Close
About
FAQ
Home
Collections
Login
USC Login
Register
0
Selected
Invert selection
Deselect all
Deselect all
Click here to refresh results
Click here to refresh results
USC
/
Digital Library
/
University of Southern California Dissertations and Theses
/
The Relationship Of Multidimensional Scaling Spaces Of Trait Adjectives For Different Reference Persons
(USC Thesis Other)
The Relationship Of Multidimensional Scaling Spaces Of Trait Adjectives For Different Reference Persons
PDF
Download
Share
Open document
Flip pages
Contact Us
Contact Us
Copy asset link
Request this asset
Transcript (if available)
Content
THE RELATIONSHIP OP MULTIDIMENSIONAL SCALING SPACES
OP TRAIT ADJECTIVES FOR DIFFERENT REFERENCE PERSONS
ty
Linda Margit Doherty
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 1973
INFORMATION TO USERS
This material was produced from a microfilm copy of the original document. While
the most advanced technological means to photograph and reproduce this document
have been used, the quality is heavily dependent upon the quality of the original
submitted.
The following explanation of techniques is provided to help you understand
markings or patterns which may appear on this reproduction.
1. The sign or "target" for pages apparently lacking from the document
photographed is "Missing Page(s)". If it was possible to obtain the missing
page(s) or section, they are spliced into the film along with adjacent pages.
This may have necessitated cutting thru an image and duplicating adjacent
pages to insure you complete continuity.
2. When an image on the film is obliterated with a large round black mark, it
is an indication that the photographer suspected that the copy may have
moved during exposure and thus cause a blurred image. You will find a
good image of the page in the adjacent frame.
3. When a map, drawing or chart, etc., was part of the material being
photographed the photographer followed a definite method in
"sectioning" the material. It is customary to begin photoing at the upper
left hand corner of a large sheet and to continue photoing from left to
right in equal sections with a small overlap. If necessary, sectioning is
continued again — beginning below the first row and continuing on until
complete.
4. The majority of users indicate that the textual content is of greatest value,
however, a somewhat higher quality reproduction could be made from
"photographs" if essential to the understanding of the dissertation. Silver
prints of "photographs" may be ordered at additional charge by writing
the Order Department, giving the catalog number, title, author and
specific pages you wish reproduced.
5. PLEASE NOTE: Some pages may have indistinct print. Filmed a s
received.
Xerox University Microfilms
300 North Zeeb Road
Ann Arbor, Michigan 48106
74-913 I
4
t
v
DOHERTY, Linda Margit, 1945- {
THE RELATIONSHIP OF MULTIDIMENSIONAL SCALING |
SPACES OF TRAIT ADJECTIVES FOR DIFFERENT |
REFERENCE PERSONS. f
University of Southern California, Ph.D., 1973 j
Psychology, clinical j
I
(
University Microfilms, A XEROX Company, Ann Arbor, Michigan j
THIS DISSERTATION HAS BEEN MICROFILMED EXACTLY AS RECEIVED.
UNIVERSITY OF SOUTHERN CALIFORNIA
TH E G RADUATE SCHO O L
U N IV E R S IT Y PARK
LOS ANG ELES. C A L IF O R N IA 9 0 0 0 7
This dissertation, w ritten by
Linda Marg i t Doherty
under the direction of h «r— . Dissertation C om
m ittee, and approved by a ll its members, has
been presented to and accepted by T h e G raduate
School, in p a rtia l fu lfillm e n t of requirements of
the degree of
D O C T O R O F P H IL O S O P H Y
O
Dean
dlSSERTATION COMMITTEE
....
Chairman
ACKNOWLEDGMENTS
The completion of this dissertation is a
culmination for me of several difficult years as a
graduate student. If it had not been for the help and
faith of several significant people in my life, I
certainly would not have completed my education. I wish
to express my appreciation to them now.
First, I would like to thank my father, Alan. In
many ways he was a man before his time. His deep
conviction that women deserve an equal place in the
professional, economic and social areas of our daily
lives was a tremendous early influence on developing my
own academic potential.
