stimuli were noted for subsequent selection. This
was repeated for each of the ten lists. At this stage it
was possible to delete five of the ten mood types from
further consideration, because no appropriate stimuli were
found or not enough stimuli were found. The five that
remained could be characterized as moods of Anger, Cheer
fulness, Sadness, Anxiety, and Concentration. The second
step involved comparing the lists of appropriate stimuli
across moods to eliminate stimuli that appeared in more
than one list. At this time Concentration was dropped
from further consideration because there was considerable
overlap with the other moods. For the third step the
remaining stick figures, line-drawn faces, and photo
graphed faces were attached to sheets of paper, one sheet
for each mood type, and presented to staff members of the
Aptitudes Research Project. They were to indicate for
each mode three stimuli that best expressed the mood indi
cated by three adjectives listed at the top of the page.
The adjectives had been selected by E on the basis of
their describing common mood states and by their factor
loadings in the four studies mentioned earlier. The most
frequently chosen stimuli for the other modes were
selected as the final sample of mood expressions to be
included in the booklets. This final sample of 48 expres
sions appears in Figure 1, and the numbering of the
stimuli is carried through in the data analyses and dis
cussion of the results. There are 13 male faces and 11
female faces, while all stick figures are male. The pho
tographs were posed for mainly by college students, while
the drawn faces exhibit more variety of ages.
Sampling of stimulus pairs. If all possible pairs
of the 48 stimuli were formed, there would be 1128 combi
nations of mood expressions, not counting any repeated
administration of some of the items. It was desired to
eliminate some of these for two reasons: testing time
and balance of expected ratings of dissimilarity. The
former consideration relates to maintaining subjects in a
proper state of motivation while completing the experi
ment. Previous exploratory work had shown that college
students could complete a booklet of 120 pairs in 15-30
53
1. angry
2. annoyed
3. grouchy
13. anxious
14. fearful
15. nervous
4.
16.
17. 18.
19.
25. cheerful
26. gay
27. happy
37. depressed
38. gloomy
39. sad
28.
29.
30.
40.
41.
42. 43.
Fig. 1. Mood expressions used in this study.
in this study
54
minutes. To keep testing time to less than one hour, a
booklet of 240 or 260 items, including repeated items,
seemed to be the maximum length for one experimental ses
sion. The latter consideration led to the particular
sampling of items that would be excluded. To do this,
each of all possible combinations was conceptualized as
being one of four types of combination: same mood-same
mode, same mood-different mode, different mood-same mode,
and different mood-different mode. The number of possible
pairs of each of these types is presented in Table Al(a),
in the Appendix. Since the ratio of different mood pairs
to same mood pairs is greater than 3:1, and since these
pairs would most probably lead to ratings of greater dis
similarity, it was decided to present only a sample from
this set of combinations. In this way, there would be a
better balance between pairs that were expected to be low
in dissimilarity (same mood pairs) and pairs that were
expected to be high in dissimilarity (different mood
pairs).
The sampling strategy is best described by a re
conceptualization of the stimulus pairs. Looking at these
pairs as the above diagonal intersections of a 48 x 48
matrix, it is easy to see that there are six blocks of
pairs that represent different mood combinations. Within
each block there are 144 stimulus pairs, and these can be
55
further broken down into 36 different mood-same mode pairs
and 108 different mood-different mode pairs. In each of
the six blocks the sampling strategy was the same: delete
four different mood-same mode pairs and 18 different mood-
different mode pairs. For each block, then, 22 of the
possible pairs would not be included in the final sample
of stimulus pairs, and summing across blocks, 132 pairs
would be deleted from the total of 1128 possible pairs.
This sampling scheme is summarized in Table A2, in the
Appendix, where the smallest block represents a three-by-
three matrix of stimuli, and the final number of types of
combinations appears in Table Al(b). The percentages that
appear in this table were put to further use in determin
ing the composition of each booklet and each page, and
in determining the types of stimulus pairs to be repeated.
Composition of booklets. Having decided on the
final sample of stimulus pairs, it remained to form four
separate booklets for actual administration. Rather than
use a strategy of random sampling of stimulus pairs, a
strategy of stratified sampling was adopted to compose
each booklet. The stratification followed the percentages
of each type of stimulus combinations indicated in Table
Al(b). Conveniently, the numbers in this table are even
multiples of four, so the number of each type of combina
tion that appears in each of the booklets is one-fourth
the numbers in Table Al(b).
Having mapped out this scheme, it remained to
select the actual items for each booklet. Initially a
random sampling approach was tried out but was discarded
when it was found that the frequency of appearance was not
equal for all stimuli. In general, the selection of items
for each type proceeded sequentially, taking care to
ensure near equal selection of the stimuli and the various
mood and mode combinations. Determining what items would
go on a page followed these considerations also.
Repeated items. Since there were 249 items
selected for each booklet and since 20 items fit on one
page, there was room for 11 more items to complete a
13-page booklet. These 11 items were used to get retest
information on the rating task and on the subjects them
selves. The proportions in Table Al(b) were applied to
the total number of items to be repeated (44) to arrive
at the number of each type that would be repeated. These
numbers appear in Table A1(c), and one-fourth of each
number gives the frequency in each booklet. The selection
of the repeated items occurred after the selection of the
249 items and followed the same guidelines. The location
of the repeated items in the booklets was two pages after
its initial appearance.
57
Subjects
Volunteers were solicited from the Introductory
Psychology course at the University of Southern Califor
nia. The students in this course are mainly freshmen and
sophomores of college age from upper middle class, Cauca
sian families. No one was rejected from participation
because of sex, color, or nationality. Complete data were
obtained from 35 participants--19 males and 16 females.
Administration
The booklets were administered in group sessions
over a six-week period of time: the first two booklets
during two weeks preceding Christmas vacation and the last
two booklets during the two weeks following the recess.
The students could come to any one of three sessions in a
given week, and completed one booklet during that time.
Most subjects followed the sequence of completing one
booklet per week, but some students started after the
vacation and had to complete two booklets in each of the
last two weeks. No one was allowed to work on more than
one booklet on a test day. The sequence of administration
was constant for all subjects.
Prior to administration of the first booklet each
student was given a sheet of instructions that described
the task. The directions (see Appendix) stated that they
were to make judgment of the dissimilarity of the pair of
58
mood expressions in each item, and assign a number to
their judgment on a rating scale that ranged from one to
nine. The group was then given two minutes to work on a
set of practice items that were of the same nature as in
the booklet. They were told to practice assigning numbers
to their judgments of the dissimilarity of the moods that
they perceived in the two expressions in an item, but
they were not required to record their ratings. Following
this the subjects began work on the booklets. Each book
let contained 13 pages, with 20 items on a page (one item
being a pair of mood expressions). The ratings were
recorded on a separate answer page that had the numbers
one through nine printed opposite each item number. For
the succeeding booklets each subject began work as soon as
he reported to the testing room. He was encouraged to
re-read the directions after vacation, but was not
required to look at more practice items. The rest of the
procedure followed that described above for the first
booklet.
Data Analysis
The data that were collected consisted of the dis
similarity ratings of 35 subjects for 1040 pairs of mood
expressions, 44 of these pairs being repeated items. All
or part of this data was used in four major types of
59
analysis: retest reliability of people, euclidianness of
individual ratings, clustering of people, and multidimen
sional scaling.
The data from the repeated items were used to
obtain information regarding the stability of the subjects
which could be used to identify and delete outliers. This
seemed to be the only formal manner of removing from the
sample subjects who might have been responding in a random
manner. Accordingly, the data were analyzed in two ways.
First, a deviation score was obtained for each person on
each of the 44 repeated items. This deviation score was
the difference between the difference of the person's two
ratings and the difference of the two means. Symboli
cally,
xij = C^ijl"xij2^ “ (x.jl'x.j2)
where
Xiji = rating of person i on the first appearance
of item j
Xij2 = rating of person i on the second appearance
of item j
X * mean rating of..first. appearance of item j
X.j2 = mean rating of second appearance of item j
While these scores were not statistically evaluated, it
was felt that if someone was consistently random, then it
would show up in an examination of his deviation scores.
60
The second way in which the data from the repeated items
were analyzed was to obtain a correlation coefficient for
each person. This correlation was between the test and
retest ratings, taken over 44 items. These also were
examined to see if there were any people with severely
low correlations, or with correlations that departed from
the range of correlations for the majority of the sample.
Once the above analyses were completed, the retest
ratings were deleted from the data matrix. All succeeding
analyses used only the data from the 996 items. The
second type of analysis was to check on the euclideanness
of the individual ratings by computing the number of vio
lations of the triangular inequality by each person.
Since one of the axioms of the euclidean model is that
dijfdik + djk, it seemed essential to know the extent that
this principle was violated. A high number of violations
would indicate that the person was responding in a pecu
liar manner (not necessarily random) and would make it dif
ficult to obtain a euclidean space of small dimensionality
in the multidimensional scaling analysis. Because of the
known error in individual ratings it was decided to count
as a violation of the triangular inequality all those
cases where dxj-d^k + dj^ + 1.0. Excluded from this count
were instances when no estimate of dij, dfk, or djfc was
obtained.
61
The third major analysis involved searching for
sub-sets of people whose ratings were more like one
another than the ratings of people not in the sub-set.
This, of course, is the individual differences analysis,
and a procedure was set up for defining the clusters of
people whose data would be averaged and treated separately
in the multidimensional scaling. The strategy was to per
form separate individual differences analyses on the rat
ings from each of the four booklets, compare the clusters
across the booklets, and then determine the final grouping
on the basis of consistency across booklets. To guard
against the possibility of getting method-dependent clus
ters, two types of cluster analysis were carried out for
each booklet: one based on level similarity and the
other based on pattern similarity.
In the former case the similarity between persons
is taken as the distance between the profile of similarity
ratings, one index being computed for all pairs of per
sons. Ward's (1963) procedure for determining the optimum
grouping of persons, as programmed by Veldman (1967), was
applied to the data from each booklet. The distance func
tion Ward used was modified by substituting the sum of
absolute differences for the sum of squared differences.
It was felt that this modification would reduce the
effects of large differences that reflected error rather
62
than perceptual differences.
The second type of cluster analysis was the Tucker
and Messick (1963) procedure. The correlation between
profiles was taken as the index of pattern similarity, and
one correlation matrix was obtained for each booklet. The
steps followed in deriving the clusters for each booklet
were to factor the correlation matrix with unities in the
diagonal, obtain the person factor matrix for an appropri
ate low dimensionality, and then use Ward's technique for
finding clusters in the person factor space. This latter
step was taken so as to avoid, or diminish, the subjective
decisions that are usually required in finding clusters in
the person factor space.
The net result of these two types of cluster anal
ysis was four sets of clusters derived from level similar
ity and four sets of clusters derived from pattern simi
larity, where a set refers to the results for one booklet.
Using this information, individuals were placed into a
final grouping on the basis of their consistency in
appearing in the same cluster across booklets and type of
cluster analysis. At this point, the data were prepared
for the fourth major analysis by obtaining within each
group (cluster) the mean rating for each of the 996 pairs
of mood expressions.
The last stage was the multidimensional scaling
analysis of the similarity ratings for each group. These
were carried out with TORSCA (Young § Torgerson, 1967), a
nonmetric program that is based on the rationale proposed
by Torgerson and Mueser (1962) and Shepard (1962a, b).
Incorporated into this computer program is Kruskal's
Stress index (Kruskal, 1964a, b), a goodness of fit index
that indicates the extent to which the order of the
reproduced distances fits the order of the raw data. The
dimensionality of the spatial representation for each
group's similarity ratings was decided by taking into
account both the goodness of fit index and the psychologi
cal meaningfulness of the configuration of mood expres
sions. In addition to the multidimensional scaling analy
ses for each group, a configuration was derived for a set
of similarity ratings that was obtained by averaging over
all subjects. This spatial representation was used in
evaluating the similarities and differences among the
group configurations by treating it as a target matrix,
and then orthogonally rotating each of the group matrices
to a least-squares best fit to it (Cliff', 1966). .
CHAPTER V
RESULTS
Repeated Measurements
Stability of persons. Across the four booklets of
stimulus pairs there were 44 items that were repeated.
This information was used to determine if there were indi
viduals in the sample who appeared to be giving random
ratings. The computer program that gave various statis
tics for the repeated measurements prepared a table that
listed a deviation score for each person on each repeated
item. This deviation score was the difference between
two other deviation scores: the difference between the two
ratings given by the person, and the difference between
the two mean ratings. These deviation scores were
examined to see if there were several high deviations,
indicating that the person might be responding in a random
fashion. This information did not lead to deleting any
individuals from the sample, as no person had consistent
high deviation scores.
To summarize the stability of the ratings for
each person, test-retest correlations were obtained over
the 44 repeated items, and are presented in Table A3 in
64
65
the Appendix, along with other descriptive statistics.
The individuals are arranged in this table according to
the groups into which they were placed as a result of the
analysis of individual differences. These correlations
are quite good, with several being .90 or greater.
Although there are a few individuals with correlation
coefficients less than .70, it was not deemed necessary to
exclude them from further analysis.
One other item in Table A3 was considered in
evaluating the goodness of individuals' ratings, and this
is the number of violations of the triangular inequality
(VTI). The entries in the table are percentages, and
range from a low of 0.9% to a high of 13.9%. If there
were individuals whose ratings gave rise to a large number
of violations, then it would be desirable to eliminate
them from the sample, as their data would not lead to a
good fit in a euclidean space. Again, however, it was
not deemed necessary to delete anyone from further
analyses.
Individual Differences
Repeated measurements. Although the two cluster
ing procedures were performed on the ratings to all items,
an analysis of variance was carried out on the data from
the 44 repeated items. In this analysis there were two
66
main effects--persons and items--and two replicates--the
two ratings to each item. The results are presented in
Table 1. It was expected that the main effects due to
TABLE 1
Analysis of Variance for Repeated Measurements
Source df MS F
Persons (A) 34 31.38 21.22*
Items (B) 43 348.77 235.88*
A X B 1462 5.14 3.48*
Within
Replicates 1540 1.48
*p<.01.
persons and items would be statistically significant, but
of primary concern was the interaction effect. A very
large ratio would indicate that there were substantial