Second, I want to express my deep appreciation to
Dr. Norman Cliff. He has been both a friend and advisor
through my graduate career. Two of his qualities, his
infinite patience and fairness, were greatly responsible
for keeping me in graduate school and helping me to
complete my degree. Besides that, I always felt he had
a faith in my abilities, even when I had lost all
confidence in myself. I can never express my gratitude
to this man.
Third, I would like to thank some other very
ii
helpful professors. Dr. Edward Conolley and Dr. Prank
Pox have always given of their time and of themselves to
help me in my endeavors.
Last, I would like to thank my family. My
husband's unending faith in me and his commitment to a
partnership marriage were truly the reasons I was able to
complete my degree. Also, I would like to thank my son,
Richard, for help on part of this dissertation and my
son, Tad, just for being Tad.
iii
TABLE OF CONTENTS
Page
ACKNOWLEDGMENTS................................. ii
LIST OF TABLES................................ vi
LIST OF ILLUSTRATIONS........................... vii
Chapter
I. DEVELOPMENT OF THE PROBLEM............... 1
General Background
Multidimensional Scaling Studies
Using Trait Adjectives
Purpose
II. METHOD................................. 21
Construction of Stimulus Materials
Subjects
Administration
Constructing Hanno and Jones
Replication
Data Analyses
Analyses for Hanno and Jones
Replication
III. RESULTS............................... 38
Repeated Measurements
Multidimensional Scaling Analyses
Congruence of Configurations
Unidimensional Favorability Ratings
Hanno and Jones Replication
IV. DISCUSSION..............'.............. 75
Multidimensional Scaling Results
Relation of Favorability Ratings to
the Multidimensional Structures.^
Hanno and Jones Replication
iv
Chapter
V. SUMMARY
REFERENCES
APPENDIX
LIST OP TABLES
Table Page
1. Trait Adjectives Used as Stimuli............ 2 2 .
2. Trait Adjectives in Hanno and Jones
Replication............................. 30
3- Stress Values............................. 40
4. Coordinate Matrices ........................ 4l
5. Rotated Coordinates ....................... 45
6. Correlations Between Corresponding
Dimensions for Pour References.......... 49;
7. Procrustes Transformations for "Astronaut"
and "Surgeon"........................... 51
8. Fit of Transformed References.............. 54:
9. Prediction of Pavorableness Ratings for
Three Dimensions....................... 6l
10. Coordinates for Doctor and Politician .... 69
11. Prediction of Pavorableness and
Likableness Ratings ................... 72;
vi
LIST OP ILLUSTRATIONS
Figure Page
1. Transformation on Idealized Hypocrite,
Ordered Condition ....................... 57
2. Transformation on Idealized Hypocrite,
Mixed Condition......................... 58
3- Location of Points in Two Dimensions,
Ordered Astronaut and Surgeon with
Direction of Pavorableness Indicated ... 62
4. Location of Points in Two Dimensions, for
Mixed Astronaut and Surgeon with
Directions of Pavorableness Indicated ... 63
5. Location of Points in Two Dimensions for
Ordered Hypocrite and Killer with
Directions of Pavorableness Indicated ... 64
6. Location of Points in Two Dimensions for
Mixed Hypocrite and Killer with
Directions of Pavorableness Indicated ... 65
7. Location of Points in Two Dimensions for
Family Doctor and Politician with
Directions of Pavorableness and
Likableness Indicated ................... 73
vii
CHAPTER I
DEVELOPMENT OF THE PROBLEM
General Background
Past studies have demonstrated that multidimen
sional scaling is useful in determining psychologically
meaningful representations of how subjects view stimuli,
specifically trait adjectives. The present study is
concerned with the application of multidimensional scaling
to a set of trait adjectives, when these adjectives are
presented under varying changes in context.
Impression formation studies. Some of the back
ground for this study includes impression formation
studies. These studies show ways in which trait adjectives
have been studied in the past. Also, this present study
approaches the area of forming judgments from the view
point that the subject uses the information presented to
him and applies his own internal organization to it.
Since Anderson (1971) conceives of impression formation
studies as dealing with the integration or cognitive
processing of information in order to make a particular
judgment or respond with an attitude, impression formation
provides a general background.