individual differences in the pattern of ratings across
the items, even allowing for the error estimated by the
within-cell variation. Although the F ratio obtained for
the interaction term is significant, it is not as large
as would be desired. The obtained F ratio for the inter
action indicates that there are some differences in pat
terns, and the differences are statistically significant,
but they are not overwhelming. This result, based on just
67
44 items, is indicative of further analyses which also
point out that there are some individual differences, but
there is more communality than not.
Cluster analysis of distances. In this first
method of cluster analysis, the index of profile similar
ity between all pairs of persons was the sum of absolute
differences between the 249 ratings in each booklet. The
computer program that attempts to determine the optimum
grouping of persons initially defines each person as a
one-man group. The algorithm proceeds by combining groups
that lead to a minimum increase in the within group devi
ations. It proceeds until there are only two groups left,
but at each step the program lists the average deviation
associated with the new group it has just formed. It is a
matter of judgment as to where an appropriate cut-off
should be made, on the basis of the increasing size of the
deviations. The strategy taken was to examine the devia
tions and find a point at which the average deviation
increases sharply as a new group is formed. Applying this
guideline to the cluster analyses for each booklet, the
number of groups retained in Booklets 1, 2, 3, and 4 was
5, 6, 6, and 5, respectively. The degree to which the
groups formed in the first booklet are consistent over the
remaining three booklets is illustrated in Table A4, in
the Appendix. A chi-square test of the independence of
68
clusters was performed for the three pairs of clusterings
in Table A4. While the chi squares are significant at the
.05 level or better, there was not as much consistency as
it was hoped there would be. Before deciding on an over
all grouping on the basis of the cluster analysis of dis
tances, an alternative method of dealing with individual
differences was explored.
Cluster analysis of factor scores. The alterna
tive method is based on the Tucker and Messick (1963) pro
cedure. This method began with a factor analysis of the
correlation matrix of persons. A factor score matrix was
determined from the three largest factors in each booklet.
The pattern of eigenvalues in each of the four correlation
matrices was quite similar: there was one large value
followed by two with values around 1.0, and then a sharp
break between the third and fourth eigenvalues. The fac
tor score matrix was then submitted to the clustering
program which determined the optimal grouping of persons
in the factor space. The consistency of grouping across
the four booklets was again evaluated with chi-square
tests, and the three chi squares were all significant at
the .05 level or better. But, as with the cluster analy
sis of distances, the groups were not so consistent as to
make the task of forming overall groups a simple and
objective procedure.
69
Final grouping. The evidence to this point
regarding individual differences was somewhat murky.
While it was clear that there was a great deal of commu-
nality among the ratings, it was also evident that there
were some differences that could possibly lead to con
trasting perceptual spaces in the multidimensional scaling
analysis. Therefore, it was decided to go ahead with the
formation of different groups of persons, using the evi
dence from the two types of cluster analyses across the
four booklets. The crucial factor in making this decision
was that in the cluster analyses there were some persons
who were frequently grouped together by both procedures,
and many persons who were never grouped together by either
method. It was felt that it was less harmful to split (the
persons into different groups than to average the data
over all persons and risk overlooking some important indi
vidual differences. If in fact there are no substantive
differences between the groups, then the multidimensional
scaling analysis would show that the perceptual spaces for
the group data are very similar to one another.
The final grouping of the persons was subjectively
determined by considering the evidence from the cluster
analyses of the data for each booklet. While an attempt
was made to ensure that each person was placed into a
group, there were three persons who did not conveniently
70
belong in the five groups that were formed. For two of
these individuals there was nothing noteworthy about their
descriptive statistics to indicate their peculiarity, but
the third person had the highest percentage of violations
of the triangular inequality. Table A3 lists the persons
according to their membership in the five groups that were
formed.
One statistic in Table A3 that has not been men
tioned yet is the mean rating for each person, taken over
996 ratings. Although the cluster analysis was not per
formed on these statistics, it is fairly clear that there
are differences between the groups in terms of the means
of the individuals in the groups. Differences among indi
viduals in overall mean rating could reflect both percep
tual differences and a scale factor. That is, some indi
viduals may be more or less disposed to give extreme
ratings. For instance, the individuals in the first group
tended to give ratings of one and nine (the end points of
the rating scale) much more frequently than the other per
sons. The use of the end points of the rating scale is
reflected in both a high mean and a high standard devia
tion. It still remains a question, though, whether or not
this characteristic is due to perception of mood similar
ity or to a response scale factor. It could very well be
that the individuals in the first group tended to see the
71
mood expressions as either very similar or very dissimi
lar, whereas the other individuals perceived fewer
extremes of similarity and dissimilarity.
Table 2 presents some information regarding the
TABLE 2
Correlations among Groups (all items--above diagonal;
repeated items--below diagonal; reliabilities--
diagonal) and Descriptive Statistics
Group
Group
1 2 •3 4 5 AVE
1 .97 .80 .92 .88 .88 .96
2 .84 .92 .86 .84 .81 .89
3 .93 .91 .98 .92 .88 .98
4 .86 .86 .92 .95 .87 .96
5 .90 .81 .87 .87
•M
.93
AVE .96 .92 .98 .95 .93 .99
Mean 6.1 4.9 5.3 5.2 5.5 5.5
s .d. 2.6 2.3 2.6 2.2 2.1 2.3
Violations of Triangular
Inequality 2.0% 6.6% 6.9% 1.9% 2.1% • •
N 8 4 7 9 4 32
72
mean ratings for each group. It should be noted that mean
ratings were also obtained by averaging over all persons,
excluding the three who were not placed into a group.
These data are referred to as AVE. Among the several sta
tistics in Table 2, of particular importance are the reli
abilities for each group and the correlations among the
groups based on the 44 repeated items. It should be noted
that in all cases the correlation between two groups is
never equal to either of the reliabilities. Although all
the correlations are substantially high, there is some
reliable variance in each group that is not common to
another group. It should also be noted that although the
correlations were based on just 44 items they should pro
vide good estimates of the correlations based on 996
items.
Multidimensional Scaling
Dimensionality. Once the mean rating to each of
the 996 items was obtained for each group, the data were
ready to be submitted to the multidimensional scaling
program. The first consideration in this analysis was to
determine the number of dimensions needed to represent the
i similarity ratings. It was hypothesized that a three-
dimensional solution would best account for the data, with
the 12 expressions for each mood quality forming homoge-
| neous clusters in the three-space.
73
The TORSCA program requires that the user provide
limits on the number of times that the program will iter
ate in trying to find the transformation of the raw data
that leads to the best fit in a given dimensionality. For
each of the six sets of ratings (five groups plus AVE),
these limits were the same. It is also necessary for the
user of TORSCA to specify the range of dimensionalities
in which the program will try to fit the data, and initial
ly this range was set at three-five. To aid in deciding
on the number of dimensions, TORSCA lists Kruskal's
Stress index at each dimensionality. This index is a
goodness of fit value, expressed as a proportion, and the
lower the proportion the better the data fits in that
dimensionality. Consideration of this index, however, is
not the sole criterion, and it is necessary to examine the
matrix of projections on the dimensions for the psycholog
ical meaningfulness of the configuration.
The strategy adopted in deciding on the number of
dimensions was to first narrow the range of possible
dimensions by examining the plot of Stress against dimen
sionality. During the course of this analysis it devel
oped that as the computer program stepped down from a
higher dimensionality to a lower dimensionality the pro
jections progressively shrunk in size. This was an unde
sirable effect, and to eliminate it further analyses were
carried out one dimensionality at a time.
74
In addition to repeating the initial analyses of
three to five dimensions, further analyses were performed
at two and six dimensions. Table 3 contains the Stress
Table 3
Stress Values
Dimensionality
Group
1 2 : 3 4 5 AVE
2 .141 .202 .159 .167 .179 .129
3 .098 .162 .123 .132 .140 .097
4 - .082 .139 .107 .113 .118 .081
5 .068 .124 .093 .098 .101 .068
6 .108 .084 .083 .091 .060
Note.--No solution where -- appear.
values at each of these dimensionalities for the six sets
of data. A first look at this table reveals that the
Stress values at three dimensions are not particularly
close to zero, and that they are not nearly equal for all
groups. However, as with examining eigenvalues in factor
analysis, it would be a rare occasion when the "correct"
number of dimensions can be inferred from just absolute
values. It is necessary to consider the size of the
Stress values in relation to the various dimensionalities,
75
and look for a break in the plot of Stress values against
dimensionality. With this strategy, it is seen that for
Group 1 and AVE there is a definite break between two and
three dimensions, and a gradual leveling off beyond three
dimensions. For the other groups there is not the same
degree of difference between the Stress values for two and
three dimensions in relation to the difference between the
values for three and four dimensions.
It was felt, however, that a more significant
indicator of the proper number of dimensions would be the
psychological meaningfulness of the configurations.
Accordingly, the three-dimensional solutions were examined
first in order to provide a basis to which the other con
figurations could be compared. For every group the three-
dimensional solution compared very favorably to a hypothe
sized configuration of four clusters of points, each
cluster composed of the 12 expressions for one mood
quality.
When the other configurations were considered, it
was immediately apparent that the five-dimensional solu
tion was not at all reasonable. In this solution the
clusters that had appeared in three dimensions were no
longer maintained. Because the Stress values for Groups
2, 3, 4, and 5 indicate that four dimensions might be the
"correct" number, these solutions were carefully
76
considered. There was considerable' similarity between
the projections on the first three dimensions in both the
three-and four-dimensional configurations. The projec
tions on the fourth dimension, however, did not lead to
reasonable interpretation for any of the groups. Further
more, while there was some consistency across groups in
the projections on the first three dimensions, there was
very little on the fourth dimension. Although the Stress
values for two dimensions are rather high, particular
attention was given to these solutions since previous
research had suggested that ratings of similarity of mood
could be accounted for by two dimensions. When these
solutions were considered, there was, as with the three-
dimensional solutions, a good deal of consistency among
the groups, and the configuration was interpretable. A
major drawback to the solutions, however, was that the 12
expressions for the mood quality of Fear-Anxiety did not
form a distinct cluster as they did in the three-
dimensional solutions. It was the third dimension on
which these expressions had their highest projections, and
eliminating this dimension apparently led to the breaking
up of the cluster.
Since there was no compelling reason to atcept a
solution in two, four, or five dimensions, the three-
dimensional configuration was retained for each group. An
77
additional source of evidence for the validity of the
three-dimensional solutions became available after the
above decisions had been made. This evidence was provided
indirectly by a computer program that performs a metric
multidimensional scaling analysis. The matrix of similar
ity ratings for each group was analyzed in accordance with
the procedure in Torgerson (1958), where the data are
treated as being a ratio scale of euclidean distances.
One step in this program computes and lists the eigen
values of the matrix of similarity ratings. The first
ten eigenvalues for each group are presented in Table 4.
TABLE 4
Eigenvalues
Eigen
values
Group
1 2 3 4 5 AVE
1 10.676 13.931 13.564 13.399 11.076 12.384
2 7.635 4.493 5.782 4.045 5.524 5.718
3 2.642 2.389 2.378 1.862 2.140 2.111
4 1.119 1.905 1.336 1.480 1.523 1.109
5 0.937 1.697 1.248 1.251 1.372 0.938
6 0.918 1.620 1.205 1.216 1.338 0.895
7 0.827 1.556 1.164 1.195 1.147 0.853
8 0.791 1.470 1.109 1.110 1.069 0.774
9 0.709 1.359 0.948 0.956 0.967 0.705
10 0.664 1.302 0.879 0.936 0.935 0.609
78
Looking at the difference between successive values leads
pretty conclusively to the decision that the "correct”
number of dimensions is no more than three.
Rotation. The matrix of projections that is
obtained from TORSCA is a Varimax rotation of the princi
pal axes. Because there was a separate matrix for each
group, it was desirable to eliminate any differences among
the groups due solely to rotational differences. To pro
vide a common frame of reference for the five coordinate
matrices, the coordinate matrix from AVE was taken as a
target, and an orthogonal, least-squares rotation (Cliff,
1966) computer program was used to rotate the other five
matrices to it. This procedure enabled the five groups
to be compared to each other on a fair and equal basis,
and also permitted a contrast between AVE and the groups.
The matrices obtained from these rotations are presented
in Table 5, along with the matrix for AVE.
Although there is a wealth of numbers in this
table, close examination of it reveals that there is a
great deal of consistency among the six matrices. In
order to facilitate verbalization of the numbers in the
table, differences among the six matrices will be tempo
rarily ignored. Furthermore, two somewhat different
aspects of the matrices need to be highlighted: the
dimensions themselves, and the clusters of mood
79
TABLE 5
R o ta te d C o o rd in a te M a t r ic e s
Grc up
Stimulus 1 2 3 4
1 2 3 1 2 3 1 2 3 1 2 3 1
1 -14 -67 -07 06 -42 10 -07 -57 02 01 -54 02 -02
2 00 -58 -05 10 -27 03 11 -44 07 13 -45 02 07
3 03 -57 -29 09 -32 -20 15 -48 -26 26 -46 -14 16
4 -11 -63 -07 -05 -27 -21 -11 -41 -08 -06 -31 -09 -16
5 -16 -59 -25 -22 -26 00 -15 -31 -23 -19 -23 -24 -17
6 -27 -57 -12 -13 -43 -14 -06 -63 -06 -23 -44 -10 -07
7 12 -51 -15 19 -23 -22 21 -34 -13 23 -32 -16 11
' 8 -04 -65 -18 10 -38 01 14 -47 -10 20 -37 04 10
9 05 -59 -16 11 -36 06 11 -46 -06 16 -40 05 04
10 -16 -66 -04 -05 -44 04 -07 -56 07 02 -49 15 -15
11 -12 -55 15 -05 -21 17 -10 -42 16 -05 -42 17 -12
12 -08 - 66 -12 13 -39 -12 05 -54 -07 22 -42 -04 12
13 -09 00 49 ' "-'96 " "96 06 -14 -03 45 -10 -03 29 -19
14 19 09 60 06 04 35 14 10 52 17 05 47 18
15 03 08 48 -02 -03 19 -01 12 44 14 -14 31 -03
16 22 03 18 24 -05 16 15 15 07 12 04 -31 15
17 40 08 17 24 15 01 33 03 25 27 02 26 28
18 -01 -23 25 -07. -26 22 04 -23 17 -01 -16 31 -14 -
19 23 30 16 13 30 -06 -01 32 -01 05 22 10 04
20 08 03 63 01 11 39 14 14 42 05 10 43 07
21 19 25 47 18 31 11 21 23 34 22 12 27 20
22 36 19 30 23 -03 03 23 01 23 22 -01 16 29
23 -27 25 39 05 -03 -11 -21 -05 03 -21 -09 -01 -19 -
24 12 IS 47 09 -08 26 08 -01 35 -09 19 25 -13
25 -56 " ' 4 9 ' - 3l -49 44 -13 -SO 42 -35 ' -T7 44 -26 -53
26 -55 38 -35 -58 31 -08 -56 33 -36 - 55 41 -25 -55
27 -59 38 -23 -48 34 -25 -58 41 -23 -61 33 -23 -54
28 -53 35 -05 -36 -03 04 -52 18 -09 -44 33 -11 -10
29 -49 27 -36 -32 21 -11 -54 41 -08 -53 35 -04 -21
30 -59 26 -21 -28 31 19 -57 31 -10 -46 34 -33 -54
31 -57 34 -29 -54 27 -12 -58 27 -25 -54 29 - 29 -49
32 -50 44 -32 -40 18 -34 -45 46 -27 -43 44 -16 -48
33 -59 37 -20 -31 33 -32 -53 44 -16 -56 35 -16 -53
34 -60 35 -30 -45 43 -22 -51 43 -24 -52 42 -17 -47
35 -57 36 -32 -52 31 -29 -53 35 -30 -54 33 -29 -49
36 -58 38 -28 -57 27 -22 -57 36 -28 -57 32 -22 -50
37 51 07 -01 “■■■■35"" 02 " 05 "41! -07 -02 46 -01 07 45
38 52 11 -05 39 -06 20 46 -01 -14 40 -09 00 46 -
39 57 17 -08 41 -06 03 54 15 -04 47 02 -07 50
40 49 14 -13 22 10 -24 34 09 -09 34 IS -06 38
41 55 22 -07 41 09 18 55 09 -08 50 04 -01 46
42 53 10 -13 35 04 -11 45 11 -07 30 13 -13 47
43 48 00 18 18 01 35 41 -12 10 36 -10 12 29 -
44 47 21 10 35 02 03 .39 08 11 36 04 08 34
45 55 11 -02 39 -09 -02 44 -04 05 40 -10 -04 41 -
46 47 18 20 21 08 23 40 03 20 37 -03 20 36 -
47 52 24 02 29 -11 09 45 06 06 40 04 04 52
48 47 18 -01 30 08 -05 38 11 02 32 12 07 35
N o te .- - D e c im a l p o i n t s o m i t t e d .
79
TABLE 5
R o ta te d C o o r d in a te M a tr ic e s
Group
2 3 4 5 AVE
3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3
-07 06 -42 10 -07 -57 02 01 -54 02 -02 -51 -04 -03 -65 00
-05 10 -27 03 11 -44 07 13 -45 02 07 -35 09 12 -50 03