1
2
Asch (19^6; 1952) contributed the early experi
mental research in which two groups of subjects listened to
lists of trait adjectives. The lists differed by only one
word, "warm", on one list and "cold" on the other. The
subject's task was to check off the appropriate component
of a bipolar adjective-pair. The frequency of positive
endorsements was then computed for each list, with the
result indicating that a change in this one central ad
jective can change the entire impression of a person.
Asch also studied other aspects of the problem such as the
context of traits, demonstrating that adjectives may change
in meaning from one context to the next.
Using a quantitative approach, Anderson (1962) has
applied mathematical models to impression formation in
extensive research. Basically, Anderson assumes that a
linear model may be applied to stimulus items using
analysis of variance procedures. The experimental pro
cedure Anderson used was quite simple and has been adapted
to several other studies. The stimuli were adjectives
independently scaled for likableness and representing
high, medium and low likability. Subjects had to rate the
likability of a person described by the combination of
these adjectives. An additive model was applied to the
obtained data and the discrepancy between observed and
predicted values was measured.
The purpose of Anderson's (1962) basic additive
study was to determine whether the likability response to
combinations of three adjectives* high (H)* medium (M) and
low (L)* could be described as the arithmetic mean. The
linear equation was Rijk = 1/3(&i+bj+c^) + + e* with
ai* bj and c^ being the psychological values of these H*
M* and L adjectives. l>ijk represented the discrepancy
from additivity and e was the amount of unreliability. If
there were perfect additivity* then d-j^^- would equal zero.
The advantage of using this model is the analysis of
variance form which allows for goodness-of-fit tests and
least-squares estimates of the psychological scale values.
In testing for nonadditivity* the study indicated that a
greater part of the subject's response behavior may be
accounted for by additivity. It was as if the subject
assigned a value to each adjective and then gave the mean
of these to each set of adjectives.
Later* Anderson (1964) refined his additive model
to E = c +k=lwkskJ c is an initial neutral impression* wk
is the weight attached to the kth stimulus and sk is its
scale value. Anderson used analysis of variance to test
for goodness of fit* as well as to provide a joint test
of the model and the scale of measurement. Using this
model controversies and further research in impression
formation developed.
Some researchers have been interested in the ways
in which the weight and scale parameters are combined
(e.g. Fishbein and Hunter, 1964; Anderson, 1965j and
Anderson and Jacobson, 1965), while others have been
interested in the comparison of models in impression for
mation (Wyer, 1969; Anderson, 1971). Most work has
assumed that the w and s values remain constant, while in
others, context-induced changes may occur. An example
would be primacy effects (Anderson and Barrios, I96I;
Anderson, 1968). The problems of stimulus configurability
is also an additional factor. Anderson (1971) stated that
multidimensional scaling might be better for representing
some complex cognitive structures. Other research further
substantiated Anderson's basic model. With an increase in
the set-size of stimuli, Anderson (1965) found that an
averaging model accounted for the data fairly well.
Gollob (1968) studied word combinations in sen
tences from an impression formation context. He first had
subjects rate the isolated sentence components on a good-
bad scale. Gollob then constructed sentences from the
adjectives, nouns and verbs, with the subjects rating
these on a good-bad semantic differential kind of scale.
First a simple additivity model was tested by an analysis
of variance procedure whereby the overall rating could be
predicted from the sum of the adjective and the predicate,
verb and object combined. This simple model of the subject
and predicate combining additively accounted for about 93$
of the total variance between sentences.
A modification of the simple ANOVA model is the
"good-had regression model" which Gollob postulated as
being one in which one obtains an evaluative rating of the
whole sentence by simply taking a linear combination of
the evaluative rating of the adjective and predicate. The
predicate variation may be further’decomposed into verb and
object main effects and verb and object interactions. When
the verb is positive, the contribution of the predicate
becomes increasingly less favorable as the goodness of the
object decreases. This linear combination of the evalua
tive ratings was further substantiated by finding factors
from a factor analysis of the data which corresponded to
the significant effects in the analysis of variance.