-29 09 -32 -20 15 -48 -26 26 -46 -14 16 -40 -17 18 -56 -21
-07 -05 -27 -21 -11 -41 -08 -06 -31 -09 -16 -31 -05 -10 -48 -08
-25 -22 -26 00 -15 -31 -23 -19 -23 -24 -17 -35 -18 -18 -40 -27
-12 -13 -43 -14 -06 -63 -06 -23 -44 -10 -07 -SO -12 -19 -62 -12
-15 19 -23 -22 21 -34 -13 23 -32 -16 11 -36 -09 23 -41 -14
-18 10 -38 01 14 -47 -10 20 -37 04 10 -44 -02 13 -55 -06
-16 11 -36 06 11 -46 -06 16 -40 05 04 -45 -10 14 -53 -06
-04 -05 -44 04 -07 -56 07 02 -49 15 -15 -47 09 -07 -63 07
15 -05 -21 17 -10 -42 16 -05 -42 17 -12 -40 10 -08 -49 17
-12 13 -39 -12 05 -54 -07 22 -42 -04 12 -43 -11 09 -60 -07
' 49" - 06 06 06 -14 -03 45 -10 -03 29 -19 01 37 -11 00 42
60 06 04 35 14 10 52 17 05 47 18 04 46 21 07 55
48 -02 -03 19 -01 12 44 14 -14 31 -03 01 33 07 00 42
18 24 -05 16 15 15 07 12 04 -31 15 11 -24 20 15 -01
17 24 15 01 33 03 25 27 02 26 28 14 12 35 06 20
25 -07 -26 22 04 -23 17 -01 -16 31 -14 -11 27 -02 -23 26
16 13 30 -06 -01 32 -01 05 22 10 04 30 18 11 32 09
63 01 11 39 14 14 42 05 10 43 07 12 42 11 12 55
47 18 31 11 21 23 34 22 12 27 20 18 37 25 22 39
30 23 -03 03 23 01 23 22 -01 16 29 04 17 31 03 26
39 05 -03 -11 -21 -05 03 -21 -09 -01 -19 -OS -24 -20 -01 00
47 09 -08 26 08 -01 35 -09 19 25 -13 17 18 04 16 33
- 3 i -49 44 -13 -50 42 -35 -47 44 -26 -55 40 -05 -62 50 -34
-35 -58 31 -08 -56 33 -36 -55 41 -25 -55 31 -19 -69 44 -34
-23 -48 34 -25 -58 41 -23 -61 33 -23 -54 34 -10 -70 44 -25
-05 -36 -03 04 -52 18 -09 -44 33 -11 -10 38 -03 -52 31 -03
-36 -32 21 -11 -54 41 -08 -53 35 -04 -21 01 -21 -52 29 -27
-21 -28 31 19 -57 31 -10 -46 34 -33 -54 25 -09 -64 39 -10
-29 -54 27 -12 -58 27 -25 -54 29 - 29 -49 27 -23 -66 37 -34
-32 -40 18 -34 -45 46 -27 -43 44 -16 -48 42 -15 -55 51 -30
-20 -31 33 -32 -53 44 -16 -56 35 -16 -53 17 -17 -66 44 -21
-30 -45 43 -22 -51 43 -24 -52 42 -17 -47 37 -28 -65 48 -27
-32 -52 31 -29 -53 35 -30 -54 33 -29 -49 36 -22 -66 42 -34
-28 -57 27 -22 -57 36 -28 -57 32 -22 -50 28 -33 -70 40 -30
-or ' 38 02 05 48 -07 -02 46 -01 07 45 02 10 54 -02 04
-05 39 -06 20 46 -01 -14 40 -09 00 46 -08 08 54 -04 -03
-08 41 -06 03 54 IS -04 47 02 -07 50 02 -02 60 08 00
-13 22 10 -24 34 09 -09 34 15 -06 38 13 -17 43 09 -15
-07 41 09 18 55 09 -08 50 04 -01 46 10 05 60 13 02
-13 35 04 -11 45 11 -07 30 13 -13 47 16 -05 51 11 -11
i 18 18 01 35 41 -12 10 36 -10 12 29 -13 27 44 -12 19
10 35 02 03 .39 08 11 36 04 08 34 05 11 45 07 09
-02 39 -09 -02 44 -04 05 40 -10 -04 41 -02 -04 52 -03 00
1 20 21 08 23 40 03 20 37 -03 20 36 -01 02 45 04 24
02 29 -11 09 45 06 06 40 04 04 52 07 -07 52 10 06
1 -01 30 08 -05 38 11 02 32 12 07 35 12 -05 44 13 02
.n ts o m i t t e d .
80
expressions within the three-dimensional space.
Dimensions. One consistent finding is that the
three dimensions are highly similar across all six matri
ces. The first dimension is strongly bipolar, with the
set of Happy expressions at the negative end, and the Sad
at the positive end. All dimensions in a multidimensional
scaling analysis are necessarily bipolar, so there is no
particular significance that there is strong bipolarity.
The Angry expressions are fairly uniformly distributed
around zero, while the Fear-Anxiety tend to be slightly
positive. No Angry expression is so negative or positive
that it extends into the range of the Happy or Sad pro
jections; and in only a couple of cases does a Fear-
Anxiety expression project as highly as a Sad expression.
Dimension 2 is also strongly bipolar, with Angry
at the negative end and Happy at the positive end. The
general tendency is for the Sad and Fear-Anxiety expres
sions to be distributed around zero, with more positives
than negatives. In no case are any of these projections
at all as high as the Angry and Happy.
The third dimension is marked by the Fear-Anxiety
at the positive end, and the Happy at the negative end.
However, the projections of these expressions are not as
large as the projections on dimensions 1 and 2, especially
the negative loadings of the Happy expressions. The set
79
TABU 5
R o tate d C o o rd in ate ‘* u t n c c *
wfOU,)
2 3
4 5 AVE
L 2 3 1 2 3 1
* »
3 1 2 3 1 2 3
36 -42 10 -07 -57 °2 '<1 - 54 02 -02 -51 -04 -03 -65 00
10 -27 03 11 -44 07 1 3 -45 02 07 -35 09 12 -50 03
39 -32 -20 15 -48 ■It • ? • - 40 -14 16 -40 -17 18 -56 -21
35 -27 -21 -11 -41 -OS - 90 -31 -09 -16 -31 -OS -10 -48 -08
22 -26 00 -IS -31 -23 - 19 -23 -24 -17 -35 -18 -18 -40 -27
L3 -43 -14 -06 -63 -00 -23 -44 -10 -07 -50 -12 -19 -62 -12
L9 -23 -22 21 -34 -13 - 32 -16 11 -36 -09 23 -41 -14
L0 -38 01 14 -47 -10 2" - 37 04 10 -44 -02 13 -55 -06
LI -36 06 11 -46 - Oo If: - 40 05 04 -45 -10 14 -53 -06
35 -44 04 -07 -56 0 7
,
* 4 * • 15 - IS -47 09 -07 -63 07
35 -21 17 -10 -42 10 * ’ 3 * - 42 17 -12 -40 10 -08 -49 17
L3 -39 -12 05 -54 -07
*
- 42 -04 12 -43 -11 09 -60 -07
)6 06 06 -14 -0T" 45 -l.» - 29 -19 01 37 -11 00 42
36 04 35 14 10 32 1- 05 47 18 04 46 21 07 55
32 -03 19 -01 12 44 14 - 14 31 -03 01 33 07 00 42
!4 -OS 16 15 IS 0 7 12 94 -31 15 11 -24 20 15 -01
!4 15 01 33 03 25
- 7
02 26 28 14 12 35 06 20
17 -26 22 04 -23 17 -91 - 16 31 -14 -11 27 -02 -23 26
.3 30 -06 -01 32 -01 u 5 22 10 04 30 18 11 32 09
11 11 39 14 14 42 0 5 10 43 07 12 42 11 12 55
.8 31 11 21 23 34
-1
12 27 20 18 37 25 22 39
!3 -03 03 23 01 23 -01 16 29 04 17 31 03 26
15 -03 -11 -21 -05 03 -21 -09 -01 -19 -05 -24 -20 -01 00
19 -08 26 08 -01 35 - 0 9 19 25 -13 17 18 04 16 33
-9 44 -13 -SO 41 -35 11 -26 -S3 40 -05 -62 50 -34
18 31 -08 -56 33 - 3o - 5 5 41 -25 -55 31 -19 -69 44 -34
8 34 -25 -58 41 -23 -61 33 -25 -54 34 -10 -70 44 -25
6 -03 04 -52 18 -09 -44 33 -11 -10 38 -03 -52 31 -03
2 21 -11 -54 41 -08 - 53 35 -04 -21 01 -21 -52 29 -27
8 31 19 -57 31 -10 -46 34 -33 -54 25 -09 -64 39 -10
4 27 -12 -58 27 -25 -54 29 -29 -49 27 -23 -66 37 -34
0 18 -34 -45 46 -27 -43 44 -16 -48 42 -15 -55 51 -30
1 33 -32 -53 44 -It- - 56 35 -16 -53 17 -17 -66 44 -21
5 43 -22 -51 43 -24 -52 42 -17 -47 37 -28 -65 48 -27
2 31 -29 -53 35 -30 - S 4 33 -29 -49 36 -22 - 66 42 -34
7 27 -22 -57 36 -28 " J t 32 -22 -SO 28 -33 -70 40 -30
8 0 2 " ITS'- ' 48 -TT7 - - '0 ! — 40 - u l 07 45 02 10 54 -02 04
9 -06 20 46 -01 -14 40 -09 00 46 -08 08 54 -04 -03
1 -06 03 54 15 -04 47 0 2 -07 50 02 -02 60 08 00
2 10 -24 34 09 -09 34 15 -06 38 13 -17 43 09 -15
1 09 18 55 09 -OS SO 04 -01 46 10 05 60 13 02
5 04 -11 45 11 -07 30 13 -13 47 16 -05 51 11 -11
8 01 35 41 -12 10 30 -10 12 29 -13 27 44 -12 19
5 02 03 39 08 11 36 04 08 34 05 11 45 07 09
9 -09 -02 44 -04 05 40 - 10 -04 41 -02 -04 52 -03 00
1 08 23 40 03 20 37 -03 20 36 -01 02 45 04 24
9 -11 09 45 06 06 40 04 04 52 07 -07 52 10 06
0 08 -05 38 11 02 32 12 07 35 12 -05 44 13 02
80
expressions within the three-dimensional space.
Dimensions. One consistent finding is that the
three dimensions are highly similar across all six matri
ces. The first dimension is strongly bipolar, with the
set of Happy expressions at the negative end, and the Sad
at the positive end. All dimensions in a multidimensional
scaling analysis are necessarily bipolar, so there is no
particular significance that there is strong bipolarity.
The Angry expressions are fairly uniformly distributed
around zero, while the Fear-Anxiety tend to be slightly
positive. No Angry expression is so negative or positive
that it extends into the range of the Happy or Sad pro
jections; and in only a couple of cases does a Fear-
Anxiety expression project as highly as a Sad expression.
Dimension 2 is also strongly bipolar, with Angry
at the negative end and Happy at the positive end. The
general tendency is for the Sad and Fear-Anxiety expres
sions to be distributed around zero, with more positives
than negatives. In no case are any of these projections
at all as high as the Angry and Happy.
The third dimension is marked by the Fear-Anxiety
at the positive end, and the Happy at the negative end.
However, the projections of these expressions are not as
large as the projections on dimensions 1 and 2, especially
the negative loadings of the Happy expressions. The set
81
of Angry expressions tend to have negative projections,
while the Sad are distributed around zero; and in only a
few cases are these projections as large as those for
Fear-Anxiety and Happy.
One of the significant problems to which this
research was directed was whether or not the dimensions
obtained in the multidimensional scaling analysis could be
interpreted as meaningful psychological constructs. A
very distinct possibility recognized at the outset was
that the dimensions would not be meaningful, but rather
that they would be artifacts of the multidimensional
scaling procedure. Such a situation would arise when the
stimuli that are being scaled are categorical in nature,
and they form relatively homogeneous clusters which tend
to be equally distant from each other. It is a situation
analogous to fitting three points in two dimensions, or
n points in n-1 dimensions. Despite these preconceived
expectancies, an attempt was made to interpret the dimen
sions as meaningful psychological constructs.
Of the three dimensions, the one for which a label
could most readily be attached was the third. With the
Fear-Anxiety expressions at one end, and the Happy at the
other end, it seemed reasonable that this dimension might
be called tension. The Happy expressions, especially the
stick figures, appear to be quite free of any tension,
82
while most of the 12 expressions in the Fear-Anxiety set
represent a considerable degree of tension. Furthermore,
the differences among the projections of Angry and Sad
expressions seemed to be accounted for by varying degrees
of tension.
Attempts to label the first dimension did not lead
to a convenient descriptive term. To call it happy-sad
did not explain very well the spread of the Angry and
Fear-Anxiety expressions on the first dimension. A some
what reasonable description that took into account vari
ation along the entire dimension was in terms of something
like inhibited-disinhibited. Alternative descriptions
might be inner-outer directed, or withdrawn-outgoing.
This concept not only accounts for the marked contrast
between the Sad (inhibited) and Happy (disinhibited)
expressions, but is also reasonable for the contrasts
among the remaining expressions.
The second dimension was least readily labeled.
A possibility that reasonably describes the difference
between Angry and Happy is social-antisocial. This con
ceptualization, however, does not seem to accurately
describe the contrast among those Fear-Anxiety and Sad
expressions that have deviant positive and negative pro
jections .
In Schlosberg’s model of emotions and in the
multidimensional scaling and semantic differential studies
83
of facial expressions, the first dimension has been
labeled Pleasant-Unpleasant, but it does not seem to be
an applicable label for either of the three dimensions
obtained here. While the Happy expressions are always at
one end of a dimension, in no case are all of the other
expressions at the other end. It is possible that a rota
tion of the first two dimensions could result in an align
ment that places the expressions on a dimension such that
the non-Happy expressions are all at one end. A term
that has been suggested for a second dimension in the
above-mentioned studies is Arousal, but again it does not
appear that this fits any of the three dimensions in this
study.
While it was possible to describe the three dimen
sions with psychological constructs, the labels seem some
what ad hoc and do not correspond to previously suggested
dimensions of facial expressions. What seems more accept
able is to consider the coordinate matrices as giving pro
jections on an arbitrary set of dimensions, and to place
more importance on the clusters obtained and the shape of
the configuration.
Mood clusters. The most overwhelming characteris
tic of the coordinate matrices is that the set of 12
expressions for each mood quality are located very close
to each other in the three-dimensional space, and do not
84
overlap the other sets of expressions. For this reason it
is conceptually easier to think of the coordinate matrices
as defining four mood clusters in a three-dimensional
space, rather than defining three dimensions of mood. In
describing the mood clusters, each mood quality will be
treated separately and in order of the numbering of Figure
1. Within each cluster, the individual expressions will
be characterized according to their location within the
spatial configuration of the cluster.
Despite the emphasis on the clusters, it was found
that considering the differences of projections on the
dimensions led to reasonable and meaningful descriptions
of the expressions over and above a simple categorical
statement that an expression is Angry, or Happy, or what
ever cluster an expression ;vas located in. It also turned
out that each mood cluster had its primary projections on
just one dimension, and that the larger and smaller pro
jections could be interpreted in terms of varying degrees
of the mood represented by the expressions in the cluster.
These dimensions for each of the mood qualities are:
Angry - 2, Fear-Anxiety - 3, and Happy and Sad - 1. But
when the projections on the secondary dimensions were con
sidered, it was not meaningful to interpret differences
here in terms of the mood quality represented by the
expressions at the end of the dimension. As an alternative
85
approach, it was meaningful in most cases to use the
descriptive labels suggested above for the three dimen
sions in making comparisons among the expressions on the
basis of their projections on the secondary dimensions.
Thus, for Angry and Fear-Anxiety, their projections on
dimension 1 are described in terms of inhibited-
disinhibited, or the other labels suggested for this
dimension. The projections of Angry, Happy and Sad on
dimension 3 are described in terms of tension, but it was
not reasonable to describe the projections of Tension and
Sad on dimension 2 in terms of social-antisocial, as it is
not immediately obvious what appropriately describes the
ground in between social and antisocial. Thus, although
the dimensions of the coordinate matrices are rejected as
mood constructs, the suggested labels are useful in
describing secondary characteristics of the expressions.
As with the dimensions, there is a considerable
degree of consistency across the groups in the mood clus
ters. However, there are some meaningful differences,
and they are manifested in the particular location of
individual mood expressions in the three-dimensional
space. That is, while the clusters are pretty much the
same, there are some expressions that have somewhat devi
ant projections on one or more dimensions, thus moving
their position within the cluster. Whenever there are
86
substantial differences among the groups in the descrip
tion of a particular expression, they will be pointed out.
In trying to sort out the similarities and differences
among the groups, it was necessary to take into account
that the size of projections was not always a valid indi
cator of differences. Rather, rank ordering of the
expressions on the three dimensions, as well as the size
of the projections, was relied upon. Small deviations
were not given much attention, because the Stress values
for each set of data are not so close to zero as to permit
generalizations based upon small differences.
Angry. Expressions 1, 6, 10, and 12 consistently
had the highest loadings on dimension 2, and can be inter
preted as being very angry. Numbers 3, 8, and 9 were
generally next highest and express a more moderate amount
of anger; and 2, 4, 5, 7, and 11 were generally lowest
and represent mild anger. There is no tendency for a par
ticular mode to be associated with degree of anger, and
there were very few deviations among the groups in this
ordering. One notable difference is that in group 1 the
stick figures 4 and 5 are more angry while 6 is much less
angry.
On dimension 1, all the stick figures, photos 10
and 11, and the adjective angry are usually negative, and
these expressions are more disinhibited and active than
87
the other expressions, which are usually positive and
more inhibited or passive in their expressiveness. There
is not much variation in the projections on dimension 3,
most of them being small positive and negative. Since the
expressions at the furthest negative end are taken as no
tension, the majority of the Angry fall in a range that
can be characterized as slight tension. Numbers 3 and 5
are frequently more negative, approaching a range charac
terized as no tension; while 11*s positive loadings indi
cate almost mild tension. These deviations are quite rea
sonable, although it is not so obvious that grouchy should
represent a mood free of tension.
Fear-Anxiety. Of all four clusters, this one was
the least homogeneous and least stable across the groups.
Only 7 of the 12 expressions were consistently high enough
on dimension 3 to be characterized as always expressing
mild anxiety or a more intense mood of fear. There was
considerable variability among the remaining 5 in their
location on dimension 3. There were two expressions, 14
and 20, that almost always were highest, and these are
fairly obviously fear. Next highest were 13, 15, and 21,
and they appear to represent anxiety, while 17 and 22 were
generally low and seem to express mild anxiety or tension.
Of the others, 16 was mild anxiety for groups 1 and 2,
just slight tension for groups AVE and 3, and no tension
88
for groups 4 and 5; 18 was anxiety for groups 2, 4, and 5
and mild for the others; 19 ranged from slight tension for
AVE, 2, 3, and 4 to some tension for 1 and mild anxiety
for 5; 23 had even a greater range, from no tension in 5
to slight tension in AVE, 2, 3, and 4 to anxiety in 1; and
24 was anxiety for all groups except 4 and 5, where it was
mild anxiety.
The differences among the groups are substantial.
While group 1 seemed to perceive all the expressions
except 17 and 19 as fear or anxiety, group 2 seemed to
perceive only 14, 18, 20, and 24 as fear or anxiety.
While negative projections on dimension 3 are rare, in
group 5 there are two expressions that are so negative as
to be described as representing no tension. There is also
a tendency for mode of expression to be associated with
perception: the stick figures and photos tend at best to
express mild anxiety, while the adjectives and faces are
anxiety or fear. It would probably be erroneous to con
clude that, in general, these relationships are true;
rather, it is more likely that the particular expressions
included in this study are not good communicators of fear
and anxiety.
The projections on dimension 1 are generally posi
tive, indicating that expressions of fear and anxiety tend
to be inhibited or passive. In some cases the projections
89
are quite close to the projections of the Sad expressions,
notably 14, 16, 17, 21, and 22. Of these, only 17 is
located close to the cluster of Sad expressions, and there
are features of this stick figure that would lead to its
being perceived as sad. There are three expressions--13,
18, 23--that tend to be negative, especially 23, and they
are more disinhibited or active in their expressiveness.
The deviancy of 23 on both dimension 1 and 3 suggests that
it is expressing a mood not represented by any of the
other expressions, and it is possible that it is atten
tiveness. The variation of the projections on dimension
2 is not great and is difficult to interpret. Stick fig
ure 18 is consistently the most negative, and the arched
back and clenched fists could quite easily be perceived
as expressing anger. The positive projections of 19, 21,
and sometimes 24 are not easily interpreted as represent
ing a social expressiveness, as was suggested for the
Happy expressions that are at the positive end of this
dimension.
Happy. This set of expressions, like those for
Angry, forms a very tight cluster and is set off by itself
as is indicated by its sizable loadings on all three
dimensions. Although there tended to be very little vari
ation in their projections on the first dimension, it was
possible to break them up into three categories as has
90
been done for the previous moods. Adjectives 26 and 27
and photo 36 are generally perceived as the most happy
(although in group 1 they were moderately happy and 33 and
34 were most happy). Generally perceived as moderately
happy were 34 and 35, while 25, 28, 29, and 32 were mildly
happy. For three of the expressions there was distinct
variability among the groups: 30 was either mild, as in
groups AVE, 2, and 4, or very happy as in the remaining
groups; 31 ranged from just moderate in groups AVE, 1, 4,
and 5 to very in 2 and 3; and 33 had greater variability,
being mild in group 2, moderate in AVE, 3, and 5, and very
happy in groups 1 and 4.
On dimension 2 the expressions are almost always
positive and there is not much variation among the projec
tions withn a group. However, in groups 2 and 3, 28 is
negative and small positive, respectively, suggesting that
these groups might have perceived some boastfulness in
this stick figure, thus accounting for its deviation from
the main cluster on this dimension. The extreme negative
projections on dimension 3 were taken to represent no ten
sion, and most of the expressions can be so characterized.
However, the three stick figures cannot be so described in
all groups. Number 28 is always around zero, and thus
slightly tense; 29 is also slightly tense in groups 2, 3,
and 4; and 30 is slightly tense in groups AVE, 3, and 5
and is anxiety in group 2.
91
Sad. Like those for Angry and Happy, the expres
sions for the mood Sad are neatly clustered together, and
there are very few differences among the groups. A trend
exhibited in the other moods for an adjective to be con
sistently perceived as one of the most intense expressions
of the mood is present here. The adjective sad and stick
figure 41 are almost always the two highest expressions
on dimension 1. The complete dejection in 41 obviously
accounts for this finding. There are five expressions
that are moderately sad: 37, 38, 42, 45, and 47, and an
obvious common feature of the non-adjectives is the head
and eyes cast downward. Least sad are 40, 43, 44, 46,
and 48, which do not have the features common to the other
expressions.
On dimension 2 there is very little variability,
all of the projections being small positive or negative.
And with regard to dimension 3, all but two of the expres
sions could be described as slightly tense. The excep
tions are 43 and 46. The former is anxiety in group 5 and
fear in group 2, while the latter is mild anxiety in AVE,
1, 3, and 4 and anxiety in 2.
Overall, then, the expressions that were selected
for each mood quality do cluster together, and it is only
in the set for Fear-Anxiety that some of the expressions
depart from the cluster. Each mood quality has its
92
highest projections on just one dimension, except Happy,
which has large loadings on dimensions 1 and 2 and smaller
loadings on dimension 3. And, as indicated earlier, there
are few substantial differences among the groups. The
most notable characteristic distinguishing the six coor
dinate matrices is the varying size of the projections,
and these variations are mainly differential stretching
or shrinking of the dimensions.
Dimension matching. In order to summarize and
quantify the similarities and differences among the coor
dinate matrices, two types of correlation analyses were
performed. The first set of correlations is presented in
Table 6, and they are the correlations of each dimension
TABLE 6
Correlations between AVE Dimensions and
Corresponding Group Dimensions
Dimension
Group
1 2 3 4 5
1 .98 .97 .99 .99 .97
2 .97 .94 .99 .98 .98
3 .95
00
r-
•
.97 .92 .90
(4)
.88
00
o
•
.53 .46 .29
93
in the five groups with the corresponding dimension in
AVE. For three dimensions, all but three of these corre
lations are .95 or greater, thus indicating the substan
tial similarity. The lowest correlation is between the
third dimension of group 2 and AVE, and this departure
from great similarity was mentioned in describing the
Fear-Anxiety cluster. Also included in the table are the
correlations between corresponding fourth dimensions, and
these are uniformly low, except for group 1.
The second type of analysis was a multiple regres
sion prediction of each dimension in AVE from the three
dimensions in each group. The regression weights for
corresponding dimensions are presented in Table 7, and
TABLE 7
Regression Coefficients for Dimensions in Groups
when Predicting Corresponding Dimensions in AVE
Dimension
Group
1 2 3 4 5
1 .97 .67 .86 .77 .79
2 1.10 .63
00
00
•
.77 .77
3 1.13 .56 .91 .78 .76
94
they are indicative of the stretching or shrinking of the
group dimensions as compared to AVE. In only group 1 does
any stretching occur, while the most shrinkage occurs in
group 2. The total effect of this stretching and shrink
ing was only to change the relative distances among the
four mood clusters, leaving the general shapes of the con
figurations unaffected.
Configuration shape. There is one further means
of illustrating the similarities and differences among
the groups, and this involves schematically locating clus
ters as four points in the three-dimensional space. To
accomplish this, distances between mood clusters were
estimated by first determining the median projection of
each of the four sets of 12 expressions on the three
dimensions, and then calculating the six euclidean dis
tances between the four points taken to represent the four
moods. A set of distances was obtained for each group,
and these were used to construct tetrahedrons which are
schematically reproduced in Figure 2. In these diagrams
the three vertices of the triangle are Angry, at the top,
Fear-Anxiety, lower left, and Sad, lower right. The loca
tion of the fourth vertex, Happy, is always above the
triangular base, and its projection on the base is indi
cated by the additional point. The distances between
Happy and the three points of the base are always greater
4
5
AVE
Fig. 2. Schematic representation of the three-
dimensional configuration of mood clusters: Anger - top
of triangle; Fear-Anxiety - lower left corner of triangle;
Sadness - lower right corner of triangle; Happiness -
point.
96
than the distances between Angry, Sad, and Fear-Anxiety.
The distance between Sad and Fear-Anxiety is always the
shortest, while Angry is generally equidistant from these
latter two. The most notable difference among these con
figurations is that Happy projects within the triangular
base for groups 1 and 3, but outside for the other groups.
Also, in groups 2 and 4, Happy is somewhat closer to Fear-
Anxiety and further from Angry than in the other groups.
To summarize, there is much similarity in the con
figuration of the clusters, the major difference being in
the overall size. While there are distinct clusters, they
are not all equidistant from each other, which would
result if ratings of similarity were based entirely on
judging whether or not pairs of expressions were the same
mood. Although Angry, Sad, and Fear-Anxiety are distinct
clusters, the distance between them is less than their
distances from Happy. The relatively short distance
between Sad and Fear-Anxiety would indicate that there are
some similarities in the expressions of these moods, at
least those included in this study.
Multidimensional scaling within mode. There was
major interest in determining if mode of expression had
any effect on perception of mood similarity, or if it was
associated with the spatial configurations. Looking at
the coordinate matrices, there appear to be only two
97
trends related to mode of expression. One is that, for
all four moods, one or more of the adjectives has the
largest projection on its primary dimension. This could
be taken to mean that verbal expression is the least
ambiguous indicator of mood. Secondly, the stick figures
tend to have the smaller projections, and also to deviate
more from the cluster. From this it would be possible to
conclude that postural cues are the most ambiguous indi
cators of mood, or that there is insufficient information
in stick figures to communicate specific mood states.
To assess further some of the characteristics of
mode of expression, separate multidimensional scaling
analyses were performed by extracting from the 48 x 48
matrix of similarities four 12 x 12 matrices. These
matrices were the ratings among the 12 expressions of each
mode type. Of primary interest in these analyses were the
dimensionalities for the various modes. Solutions were
obtained at 2, 3, and 4 dimensions, and the Stress values
are presented in Table 8 and the three-dimensional coor
dinate matrices in Table 9. Although there is much use
ful information in these tables, it must be cautioned that
the data upon which they are based were not obtained under
separate experimental conditions. There is no telling
what effect the presence of all modes in the booklets had
on the ratings of same mode pairs.
98
TABLE 8
Stress Values: Within Mode
Group
Dimensionality
1 2 3 4 5 AVE
Adj ectives
2 .068 .118
— » — « —
.052
3 .016 .069 .036 .031 .037 .015
4 .017 .048 .029 .015 .033 .015
Stick Figures
2 .050 .078
_ —
.064
3 .034 .120 .053 .095 .093 .042
4 .015 .075 .040 .054 .059 .032
Faces
2 .075 .135 .078
~ — •» —
.047
3 .029 .069 .048 .039 .028 .028
4 .026 .046 .026 .021 .026 .015
Photos
2 .043 .083 .040 .067 .064 .027
3 .025 .057 .033 .040 .043 .021
4 .015 .046 .022 .029 .035 .019
Note.--No solution obtained where -- appear.
99
TABLE 9
H o ta te d C o o r d in a te M a t r i c e s : W ithin Mode
Group
S tim u lu s 1 2 3 4
1 2 3 1 2 3 1 2 3 1 2 3 1
1 -11 -66 -14 47 -28 22 -08 -49 00 -03 -56 00 28
2 -07 -61 -13 12 -14 34 10 -52 -08 08 -45 16 -02
3 -03 -63 -09 30 -37 -12 21 -41 06 22 -33 11 32
13 -08 12 54 01 -04 34 -13 -03 44 -06 -13 36 -11
14 01 02 61 10 35 00 04 -11 57 24 17 42 07
.15 -09 07 53 16 12 22 04 09 48 08 -20 35 02
25 -51 50 -35 -88 -02 12 -56 61 -45 -59 56 -55 -78
26 -51 46 -37 -76 -02 30 -53 59 -52 -65 60 -47 -85
27 -51 49 -27 -90 -03 -07 -63 59 -40 -74 48 -50 -78
37 62 04 -09 48 -04 -28 47 -17 03 52 -06 04 60
38 63 10 -12 32 22 -42 50 -04 -14 43 -15 -05 60
39 64 09 -12 57 25 04 57 -01 00 51 08 12 65
4 35 -62 -16 -01 -29 -13 27 -53 -01 30 -48 06 -15
5 31 -63 -09 -28 -08 03 -10 -46 -08 -08 -41 00 -12
6 25 -50 -42 -16 -41 01 10 -55 46 -06 -24 -59 00
16 14 20 49 03 -06 26 04 44 07 19 10 -38 15
17 22 50 04 25 00 06 24 26 18 20 01 55 21
18 27 -05 -02 14 -11 -19 45 -10 02 02 28 15 OS
28 -74 -19 -03 -32 07 -18 -66 -19 -41 -63 -19 07 -22
29 -68 -06 -32 -22 22 -37 -54 08 -67 -69 18 -17 -43
30 -75 -11 -14 01 42 -14 -72 22 -34 -58 19 19 -69
40 19 52 13 33 14 02 07 21 41 46 -03 22 40
41 23 49 25 02 09 39 43 39 33 45 38 -19 34
42 22 44 27 20 01 25 44 23 05 42 21 08 47
7 14 -57 09 17 -53 -31 27 -45 - 23 35 ' - 4 0“ -32 02
8 05 -66 -04 19 -52 29 36 -56 -03 18 -55 05 09
9 03 -68 -11 -03 -47 06 11 -62 01 17 -53 07 08
19 15 31 -02 04 31 06 -26 14 -03 00 27 01 -07
20 02 03 68 -07 07 61 14 12 62 00 01 63 00
21 18 23 47 31 14 -01 23 40 16 17 -07 42 19
31 -69 29 -32 -71 29 -48 -77 26 -25 -80 46 -39 -56
32 -73 28 -26 -6 6 30 -14 -79 23 00 -59 61 -39 -67
33 -66 34 -27 -46 54 -43 -69 50 -20 -79 51 -21 -62
43 53 03 -14 53 -10 -18 52 -02 -24 34 -15 13 55
44 44 24 03 21 -05 38 36 09 09 46 10 -07 45
45 53 16 06 49 03 15 52 -10 11 49 -25 06 55
. 10 -19 -75 01 -13 - 55 r r -16 -61 22 01 -55 -11 00
11 -21 -61 22 27 -34 -11 04 -60 -11 -04 -67 02 -02
12 -17 -75 -07 18 -48 08 03 -72 10 22 -60 24 20
22 32 13 40 36 -16 22 27 -04 09 22 -09 27 44
23 -08 32 06 -01 -01 05 -07 04 -19 -08 16 28 -19
24 06 07 51 44 -04 02 14 -01 34 -04 05 -26 -03
34 -50 35 -47 -69 50 -43 -53 65 -19 -66 45 -31 -70
35 -49 39 -53 -64 57 -49 -66 59 -32 -65 63 -19 -70
36 -50 40 -49 -68 60 -40 -64 55 -50 -72 64 04 -69
46 53 02 20 49 01 09 49 -13 29 57 -07 -22 45
47 61 18 08 20 -11 41 61 05 06 61 -09 17 64
48 61 24 08 21 01 45 49 23 22 56 16 08 59
N o t e .- - D e c i m a l p o i n t s o m it t e d .
i
I
■ 1
i
99
i
TABU; 9
lio ta te d C o o r d in a te M atri o p s : W ithin Mode
Group
2 3 4 5 AVE
3 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3
■14 47 -28 22 -08 -49 00 -03 -56 00 28 -59 -16 -05 -64 -04
•13 12 -14 34 10 -52 -08 08 -45 16 -02 -47 08 04 -58 -09
•09 30 -37 -12 21 -41 06 22 -33 11 32 -44 25 11 -55 -05
54 01 -04 34 -13 -03 44 -06 -13 36 -11 07 -26 -12 05 38
61 10 35 00 04 -11 57 24 17 42 07 03 48 20 -02 57
53 16 12 22 04 09 48 08 -20 35 02 19 -18 -OS 02 39
•35 -88 -02 12 -56 61 -45 -59 56 -55 -78 47 -13 -63 60 -35
•37 -76 -02 30 -53 59 -52 -65 60 -47 -85 35 -09 -64 60 -35
•27 -90 -03 -07 -63 59 -40 -74 48 -50 -78 36 -29 - 66 60 -25
•09 48 -04 -28 47 -17 03 52 -06 04 60 -01 -04 56 -07 -06
•12 32 22 -42 50 -04 -14 43 -15 -05 60 -07 17 59 -02 -11
■12 57 25 04 57 -01 00 51 08 12 65 11 17 65 00 -03
■16 -01 -29 -13 27 ---S T- - o r 30 -48 06 -15 -40 28 36 -50 -24
•09 -28 -08 03 -10 -46 -08 -08 -41 00 -12 -51 11 23 -55 -18
■42 -16 -41 01 10 -55 46 -06 -24 -59 00 -65 -26 24 -40 -63
49 03 -06 26 04 44 07 19 10 -38 15 -11 -50 19 11 45
04 25 00 06 24 26 18 20 01 55 21 45 17 17 52 02
•02 14 -11 -19 45 -10 02 02 28 15 05 -02 54 34 -03 05
■03 -32 07 -18 -66 -19 -41 -63 -19 07 -22 49 -07 -68 -35 01
•32 -22 22 -37 -54 08 -67 -69 18 -17 -43 -02 20 -79 -15 01
■14 01 42 -14 -72 22 -34 -58 19 19 -69 12 -24 -80 -06 -10
13 33 14 02 07 21 41 46 -03 22 40 13 -25 21 50 12
25 02 09 39 43 39 33 45 38 -19 34 37 08 27 52 24
27 20 01 25 44 23 05 42 21 08 47 IS -06 27 41 24
09 17 - 53 "rn " - 27 ' -45 -23 35 --r40 -32 02 -57 07 33 -39 -34
•04 19 -52 29 36 -56 -03 18 -55 05 09 -53 -05 16 -66 -05
•11 -03 -47 06 11 -62 01 17 -53 07 08 -59 -06 12 -62 -18
■02 04 31 06 -26 14 -03 00 27 01 -07 22 -08 -01 27 00
68 -07 07 61 14 12 62 00 01 63 00 -08 59 11 02 59
47 31 14 -01 23 40 16 17 -07 42 19 04 55 25 20 43
■32 -71 29 -48 -77 26 -25 -80 46 -39 -56 62 -45 -77 46 -24
•26 -66 30 -14 -79 23 00 -59 61 -39 -67 59 -34 -69 52 -31
•27 -46 54 -43 -69 50 -20 -79 51 -21 -62 49 -48 -83 36 -20
•14 53 -10 -18 52 -02 -24 34 -15 13 55 -06 -07 31 -22 23
03 21 -05 38 36 09 09 46 10 -07 45 -04 23 48 18 00
06 49 03 15 52 -10 11 49 -25 06 55 -09 07 54 -12 07
01 -13 -53" 11 ' -TS -S I'" 11 01 -SS -11...... 00 -66 18 -05 -67 16
22 27 -34 -11 04 -60 -11 -04 -67 02 -02 -58 48 -08 -66 05
•07 18 -48 08 03 -72 10 22 -60 24 20 -24 62 16 -67 -14
40 36 -16 22 27 -04 09 22 -09 27 44 -06 03 37 -02 -01
06 -01 -01 05 -07 04 -19 -08 16 28 -19 04 19 -01 11 -17
51 44 -04 02 14 -01 34 -04 05 -26 -03 16 -45 11 04 23
•47 -69 50 -43 -S3 65 -19 -66 45 -31 -70 22 -19 -68 54 -18
•53 -64 57 -49 -66 59 -32 -65 63 -19 -70 43 -22 -69 57 -32
•49 -68 60 -40 -64 55 -50 -72 64 04 -69 18 -40 -70 58 -35
20 49 01 09 49 -13 29 57 -07 -22 45 13 06 42 -06 38
08 20 -11 41 61 05 06 61 -09 17 64 08 -21 62 04 14
08 21 01 45 49 23 22 56 16 08 59 31 -09 53 19 20
o m it t e d .
i
100
The Stress values for the adjectives and faces
point very clearly to three dimensions for all groups.
For the stick figures, the Stress values are low at three
dimensions only for group AVE and 1; four or more dimen
sions are indicated for the other groups. The Stress
values for the photos are very interesting, as they sug
gest that the dimensionality might be 2 rather than 3 for
some of the groups. Since previous scalings of facial
expressions have been interpreted in two dimensions, the
coordinate matrices for the 12 photos were examined more
closely than the other modes. In all groups the config
uration is basically the same: Happy, Angry, and Sad
form an isosceles triangle, with the latter two forming a
short base; and Fear-Anxiety is scattered inside the tri
angle, with 22, and sometimes 24, being fairly close to
Sad. This configuration of the mood clusters appears to
resemble the configurations obtained in the multidimen
sional scaling studies of the Lightfoot faces (Abelson §
Sermat, 1962; Cliff § Young, 1968; Shepard, 1962b). The
correspondence depends on the extent to which photographs
labeled as Anger, Grief, and Pleasant Surprise in Shepard
(1962b) are similar to the mood clusters Anger, Sad, and
Happy in this study. While not obtaining a third dimen
sion or a cluster of Fear-Anxiety in the analysis of the
photographs could be taken as an indication that there is