In general, these studies were representative of
the kind of research which used adjectives in combinations
in impression formation. These studies tended to be narrow
in scope, since they were only concerned with unidimen
sional data and interpretation. They were included to
provide background material for the more recent develop
ments. The next sections explore the development of
multidimensional scaling and its application in studies
concerned with the scaling of personality and trait
adjectives.
Metric and nonmetric multidimensional scaling. The
basic purpose of the multidimensional scaling analysis is
to obtain a spatial representation of the trait adjectives
6
in as small a space as is psychologically meaningful. The
results of a multidimensional scaling analysis yield a
matrix of numbers containing the projections, coordinates
or loadings on a set of orthogonal axes. The stimuli which
were judged to be most dissimilar will tend to be far
apart in the space and the ones most similar will appear
closest. The number of dimensions to be retained which
best account for the data are determined by two criteria,
a statistical method and psychological meaningfulness of
the solution at different dimensionalities. This problem
is analogous to determining the number of factors in factor
analysis and the decision is also the result of a com
promise in the criteria.
The techniques of multidimensional scaling have
changed since they were first devised. Torgerson (1958)
developed a metric multidimensional scaling, since it was
assumed that the input data represented a ratio scale of
Euclidean distances. With this assumption, it was a
straightforward procedure to obtain the coordinates on the
dimensions. The symmetric matrix was converted to a
matrix of scalar products and the principal axes solution
was obtained. If the data were assumed to be on an inter
val scale, it was necessary to solve for the additive
constant.
The main difficulty with metric multidimensional
scaling was the strong assumptions regarding the data.
7
Shepard (1962a, 1962b) introduced nonmetric methods by
providing the rationale and an algorithm for computations.
Since the scaling assumptions have been relaxed so that
ordinal information may be used as inputs while metric
information is the output, a wider variety of data may be
used. The input data is only ordinal,, but the distances
in the Euclidean space must be monotonically related to the
initial proximity measures. The idea is to find the mono
tonic transformation that leads to a spatial representa
tion whose distances best reproduce the order of distances
of the input data.
Nonmetric multidimensional scaling was expanded by
others. Kruskal (1964a., 1964b) developed a statistical
index of how well the transformed distances fit the raw
data. Kruskal's stress is a root-mean-square residual of
the difference between the true configuration and the
"true" dissimilarities resulting from some monotone dis
tortion of the interpoint distances divided by the true
distances. The best fitting configuration would then be
when stress is at a minimum. Other approaches to the
problem have been developed by McGee (1966) and Guttman
(1968).
Interpretation and rotation problem. Once the
projections on each dimension are obtained, the next
question is how the dimensions may be interpreted.
Labeling the dimensions is an arbitrary procedure and may
be done by using some methods such as the prior selection
of stimuli so that only a few characteristics are varying
in some known way, or asking the subjects to state the
criteria they used in forming the Judgments. A related
problem of interpretation concerns the orientation of the
axes in the space. Since Euclidean multidimensional
scaling solutions are unique only up to a similarity
transformation - rotation, translation, reflection and
uniform stretching and shrinking of the axes, procedures
may be required to reorient the axes. The techniques used
in factor analysis are applicable in multidimensional
scaling, but usually the principal axes solution has been
the one which is interpreted. Another way to interpret
the obtained configuration may be in terms of the shape or
structure of the configuration. The clusters of stimuli
lying close together may represent a common property and
this shape should be investigated.
Background material from both traditional impres
sion formation studies and multidimensional scaling
analysis have been included in this report because they
indicate the development of how adjectives have been
studied. In impression formation studies, it was assumed
that the trait adjectives represented only one dimension,
generally likability or favorability, and that the uni
dimensional scaling was an adequate representation of the
data. However, it is reasonable to assume that trait
9
adjectives may not be necessarily unidimensional by nature.
The development of multidimensional scaling represents a
procedure whereby trait adjectives may be scaled in more
than one dimension which is psychologically meaningful.