101
an interaction between mode of expression and configura
tion, an equally likely explanation is that the photos
used to represent Fear-Anxiety do not express this mood
well, and that a better selection of faces would yield a
three-dimensional solution and a Fear-Anxiety cluster.
As for the configurations in the analyses of the
other modes, four mood clusters emerge for the adjectives
and line-drawn faces, but the results are not nearly that
clear for the stick figures. For the adjectives and the
faces, deviations from the complete analysis generally
involve the change of a stimulus' location on one dimen
sion. In group 5, the adjectives for Fear-Anxiety do not
cluster together as they did in the complete analysis, but
this is the only group in which a mood cluster does not
clearly emerge. For the stick figures, in only groups 1
and AVE do the Stress values indicate three dimensions
and it is in only these groups that there is a tendency
for mood clusters to appear. The Fear-Anxiety expressions
are scattered on all three dimensions, but the other
moods are present as clusters. In the other groups four
or more dimensions would be needed to account for the
similarity ratings, and in the three-dimensional configu
rations the expressions for the different moods do not
fall into four clusters as neatly as they do in the com
plete analyses. Although there is a tendency for same
102
mood expressions to be near each other, excepting Fear-
Anxiety again, there is much more scattering of the
expressions in the three-dimensional space. The general
weakness of the stick figures in the complete analysis to
be explicit and univocal mood expressions is further
enhanced when the relations among only the stick figures
are analyzed. This suggests that the information carried
by the stick figures is somewhat vague, and becomes more
specific when an additional cue is available, such as an
adjective or a face. It does not seem as likely for the
stick figures as it might for the photographs that they
are just poor expressions of one mood. Rather, it is more
likely that the postural cues in the stick figures do not
express specific moods by themselves. The inconsistency
in the judgments among stick figures, as reflected by the
higher Stress values, is probably a manifestation of this
ambiguity in the stick figures. Both the ambiguity and
inconsistency are reduced, however, when a different mode
of expression is paired with a stick figure.
CHAPTER VI
DISCUSSION
Repeated Measurements
The analyses of repeated measurements played a
minor role in this research, serving mainly as a quality
control to indicate the trustworthiness of the individ
uals' ratings. The two procedures used to detect individ
uals who might have been responding in a random manner did
not lead to the deletion of any outliers. Either these
procedures were not sensitive enough, or the ratings were
not so deviant as to arouse suspicion. The size and range
of the test-retest correlations for individuals in this
sample compares very favorably to similar correlations
over 45 ratings of similarities of personalities that were
reported by Weksel and Ware (1967). There were 43 persons
in their sample, and the range of reliabilities was .34 to
.86, with a median of .71. In this sample of 35 the range
is .54 to .98, with a median of .83.
The failure to detect outliers and the generally
high level of reliabilities is an encouraging sign that
subjects can be consistent when required to make a large
number of similarity judgments. Even greater test-retest
103
104
consistency was obtained when ratings were averaged over
subjects, although this is certainly expected. It is
important, though, to have this level of consistency when
mean ratings for different groups are analyzed separately
if an investigator is going to have confidence in the
results obtained from the separate analyses. This confi
dence is especially necessary when, as was found here,
correlations among the group data indicate that there are
small differences between them.
Individual Differences
The search for clusters of people whose ratings
were very similar to one another and different from rat
ings of persons not in the cluster met with very little
success. From two standpoints it was quite obvious that
the data from the five groups were not indicative of sub
stantial differences among the groups. First of all, the
inter-group correlations in Table 2 are quite high, rang
ing from .80 to .92, and the correlations between the
groups and AVE are even higher, ranging from .89 to .98.
A study by Bradley and Cliff (1968) that included an anal
ysis of individual differences in ratings of similarity
of mood adjectives also found high correlations among the
group data. Although the group correlations were high,
they were lower than the test-retest reliabilities, and
it was felt that the multidimensional scaling might lead
105
to differences between groups. But even here the similar
ity among the configurations of mood expressions made it
difficult to conclude that the configurations represented
different points of view. Although the differences among
the configurations were meaningful, they were minor and
generally involved the shift of a small number of expres
sions within their own mood cluster. There were no dif
ferences that could be attributed to, for example, one
group making judgments differentially within mode of
expression, or a group making judgments of similarity on
the basis of pleasantness or intensity of mood. Cliff
(1965) reported a failure to find substantial individual
differences, even when separate groups of raters were
either instructed or induced to attend to different char
acteristics of facial expressions.
It is possible that the failure to find substan
tive individual differences arose from the use of weak
cluster analysis methods. This does not seem likely with
regard to the Tucker and Messick (1963) procedure, which
is based on factor analysis. The eigenvalues obtained
from the correlation matrices for each booklet always
indicated one very large factor, with the second and suc
ceeding factors being very small. The robustness of the
Ward (1963) procedure is an unknown quantity, but it did
yield clusters that were somewhat invariant over the four
106
booklets, as indicated by the significant chi-square
values obtained in Table A5. The most likely reason for
the failure to obtain substantive individual differences
is that there just weren't any in the sample that provided
the data. Sampling outside introductory psychology clas
ses, outside the student population, or outside the normal
population may be necessary in order to obtain ratings of
mood similarity that reflect different organizations of
mood expressions. That substantive individual differences
are hard to come by has been noted by Cliff (1968). There
has been a failure in not only the mood and personality
domain (Messick and Kogan, 1966; Walters § Jackson, 1966) ,
but also in the attitude domain (e.g., Messick, 1961).
A possible, but yet untried, source of individual
differences in mood organization might be the variable of
complexity of conceptual systems as formulated by Harvey,
Hunt, and Schroder (1961). Conceptual systems vary on
this dimension from concreteness to abstractness, where a
conceptual system is a way of organizing environmental
information. It would be of interest to contrast the con
figuration of mood expressions for concrete individuals to
that for abstract individuals. The predominant similarity
among the configurations for the various groups in this
study suggests that this sample was composed of all con
crete or all abstract individuals. Alternatively, it may
107
be that in this sample individual differences in complex
ity of conceptual systems did not lead to individual dif
ferences in judging similarity of mood expressions.
Multidimensional Scaling
Dimensionality. One of the most important con
cerns of this study was determining the number of dimen
sions that would be required to account for the similarity
ratings between expressions representing four moods.
Although previous multidimensional scaling studies of
facial expression had usually found that two dimensions
were sufficient, it was expected here that three dimen
sions would be necessary. This hypothesis was put forth
because it was expected that the results would be more
interpretable in terms of clusters of mood rather than
dimensions of mood expression. Although three dimensions
were retained for both the group data and the average
data, it is necessary to evaluate the adequacy of these
solutions.
Considering only the Stress values for the group
and overall average data, it is seen that in only two
cases is there strong evidence for three dimensions.
These cases are group 1 and AVE, and the Stress values are
9.8% and 9.7%, respectively. In terms of the qualitative
labels attached to various levels of Stress (Kruskal,
1964a), these solutions would be fair. The Stress values
for the other groups at three dimensions range from 12.31
to 16.2%, which, qualitatively, would be fair to poor.
However, these are only qualitative labels, and do not
indicate how likely or unlikely such Stress values are for
a multidimensional solution of 48 points in three dimen
sions. There is evidence that the Stress values for all
groups are highly unlikely to occur by chance (Stenson §
Knoll, 1969). In their multidimensional scaling analysis
of three different random rankings of distances among 50
points, they obtained an average Stress value of 28% at
three dimensions. Furthermore, on the basis of several
such analyses, they suggest that the distribution around
these so-called null values is very peaked, and that
experimentally obtained Stress values which are 2% differ
ent are not likely to occur by chance. Evaluating the
Stress values obtained here in terms of a null value of
28%, it is seen that even the worst solution is 12% away
from the null value, and that all Stress values are not
likely to occur by chance. Further evidence in support
of the three-dimensional solutions is seen from the dis
tribution of eigenvalues in Table 4. For almost every set
of data, and certainly for AVE, there is a clear break
between the third and fourth roots, with the differences
between the fourth and succeeding roots being much smaller
109
than the difference between the third and fourth root.
Further support for three dimensions is found in a monte-
carlo study by Sherman and Young (1968). They found, for
a given dimensionality, that as the number of points
increased, the extent to which the reproduced distances
correlated with the true distances also increased, even
though the Stress values became poorer. Generalizing this
finding to the situation here, it is seen that the three-
dimensional configurations of the 48 expressions would
predict the true distances quite well, but the obtained
Stress values do not indicate this.
Statistical goodness of fit is only one consider
ation in determining the number of dimensions. The over
whelming evidence in favor of the three-dimensional solu
tions was the psychological meaningfulness of the config
urations in three dimensions. This meaningfulness was not
in terms of the dimensions, however, but in terms of the
clusters of mood expressions. Because the Schlosberg
dimensions have been replicated in several studies of
facial expressions, it was important to ascertain whether
or not any of the three dimensions obtained here corres
ponded to any of the Schlosberg dimensions. The major
axis in Schlosberg's model is Pleasantness-Unpleasantness,
and when this axis is superimposed on the circular order
ing of the Woodworth scales (Schlosberg, 1941), one end
110
meets the category of Love, Mirth, Happiness, and the other
meets the category of Anger. Dimension 2 in this study is
defined at one end by Happy and at the other end by Anger,
so this might be pleasantness-Unpleasantness. One diffi
culty in this correspondence is that Happy also defines one
end of dimension 1, whereas in Schlosberg's model Love,
Mirth, Happiness is zero on the second dimension. Beyond
this incongruence, it did not appear that any of the dimen
sions here could be interpreted as Sleep-Tension or Atten-
tion-Rejection, the other two dimensions in Schlosberg's
model. Dimension 3 in this was tentatively called tension,
but it did not range from sleep to tension. Rather, it
seemed to range from fear to relaxation, as the Happy
expressions were at one end and the Fear-Anxiety expres
sions were at the other end. It could be that the
Schlosberg dimensions apply only to facial expressions, and
not to all modes of mood expression, as there was some
indication that the relations among the photographs used in
this study could be accounted for by two dimensions. How
ever, the analysis of the line-drawn faces clearly supported
three dimensions, and it would seem to be very inefficient
to have different models for different modes of facial
expression. What is more likely is that the photographs
used in this study were poor expressions of Fear-Anxiety,
and that two dimensions were obtained because there was no
cluster for Fear-Anxiety.
Ill
The fact that the most important characteristic of
the coordinate matrices was the separation of the expres
sions into four mood clusters did not preclude any
attempts to use the dimensions in interpreting differences
of coordinates on them. The differences among the expres
sions on each dimension were interpreted in two ways. One
was to aid the verbalization of describing differences
among the expressions on the dimensions that were consid
ered secondary for a particular mood. It was suggested
that variability on the first dimension, excepting Happy
and Sad, indicated differences in inhibition and disinhi-
bition; and variability on the third dimension, excepting
Fear-Anxiety, indicated differences in tension. A label
suggested to account for the bipolarity of Happy and Angry
on the second dimension was social-antisocial, but this
did not reasonably explain differences among Sad and Fear-
Anxiety on this dimension. These labels were merely sug
gestive and temporarily useful, and were not thought to
be salient dimensions of mood expressions. The efficacy
of these labels, or other possible labels, in describing
the three dimensions would have to be determined by
obtaining ratings of the mood expressions on such psycho
logical characteristics and relating the ratings to the
coordinate matrices.
A second way in which the dimensions were used was
112
in interpreting differences among the expressions for one
mood on the primary dimension for that mood. Thus, dif
ferences among the projections of the Happy and Sad
expressions on dimension 1 were interpreted in terms of
greater and lesser degrees of Happiness and Sadness. Sim
ilar explanations were invoked for differences among the
projections of the Angry expressions on dimension 2 and
of the Fear-Anxiety expressions on dimension 3. While no
substantiating data for such explanations was gathered in
this study, there is some evidence that differences among
the adjectives are related to degree of intensity of feel
ing or degree of affect. Jacobs, Copek, § Meehan (1959)
and Johnson and Myers (1967) obtained ratings of degree
of affect and degree of intensity, respectively, for a
set of adjectives that included some of those used in this
study. In both these studies the adjectives represented
four mood dimensions: Anger, Fear, Happiness, and Depres
sion. For the common adjectives, the rank order on the
ratings followed the rank order on the dimensions, except
for the Sad adjectives. Whereas the adjective "sad" had
the highest projection, it had the lowest rating of the
three adjectives. The adjective "depressed" had the high
est rating, and "gloomy" was in the middle, but these two
adjectives always had lower projections than "sad." A
possible explanation for this discrepancy is that the
113
adjectives were rated on Depression, not Sadness. It
would be necessary to obtain similar ratings for the stick
figures and faces in order to verify their ordering on
degree of affect.
Mood Clusters. The dominance of the clustering of
the mood expressions in the coordinate matrices over the
significance of the dimensions was certainly the most sig
nificant outcome of the multidimensional scaling analysis.
This result was consistent across all five groups, as well
as in the analysis of the overall mean ratings. About the
only major difference among the configurations was the
size of the space in which the four mood clusters were
located. That is, there appeared to be only stretching
or shrinking of the distances between the four mood clus
ters , and that one configuration could be obtained from
another by a suitable constant transformation of the
inter-cluster distances.
The mood clusters that did emerge were exactly the
ones for which the expressions were selected. The only
set of 12 mood expressions that did not result in all
expressions falling very close to one another was that of
Fear-Anxiety. At best, nine of'these expressions could be
said to represent the mood (group 1), and at worst, only
six seemed to be perceived as Fear-Anxiety (group 2). In
all cases there was a small gap between the two
114
expressions of fear (numbers 14 and 20) and the remaining
expressions of anxiety. This cluster, although not a
strong one, represents the merger of two mood qualities
that previously have not been paired together. Fear has
been postulated in the theories of Plutchik (1960) and
Tomkins (1962), and has appeared in Woodworth's scale of
facial expressions (1938), and in two rating scales
(Jacobs, et al., 1959; Johnson § Myers, 1967). Anxiety,
or Tension, however, is not included in these theories
or scales, but has emerged as a factor in three factor-
analytic studies of mood adjective check lists (Lorr, et.
al. , 1967; McNair § Lorr, 1964; and Nowlis, 1965). The
other three mood clusters represent mood qualities that
have been postulated in theories, have appeared in rating
scales, and have emerged in factor analyses. The consis
tency with which Angry, Happy, and Sad have been isolated
might indicate their status as being very basic and com
mon emotions or moods, and also the ease with which they
are distinguished from one another in all modes of expres
sion. The 12 expressions for each of these moods were
located very close to one another in the three-dimensional
space, and there were only a few instances where an
expression tended to wander from the cluster. In these
cases the direction in which an expression moved was
towards another cluster which contained expressions that
115
had some similarity with the deviant expression.
Because of these types of movement away from a
cluster, the configurations that were obtained might be
said to be a permeable class structure (Torgerson, 1965).
That is, there are clusters of expressions with similar
properties, but the expressions may belong to more than
one cluster. As was suggested earlier, this kind of
structure may open the possibility of obtaining meaningful
relations among the clusters. As evidence for something
like this, it should be noted that the distances between
the mood clusters are not all equal, which would occur if
the ratings were made on the basis of whether or not the
moods were the same or not the same. Rather, it was found
that the distances between Happy and the other mood clus
ters were the largest, and that the distance between Sad
and Fear-Anxiety was the smallest. The relative distances
between clusters might be indicative of the probability
with which these moods would be confused with one another,
or the degree to which one mood co-occurs with another.
In line with the latter possibility, Lorr, et al. (1967)
report correlations of .50 between Tension and Depression,
.29 between Tension and Anger, .14 between Depression and
Anger, and small negative correlations between Cheerful
ness and the other three. Similarly, McNair and Lorr
(1964) report correlations of .46 between Tension and
116
Depression, .52 between Tension and Anger, and .38 between
Anger and Depression, while Friendliness correlates nega
tively with these three mood factors. While the order of
these correlations do not always coincide with the order
of the distances between the mood clusters, the general
pattern is similar.
There have been some studies concerned with the
identification of moods in facial expressions that have
yielded confusion data, but it is difficult to relate that
data to the configurations here (Levy § Schlosberg, 1960;
Thompson 5 Meltzer, 1964; Tomkins § McCarter, 1964). The
difficulty is that the categories into which the facial
expressions can be sorted do not include all four mood
types used here, and furthermore, the categories are very
frequently similar to one another. For instance, in the
Thompson and Meltzer (1964) study, the categories included
Love, Happiness, Anger, Disgust, and Contempt, as well as
others, and, not surprisingly, Love and Happiness were
most frequently confused with each other, as were Anger,
Disgust, and Contempt. Sadness and Tension were not
included in their list of categories. Similarly, in Levy
and Schlosberg (1960), Anger, Disgust, and Contempt were
most frequently confused with one another, and in Tomkins
qnd'McCarter (1964) Anger and Disgust were confused.
There is very little support in any of these studies for
117
Contempt and Love, Mirth, Happiness to be adjacent mood
categories as they are in the circular ordering of the
Woodworth scale (Schlosberg, 1941). Unfortunately, in
none of these studies were the mood categories of Sadness
and Anxiety, or Tension, included as facial expressions
or as labels into which faces could be sorted. It is
quite likely that if these two moods were included they
would be confused with one another more frequently than
with any other mood.
Mode of expression. A third principal concern in
this study was with the effects that the various modes of
expression might have on the perceived similarity of mood.
Although it was not expected that mode of expression would
interact with mood in any peculiar or systematic way, and
the configurations pretty much confirm this expectation,
there are some interesting possibilities. One would be
for the 12 expressions of each mood to be gathered into
three or four sub-clusters on the basis of mode of expres
sion. If it makes no difference whether a facial expres
sion is line-drawn or a photograph, then there would be
three sub-clusters; otherwise, there would be four. Such
sub-clustering might occur if same mood, same mode pairs
of expression were perceived as more similar than same
mood, different mode pairs of expressions. This did not
seem to happen in any of the data matrices that were
118
multidimensionally scaled, but a somewhat opposite effect
may have occurred with the stick figures. When the simi
larity ratings for all 48 expressions were analyzed, most
of the stick figures appeared in the cluster of mood
expressions for which they had been selected. But when
the similarity ratings for the 12 stick figures were anal
yzed separately, there was not the same degree of cluster
ing into four mood groups. This situation could arise
when the stick figures are seen as more similar to other
expressions of the same mood than they are to stick fig
ures of the same mood. The general failure to obtain
clear mood clusters when only the stick figures were
multidimensionally scaled may be supporting evidence for
Ekman and Friesen's (1967) hypothesis that body position
and head orientation do not carry information about spe
cific moods, but communicate information about general
affective state. It could be that when a stick figure is
paired with a face or adjective these latter modes provide
situational cues and thereby lead to a reduction of ambi
guity in the stick figures. It is as if the stick figures
are capable of expressing more than one mood, or are asso
ciated with more than one mood state, and further infor
mation is needed in order for a specific emotion to be
identified. It has been suggested that low scores in the
identification of emotions in facial expressions are the
119
result of the removal of situational cues (Bruner §
Tagiuri, 1954; Frijda, 1958). Although there was no dif
ficulty in obtaining the mood clusters in the analysis of
line-drawn faces or photographs, it is quite likely that
identification of emotion in stick figures is dependent
on situational cues.
Another possible effect of mode of expression
interacting with mood would be for mode dimensions to
appear in the multidimensional space. This would be a
remote possibility, but is related to the issue of ana-
lyzability that has been attended to by Shepard (1964) and
Torgerson (1965). It could also be analogous to the third
type of class structure that Torgerson (1965) suggested,
where in addition to classes of stimuli there are also
quantitative dimensions. The correspondence to mood
expressions that differ in mode would be for both mood
clusters and additional dimensions, albeit qualitative,
for the various modes to appear. There was no evidence
for this type of structure to emerge, as in no case did
the fourth or succeeding dimensions appear to be mode
dimensions.
In the complete analyses, then, the appearance of
the different modes of expression did not disrupt in any
way the judgments of mood similarity. The configurations
were overwhelmingly defined in terms of categories of mood
120
and the structure was very clear. Shepard (1964) found
that quite inconsistent results were obtained when the
stimuli that were to be multidimensionally scaled con
tained different characteristics that competed for the
rater's attention. The different modes of expression
included in this study did have different characteristics
that might have competed for attention. The results indi
cate, however, that the subjects were capable of ignoring
these mode differences and attending exclusively to the
mood similarities and dissimilarities. Torgerson (1965)
expressed a concern for the appropriateness of the euclid
ean model when the process underlying judgments of simi
larities tended to be more cognitive than perceptual.
He suggested that the judgment process might become more
cognitive as more and more obvious perceptual differences
are introduced in the stimuli. The different modes of
expression in this study do contain obvious perceptual
differences, but the results do not indicate that the
euclidean model was inappropriate. This further supports
the conclusion that the judgments were based only on the
perceived similarity of mood, and not on any mode charac
teristics .
Beyond examining the configurations of the 48 mood
expressions for possible mood-mode interactions, one other
step was taken to determine if the number of dimensions or
mood clusters were different in the four matrices of same
mode expressions. Because these matrices were taken from
the 48 x 48 matrices for each group, an expected outcome
would be that the dimensionality and configurations would
be the same for each mode of expression as they were in
the overall analysis. That is, there would be three
dimensions and four mood clusters. The Stress values in
Table 8 and the coordinate matrices in Table 9 indicate
very clearly that there are three dimensions and four mood
clusters for the adjectives and line-drawn faces. In most
cases the Stress values are good to excellent, in terms of
Kruskal's (1964a) evaluation, even for group 2 which had
the poorest fit in three dimensions for all 48 expres
sions. Relating these Stress values to the null values
that Stenson and Knoll (1969) obtained for 10 points in
three dimensions, all except group 2 are well below the
null value of 10%. The mood clusters, furthermore, are
very much like the clustering of the expressions in the
complete analysis.
For the stick figures and photographs, however,
the dimensionality and configurations are different from
the adjectives and the faces, and somewhat different from
the complete analysis. The Stress values in Table 8 indi
cate that four dimensions are needed in some cases to
obtain a good fit for the stick figures, and the coordi
nate matrices in Table 9 indicate that the stick figures
122
do not readily fall into four mood clusters. The results
for the photographs, however, suggest that for some data,
especially AVE and group 3, only two dimensions are needed.
The reason for this is found in the coordinate matrices,
where it is seen that the photographs for Fear-Anxiety do
not cluster together away from the other moods. Rather,
they tend to fall close to the Sad cluster, somewhat as
they do in the configuration of all 48 expressions. The
consequent similarity of the configuration of these photo
graphs in two dimensions to the configurations of the
Lightfoot faces in two dimensions (Cliff § Young, 1968;
Gladstones, 1962; Shepard, 1962b) is not taken as support
for a two-dimensional model of facial expressions. It is
felt that a better sample of photographs for Fear-Anxiety
would restore the third dimension and the fourth mood
cluster. Such is not the case for the stick figures, how
ever. As was discussed above, it is felt that the stick
figures, by themselves, are not likely to communicate
information about specific moods, but when paired with
other modes of expression they can take on specific mean
ing. It should be noted again that the validity of these
inferences about the dimensionality and configuration of
mood expressions within each mode needs to be tested
against data that is obtained from subjects who are judg
ing the similarity of mood expressions that do not vary in
mode.
123
The focus of this study was on the configuration
of mood expressions as revealed through multidimensional
scaling. Its function was to clarify some discrepancies
in previous research as to the number of dimensions of
mood or emotion. The strategy here of systematically
varying mood and mode of expression was very successful in
serving the function of this research. The results have
shown quite conclusively that the number of dimensions
underlying ratings of mood similarity was a function of
the number of mood qualities represented in the sample of
mood expressions. The fact that the moods were expressed
in different modes did not have any effect on the dimen
sionality of the space. The multidimensional configura
tion was best characterized by the appearance of four dis
tinct mood clusters that corresponded to the four mood
qualities for which the stimuli were selected.
There are some implications of these results that
need to be mentioned. It is not felt that the three
dimensions obtained here represent the final word in the
number of dimensions of mood. The dimensionality is
likely to increase as more mood qualities are included in
the sample of stimuli that are multidimensionally scaled.
In order for more dimensions to emerge, it appears that
there would have to be several expressions of each addi
tional mood quality, although not necessarily as many as
124
were included in this study. There is an analogy here to
the strategy that Comrey (1961) has used in investigating
the personality domain. Rather than analyzing correla
tions among individual adjectives, he first builds little
tests that are composed of several adjectives, and then
factors the correlations among the tests. One weakness of
the multidimensional scaling studies of the Lightfoot
faces is that there were no "tests," only "items."
This strategy could be extended by developing
"tests" for the several mood qualities that have been
hypothesized elsewhere, but were not included in this
study. There is certainly no lack of adjectives or photo
graphs from which to sample, and it would not be difficult
to obtain line-drawn faces and stick figures. A multidi
mensional scaling of mood expressions that include several
such "tests" could then be used to determine the relation
ships among the mood qualities in the multidimensional
configuration. It may turn out that some of the moods are
not too distant from each other, as Sad and Fear-Anxiety
were in the three-dimensional structure obtained here. An
immediate necessity is to build separate "tests" for Fear
and Anxiety, or Tension, to see if they are as similar as
they appeared to be in this study.
Because it was found that the two types of faces
used were perceived as being similar to adjectives for the
125
same mood, it is possible that a check list could be
developed for assessing individual mood states that either
was composed of just faces or both faces and adjectives.
It is not likely that stick figures would be effective in
such a measuring instrument.
Conclusions
1. Consistent and reliable data were obtained from sub
jects who made many ratings of perceived mood similar
ity. These ratings were not only stable over a short
period of time, but they also contained few violations
of the triangular inequality.
2. While there were some reliable differences among the
subjects' ratings, they were small and did not obscure
the preponderant communality among the entire set of
ratings.
3. The ratings of perceived similarity among mood expres
sions were best accounted for, psychologically and
statistically, by a three-dimensional spatial repre
sentation.
4. In this spatial representation it was difficult to
make meaningful interpretations of the dimensions as
psychological constructs that corresponded to some
characteristics of mood.
5. The most salient interpretation of the spatial repre
sentation was in terms of the four clusters of mood
expressions that were located at unequal distances
from each other. The clusters were easily identified
as moods of Anger, Fear-Anxiety, Happiness, and Sad
ness, and, except for Fear-Anxiety, all 12 expressions
selected for each mood l^ere located in their appropri
ate cluster.
In the spatial representation of the 48 mood expres
sions, there were no sub-structures or axes that dif
ferentiated mood on the basis of mode of expression.
However, a separate analysis of the stick figures
indicated that this type of expression, unlike the
other modes, may not communicate information about
specific mood states.
CHAPTER VII
SUMMARY
A persistent concern in the affective domain has
been the identification of basic dimensions of mood or
emotion. One approach to this problem has involved multi
dimensional scaling analyses of facial expressions or
semantic differential analyses of facial expressions. The
results of these types of studies have been quite consis
tent in finding at least two similar dimensions. Another
approach, however, has involved factor analyses of mood
adjective check lists. These studies, however, have not
only found a larger number of dimensions, but the dimen
sions have also been of a different nature than the dimen
sions underlying facial expressions.
In order to investigate these differences, a
strategy was adopted to select four mood dimensions that
had been consistently identified in factor-analytic
studies, and that also had been hypothesized in theories
of emotion, and then select expressions of these four
moods that varied in mode of expression. The four moods
were Anger, Fear-Anxiety, Happiness, and Cheerfulness.
The four modes were adjectives, stick figures, line-drawn
127
128
faces, and photographed faces. Three expressions were
chosen for each of the 16 mood-mode combinations, yielding
a total sample of 48 mood expressions.
Ratings of perceived similarity were obtained for
approximately 90% of all possible pairs of the 48 expres
sions, with 44 pairs of expressions being repeated for the
purpose of identifying subjects who may have been respond
ing in a random manner. An analysis of the repeated mea
surements did not lead to the deletion of any of the 35
introductory psychology students who participated in the
study. Prior to the multidimensional scaling analysis,
the ratings were subjected to an individual differences
analysis. Five groups of subjects were formed on the
basis of information provided by a cluster analysis of the
ratings and by a cluster analysis of factor scores derived
from a Tucker and Messick points-of-view analysis. The
correlations among the mean ratings for these five groups
were quite high, but there were indications that there
were some reliable differences among the groups. Conse
quently, separate multidimensional scaling analyses were
performed for each of the sets of group data, and a sixth
analysis was performed on a set of data that was obtained
by averaging over all subjects.
The final solution for each of these analyses was
a three-dimensional configuration. This decision was
based on considerations of Kruskal’s Stress values and the
129
psychological meaningfulness of the solutions at various
dimensionalities, with the latter assuming the dominant
role. An attempt was made to interpret the three dimen
sions in terms of Schlosberg's model of facial expressions
or other meaningful psychological constructs of mood.
These were not only unsuccessful, but it was more meaning
ful to disregard the dimensions and focus on the four mood
clusters that were located in the three-dimensional space.
These clusters were easily identified as Angry, Fear-
Anxiety, Happy, and Sad moods, and they contained the 12
expressions that had been selected for each mood quality.
There was no overlap among the clusters, but there was
varying distance between the clusters, with Sad and Fear-
Anxiety being the two closest moods. Furthermore, within
each mood cluster there was no tendency for sub-clusters
to appear that corresponded to expressions of the same
mode.
Additional multidimensional scaling analyses were
performed on the four sets of ratings for same mode
expressions. These analyses suggested that the stick fig
ures may not communicate information about specific mood
states, as the other modes of expressions do. It was con
cluded that similarity of mood expressions is a function
of the specific mood that is expressed, and not a function
of general mood characteristics or mode of expression.
REFERENCES
REFERENCES
Abelson, R. P., § Sermat, V. Multidimensional scaling
of facial expressions. Journal of Experimental
Psychology, 1962 , 63_, 546-554.
Borgatta, E. F. Mood, personality, and interaction.
Journal of General Psychology, 1961, 6£, 105-137.
Bradley, P. A., § Cliff, N. Multidimensional scaling
and the measurement of mood. Paper presented at
the meeting of the Western Psychological Associa
tion, San Diego, March, 1968.
Brown, R. Social psychology. New York: The Free Press,
196^---- ----
Bruner, J. S., § Tagiuri, R. The perception of people.
In G. Lindzey (Ed.), Handbook of social psychology.
Vol. 2. Cambridge: Addison-M/esley, 1954.
Carroll, J. D., § Chang, J. J. A new method for dealing
with individual differences in multidimensional
scaling. Bell Telephone Laboratories, Murray
Hill, New Jersey, 1969 (mimeo).
Cliff, N. Multidimensional scaling and cognition:
I. Relations among judgments obtained under vary
ing conditions. Technical Report No. 1, November,
1965, University of Southern California, Los
Angeles.
Cliff, N. Orthogonal rotation to congruence. Psycho-
metrika, 1966, 31_, 33-42.
Cliff, N. The "idealized individual" interpretation of
individual differences in multidimensional scal
ing. Psychometrika, 1968, 3_3, 225-232.
Cliff, N., 8 Young, F. W. On the relation between uni
dimensional judgments and multidimensional scaling.
Organizational Behavior and Human Performance,
1968, 3, 269-285. ---- ----------------------
131
Comrey,
Davitz,
Ekman,
Ekman,
Ekman,
Ekman,
Engen,
Engen,
Engen,
Frijda,
Frijda,
Gage, N
132
A. L. Factored homogeneous item dimensions in
personality research. Educational and Psychologi
cal Measurement, 1961, 21, 417-451.
J. R. (Ed.). The communication of emotional mean
ing . New York! McGraw-Hill, 1964.
P. Body position, facial expression, and verbal
behavior during interviews. Journal of Abnormal
and Social Psychology, 1964, 68, 295-501.
P. Communication through non-verbal behavior: A
source of information about an interpersonal rela
tionship. In S. S. Tomkins and C. E. Izard (Eds.),
Affect, cognition, and personality. New York:
Springer, 1965. (a)
P. Differential communication of affect by head
and body cues. Journal of Personality and Social
Psychology, 1965, 2_, 726-735. (FJ
P., § Friesen, W. Head and body cues in the judg
ment of emotion: A reformulation. Perceptual
and Motor Skills, 1967, 24, 711-724.
T. Schlosberg*s model for facial expressions.
Paper presented at the meeting of the Eastern Psy
chological Association, Atlantic City, 1965.
T., Levy, N., § Schlosberg, H. A new series of
facial expressions. American Psychologist, 1957,
1 2 , 264-266.
T., Levy, N., § Schlosberg, H. The dimensional
analysis of a new series of facial expressions.
Journal of Experimental Psychology, 1958, 55, 454-
458. „ .
N. H. Facial expression and situational cues.
Journal of Abnormal and Social Psychology, 1958,
57, 149-154.
N. H., § Philipszoon, E. Dimensions of recogni
tion of expression. Journal of Abnormal and
Social Psychology, 1963, 66, 45-51.
. L., § Cronbach, L. J. Conceptual and methodo
logical problems in interpersonal perception.
Psychological Review, 1955, 62^, 411-422.
133
Gladstones, W. H. A multidimensional study of facial
expression of emotion. Australian Journal of
Psychology, 1962, 1_4, 95-100.
Gollob, H. F. Confounding of sources of variation in
factor-analytic techniques. Psychological Bulle
tin, 1968, 70, 330-344.
Guiiford, J. P. The nature of human intelligence. New
York: McGraw-Hill, 1967.
Guttman, L. A general nonmetric technique for finding the
smallest coordinate space for a configuration of
points. Psychometrika, 1968, j53, 469-506.
Harvey, 0. J., Hunt, D. E., § Schroder, H. M. Conceptual
systems and personality organization. New York:
Wiley,'1961. ---- -------------- --
Hulin, W. S., 6 Katz, D. The Frois-Wittman pictures of
facial expressions. Journal of Experimental Psy-
chology, 1935, 18, 482-498.
Jacobs, A., Copek, L., § Meehan, J. P. The development of
an adjective checklist to measure affective states.
Psychological Newsletter, 1959, 10, llS^-llS.
Johnson, E., § Myers, T. I. The development and use of
the primary affect scale. Research Report No. 1,
July, 1967, Naval Medical Research Institute,
Bethesda, Maryland.
Kauranne, V. Qualitative factors of facial expression.
Scandinavian Journal of Psychology, 1964, 5,
136-142.
Kruskal, J. B. Multidimensional scaling by optimizing
goodness-of-fit to a nonmetric hypothesis. Psy
chometrika, 1964* 29_, 1-28. (a)
Kruskal, J. B. Nonmetric multidimensional scaling: A
numerical method. Psychometrika, 1964, 29, 115-
130. (b) "
Levy, N. , 8 Schlosberg, H. Woodworth scale, values of the
Lightfoot pictures of facial expression. Journal
of Experimental Psychology, 1960, 6£, 121-125.
134
Lorr, M., Datson, P., § Smith, I. R. An analysis of mood
states. Educational and Psychological Measure
ment, 1967, 27, 89-96"!
Lorr, M., § Lyerly, S. B. Conference on cluster analysis
of multivariate data. Final Report, June, 1967,
Catholic University of America, Washington, D. C.
McGee, V. E. The multidimensional analysis of "elastic"
distances. The British Journal of Mathematical
and Statistical Psychology, 1966, 19, 181-196.
McGee, V. E. Multidimensional scaling of N sets of simi
larity measures: A nonmetric individual differ
ences approach. Multivariate Behavioral Research,
1968, 3, 233-248.
McNair, D. M., 8 Lorr, M. An analysis of mood in neurot
ics. Journal of Abnormal and Social Psychology,
1964, F9, ' 620-627. -------- -----------------
Mordkoff, A. M. A factor analytic study of the judgment
of emotion from facial expression. Journal of
Experimental Research in Personality, 1967, 2,
80-85.
Messick, S. The perceived structure of political rela
tionships. Sociometry, 1961, 24_, 270-278.
Messick, S. , § Abelson, R. P. The additive constant
problem in multidimensional scaling. Psychomet
rika, 1956, 21, 1-15.
Messick, S., § Kogan, N. Personality consistencies in
judgment: Dimensions of role constructs. Multi
variate Behavioral Research, 1966, 1^, 165-175.
Nowlis, V. Research with the mood adjective checklist.
In S. S. Tomkins and C. E. Izard (Eds.), Affect,
cognition, and personality. New York: Springer,
T9oin
Osgood, C. E. Dimensionality of the semantic space for
communication via facial expressions. Scandina
vian Journal of Psychology, 1966, 7., 1-3TTI
Osgood, C. E., 8 Succi, G. J. A measure of relation
determined by both mean difference and profile
information. Psychological Bulletin, 1952, 49,
251-262.
135
Osgood, C. E., Succi, C. J., § Tannenbaum, F. H. The mea
surement of meaning. Urbana: University oT
Illinois Press, 1957.
Pedersen, D. The measurement of individual differences in
perceived personality-trait relationships and
their relation to certain determinants. Journal
of Social Psychology, 1965, (i5, 233-258.
Plutchik, R. The multifactor-analytic theory of emotion.
Journal of Psychology, 1960, 5£, 153-171.
Plutchik, R. The emotions: Facts, theories, and a new
model. Hew York: Random House, 1962.
Plutchik, R. What is an emotion? Journal of Psychology,
1965, 61, 295-303.
Plutchik, R. Multiple rating scales for measurement of
affective states. Journal of Clinical Psychology,
1966, 22, 423-425.
Rosenberg, S., Nelson, C., 8 Vivekananthan, P. S, A
multidimensional approach to the structure of per
sonality impressions. Journal of Personality and
Social Psychology, 1968, 9_, 283-294.
Ross, J. A remark on Tucker and Messick's "points of
view" analysis. Psychometrika, 1966, _31, 27-31.
Sarbin, T. R., 8 Hardyck, C. D., Conformance in role per
ception as a personality variable. Journal of
Consulting Psychology, 1955, 1£, 109-111.
Schacter, S., 8 Singer, J. E. Cognitive, social, and
physiological determinants of emotional state.
Psychological Review, 1962, 6£, 379-399.
Schlosberg, H. A scale for the judgment of facial expres
sions. Journal of Experimental Psychology, 1941,
29, 497-mn------------------ —
Schlosberg, H. The description of facial expression in
terms of two dimensions. Journal of Experimental
Psychology, 1952, 44, 2 2 9 -J T T .
Schlosberg, H. Three dimensions of emotion. Psychologi
cal Review, 1954, 61, 81-88.
136
Shepard, R. N. The analysis of proximities: Multidimen
sional scaling with an unknown distance function:
I. Psychometrika, 1962 , 21_, 125-140. (a)
Shepard, R. N. The analysis of proximities: Multidimen
sional scaling with an unknown distance function:
II. Psychometrika, 1962 , 27_, 219-246. (b)
Shepard, R. N. Attention and the metric structure of the
stimulus space. Journal of Mathematical Psycholo
gy, 1964, 1, 54-87:
Shepard, R. N. Metric structures in ordinal data. Jour
nal of Mathematical Psychology, 1966, 3 , 287-315.
Sherman, C. R., 8 Young, F. W. Nonmetric multidimensional
scaling: A monte carlo study. Proceedings of the
76th Annual Convention of the American Psychologi
cal Association: 1968, 3^, 207-208.
Stenson, H. H., § Knoll, R. L. Goodness of fit for random
rankings in Kruskal's nonmetric scaling procedure.
Psychological Bulletin, 1969, 7JL, 122-126.
Sweeney, D. R., Tinling, D. C., Eby, L. A., 8 Schmale,
A. H., Jr. Factor analytic studies of four
expressive modes of emotion. Proceedings of the
76th Annual Convention of the American Psychologi
cal Association, 1968, 5, 169-170.
Thompson, D. F., 8 Meltzer, L. Communication of emotional
intent by facial expression. Journal of Abnormal
and Social Psychology, 1964, 68, 129-135.
Tomkins, S. S. Affect, imagery, consciousness. New York:
Springer, 1962.
Tomkins, S. S. Simulation of personality: The interrela
tionships between affect, memory, thinking, per
ception, and action. In S. S. Tomkins and
S. Messick (Eds.), Computer simulation of person
ality. New York: Wiley, 1963.
Tomkins, S. S., § McCarter, R. What and where are the
primary affects? Some evidence for a theory.
Perceptual and Motor Skills, 1964, 18^, 119-158.
Torgerson, W. S. Theory and methods of scaling. New York:
Wiley, 1958.
137
Torgerson, W. S. Multidimensional scaling of similarity.
Psychometrika, 1965, 3£, 379-393.
Torgerson, W. S., § Mues.er, G. Informal notes on
Torgerson and Mueser's IBM-7090 program for multi
dimensional scaling. Johns Hopkins University,
Baltimore, Maryland, 1962 (mimeo).
Tucker, L. R., § Messick, S. An individual differences
model for multidimensional scaling. Psychometrika,
1963, 28, 333-367.
Veldman, D. J. FORTRAN programming for the behavioral
sciences" New York: Holt^ 1967.
Walters, H. A., § Jackson, D. N. Group and individual
regularities in trait inference: A multidimen
sional scaling analysis. Multivariate Behavioral
Research, 1966, 1^, 145-163.
Ward, J. H., Jr. Hierarchical grouping to optimize an
objective function. Journal of the American Sta
tistical Association, 1963, 58, 236-244.
Weksel, W., 6 Ware, E. E. The reliability and consistency
of complex personality judgments. Multivariate
Behavioral Research, 1967, 2_, 537-5TF!
Woodworth, R. S. Experimental psychology. New York:
Holt, 1938.
Young, F. W., Appelbaum, M. I. Nonmetric multidimen
sional scaling: The relationship of several
methods. L. L. Thurstone Psychometric Laboratory
Report No. 68, Chapel Hill, North Carolina, 1968.
Young, F. W., § Torgerson, W. S. TORSCA, A FORTRAN IV
program for Shepard-Kruskal multidimensional scal
ing analysis. Behavioral Science, 1967, 1 2 , 498.
APPENDIX
139
TABLE A1
Classification of stimulus pairs on the basis of similar
ity of mood and mode. Frequency and percentage of each
type in (a) all possible pairs, (b) sampled pairs, (c)
repeated items.
Mood
Same Differ.
48 216
Same (4.3%) (19.1%)
Mode
216 648
Different (19.1%) (57.4%)
(a)
48 192
Same (4.8%) (19.3%)
Mode
216 540
Different (21.7%) (54.2%)
(b)
2 9
Same. (4.5%) (20.5%)
Mode
9 24
Different (20.5%) (54.5%)
(c)
140
TABLE A2
Number of items deleted from blocks that represent nine
stimulus pairs. Code for identifying stimulus sets:
A - Anger, F-A - Fear-Anxiety, H - Happy, S - Sadness;
1 - adjectives, 2 - stick figures, 3 - faces, 4 - photos
Mood
A F-A H S
Mood Mode
Mode Mode Mode Mode
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
1 0 0 0 0 1 2 1 1 1 2 1 1 1 2 12
A
2 0 0 0 0 2 1 2 2 2 1 1 1 1 12 1
A
3 0 0 0 0 2 1 1 1 1 2 1 2 1 2 11
4 0 0 0 0 2 1 1 1 2 1 2 1 2 12 1
1 0 0 0 0 1 2 1 1 1 2 2 1
2 0 0 0 0 2 1 2 1 2 1 1 2
F-A
3 0 0 0 0 1 1 1 2 2 1 1 1
4 0 0 0 0 1 2 2 1 1 2 11
1 0 0 0 0 1 2 12
2 0 0 0 0 1 12 2
n
3 0 0 0 0 1 2 11
4 0 0 0 0 1 2 11
1 0 0 0 0
2 0 0 0 0
S
3 0 0 0 0
4 0 0 0 0
141
TABLE A3
Descriptive Statistics for Individuals
Group Person Mean s.d. VTIa
b
rl, 2
Sex
1 5.9 2.5 4.2 .86 M
2 6.2 3.1 4.9 .90 F
3 6.3 3.1 5.3 .83 M
1 4 6.4 2.9 6.6 .70 M
16 7.2 3.0 2.7 .98 M
18 6.0 3.5 5.4 .95 F
22 5.7 3.4 11.1 .95 M
31 5.5 3.1 5.0 .82 F
25 4.9 2.7 8.9 .90 M
O
33 5.2 2.6 5.7 .92 M
L
34 4.8 2.8 12.9 .56 M
35 4.7 2.8 10.3 .78 F
9 5.2 3.0 9.4 .92 M
10 5.3 2.9 7.2 .83 F
12 5.7 3.2 9.2 .87 F
3 13 5.5 2.9 5.7 .91 F
14 5.3 2.7 5.0 .90 M
17 5.2 2.8 5.6 .81 M
30 5.3 3.2 13.9 .74 F
7 5.2 2.7 6.4 .85 F
8 5.8 3.1 10.0 .85 M
11 5.8 2.3 5.4 .54 F
15 5.2 . 2.8 8.9 .77 F
4 21 5.1 2.7 5.2 .77 M
23 5.0 2.7 6.9 .92 F
24 5.1 2.4 3.7 .64 M
28 4.8 . 2.8 5.9 .86 F
32 5.0 2.4 4.3 .60 M
5 5.2 2.4 2.8 .81 M
20 6.2 2.6 2.5 .88 F
5
26 5.4 1.9 0.9 .73 M
29 5.1 2.9 9.2 .69 M
aViolations of triangular inequality (VTI) are given
in percentages.
^Test-retest correlation over the 44 repeated items.
142
TABLE A4
Contingency tables and chi-square values for clusters
based on distances.
Booklet Group
Booklet 1
x2
Group
A B C D E
A 1 2 1
B 1 3 1
C 3 1
Booklet 2
D 2 2 2
E 1 5
F 2 5 1 43.60**
A 2 1
B 2 2 3
Booklet 3
C 2 2
D 1 3 1
E 2 4 1
F 1 3 3 31.82*
A . 3 1
B 2 3 4
Booklet 4 C 1 5 7
D 1 1
E 1 2 2 36.05**
*p<.05
**p<.01
143
DIRECTIONS FOR SIMILARITY OF MOOD RATINGS
To express our moods and emotions we use a variety
of cues, and it is usually the totality of these cues that
express the particular mood. For another person to per
ceive our mood states he must rely on these cues, piecing
them together to arrive at an inference. Three of these
cues are facial expressions, posture, and words.
In this booklet several mood qualities are
expressed in drawn and photographed facial expressions, in
stick figures, and in mood adjectives. These stimuli are
presented to you in pairs. The mood qualities expressed
in the two stimuli vary from quite similar to quite dis
similar in meaning. Your task is to judge the extent to
which the mood quality of the stimuli in the pair are
similar. You express your judgment by assigning it a num
ber from 1 to 9. If you think that the mood qualities are
quite similar, use the low numbers; if you think the mood
qualities are quite dissimilar, use the high numbers. Use
the middle numbers for intermediate degrees of mood simi
larity. When you have decided which number to assign to
your judgment, circle the number on the answer sheet.
A word of caution regarding the rating scale: you
are not judging sameness and oppositeness, but similarity
and dissimilarity. Therefore, if you verbalize one pair
as being nearly opposite in mood quality, and another pair
as being quite different from each other, then both judg
ments should be assigned the same number (and likewise for
pairs that are verbalized as exactly the same and quite
similar).
Think of each stimulus as expressing the current
mood of a student you are meeting for the first time. You
should ask yourself if the mood expressed by the first
stimulus is similar or dissimilar to the mood expressed by
the second stimulus. Indicate your opinion of the degree
of similarity or dissimilarity by marking the appropriate
number on the rating scale for that item.
You should work steadily, recording your first
impression. Please answer each item. Omitting items will
cause more difficulty in this research than answering when
you are unsure of how you feel about a particular item.
To provide you with some warm up, a single page of
practice items is included in the envelope. Take a few
minutes to look at it, but do not mark any numbers on the
answer sheet yet. If you wish you may write them on the
practice page.
DO NOT TURN THE PAGE UNTIL. YOU ARE TOLD TO DO SO

Linked assets

University of Southern California Dissertations and Theses

Conceptually similar

PDF

Figural And Symbolic Divergent-Production Abilities In Adults And Adolescents

PDF

A Factor Analysis Of The Symbolic-Evaluation Abilities

PDF

Factorial Stability As A Function Of Analytical Rotational Method, Type Of Simple Structure, And Size Of Sample

PDF

The Role Of Intellectual Abilities In Concept Learning

PDF

A Rotational Approach To Psychological Invariance

PDF

A Factor Analysis Of The Semantic-Evaluation Abilities

PDF

A Monte Carlo Evaluation Of Interactive Multidimensional Scaling

PDF

The Effect Of Subject Sophistication On Ratio And Discrimination Scales

PDF

Some Aspects Of The Dimensionality Of School Adjustment Of Fifth Grade Boys

PDF

Acculturation And Value Change

PDF

Measuring Thought Process As An Ego Function In Schizophrenic, Mentally Retarded And Normal Adolescents By Means Of The Rorschach

PDF

The Development Of Attitude Scales To Predict Accident Repeater And Moving Violator Drivers

PDF

A Factor Analysis Of The Figural-Evaluation Abilities

PDF

The Relationship Of Multidimensional Scaling Spaces Of Trait Adjectives For Different Reference Persons

PDF

Methodological Problems In The Identification Of Suicidal Behavior By Means Of Two Personality Inventories

PDF

A Preliminary Study Of The Dimensions Of Future Time Perspective

PDF

A Multidimensional Similarities Analysis Of Twelve Choice Probability Learning With Payoffs

PDF

A Monte Carlo Evaluation Of A Computer-Interactive Extended Transitivity Dominance Scaling Model

PDF

A Study Of Criteria Of Social Perception And Some Related Variables

PDF

Creativity In Children: A Study Of The Relationship Between Temperament Factors And Aptitude Factors Involved In The Creative Ability Of Seventh Grade Children With Suggestions For A Theory Of C...

Asset Metadata

Creator Bradley, Paul Arthur (author)

Core Title A Multidimensional Scaling Of Mood Expressions

Degree Doctor of Philosophy

Degree Program Psychology

Publisher University of Southern California (original), University of Southern California. Libraries (digital)

Tag OAI-PMH Harvest,psychology, general

Language English

Contributor Digitized by ProQuest (provenance)

Advisor Cliff, Norman (committee chair), Guilford, Joy P. (committee member), Lasswell, Thomas E. (committee member)

Permanent Link (DOI) https://doi.org/10.25549/usctheses-c18-385717

Unique identifier UC11361845

Identifier 7008517.pdf (filename),usctheses-c18-385717 (legacy record id)

Legacy Identifier 7008517.pdf

Dmrecord 385717

Document Type Dissertation

Rights Bradley, Paul Arthur

Type texts

Source University of Southern California (contributing entity), University of Southern California Dissertations and Theses (collection)

Access Conditions The author retains rights to his/her dissertation, thesis or other graduate work according to U.S. copyright law. Electronic access is being provided by the USC Libraries in agreement with the au...

Repository Name University of Southern California Digital Library

Repository Location USC Digital Library, University of Southern California, University Park Campus, Los Angeles, California 90089, USA