Multidimensional Scaling Studies Using Trait Adjectives
Several researchers have used multidimensional
scaling techniques as applied to verbal stimuli. These
studies (Gerjuoy and Aaronson, 1970; Friendly and
Glucksberg, 1970; Rosenberg, Nelson and Vivekananthan,
1968) in general relate to the perception of personality
traits, the individual differences in this perception and
the relationship of these data to some extrinsic person
ality or cognitive correlates. The multidimensional
scaling analyses are assumed to be an adequate representa
tion of the individual subject's ratings and the outside
judgments are chosen so that there might be a close
relationship between the multidimensional scaling struc
ture and these external measures.
Other studies have attempted to confirm the idea
that multidimensional scaling is useful in demonstrating
the relationship between stimuli in a psychologically
meaningful way. These studies (Cliff and Young, 1968;
Cliff, 1970; Hanno and Jones, 1973) have shown that
undimensional judgments often bear a simple quantitative
relationship to the multidimensional scaling structures.
10
This lends impressive support for the validity of the
obtained structures. The following sections will explore
some of the recent research in both of these kinds of
studies, concentrating primarily on studies which use trait
adjectives as the stimuli. In this way the major research
questions of this present investigation will be developed.
Walters and Jackson (1966) were interested in
multidimensionally scaling personality traits with
respect to the question of individual differences.
Similarity judgments, using 30 trait-descriptive adjec
tives, as well as other cognitive and personality tests
for all subjects were obtained. They used Tucker and
Messick's points of view analysis (1963) and identified
two idealized individuals in two separate random groups,
indicating that subjects make different inferences about
the relationships between traits. Walters and Jackson
also showed that these traits tended to be stable over
groups. However, they did not find any relationship
between the points of view and the cognitive and personal
ity characteristics. This may have been because of the
very large amount of variance represented in the first
dimension of the multidimensional scaling analysis and the
high similarity regarding the trait structures, since
there were only two points of view.
Boyd and Jackson (1967) used two discrete classes
of social information, attitude statements and behaviorally
11
oriented person descriptions in a multidimensional study
of similarities. This was to study the perceived relation
ship between impressions formed about a person from
behavioral information, the latter consisting of dimen
sions labeled as authoritarianism and the likelihood of
endorsement of attitude items reflecting components of
authoritarian ideology. They found that the hypothesized
three dimensions were present, and, in addition, the
attitude statements and person descriptions can be
perceived and interpreted in the common multidimensional
scaling framework. Primarily, this study shows that
different classes of social objects can be included in the
same framework and the salient features of these pieces of
information may be extracted to form impressions of
people.
The following studies are examples of how multi
dimensional scaling and trait adjectives were studied in
different ways with a primary interest in the obtained
structures themselves. These were related to other
external measures, but not necessarily in a quantitative
way.
Gerjuoy and Aaronson (1970) were concerned with
the multidimensional scaling of personality traits which
clinical psychologists use to describe an individual's
personality. They wanted to find the relationships
between traits on numerous previously determined unidi-
12
mensional scales. The conclusion was that a two-dimension
al configuration was appropriate, indicating redundancy
among the previously used undimensional scales. The con
figuration from multidimensional scaling analysis demon
strated the relationship between these adjectives.
Friendly and Gluoksberg (1970) were interested in
the change of the multidimensional structure of trait-
adjectives as a function of experience in a university
setting. The traits were scaled in two dimensions by the
freshmen, while for the seniors, the adjectives were
scaled in three dimensions. Axes for the solutions were
determined by unidimensional judgments provided by a
separate group of subjects. The differences between the
two groups of students, freshmen and seniors, were evident
in both the added dimensions of social desirability
within the ■university cultural group, as well as in the
differences between the locations of the traits in the
multidimensional space.
Partington and Jackson (1968) questioned the
adequacy of traditional impression formation models
because of a scaling problem. The stimulus and response
categories may be unrepresentative conditions. In this
particular study, the subjects gave similarity judgments
for all pairs of hypothetical persons, all of whom had
been described by a differing set of adjectives. They
argued that multidimensional scaling indicated that no
13
simple arithmetic model provides a satisfactory approxima
tion to the data.
Other multidimensional scaling studies have used
personality traits and external judgments to demonstrate
that there is a close relationship between multidimensional
structures and other ratings of desirability or preference.
Rosenberg, Nelson and Vivekananthan (1968) utilized a
trait association technique rather than a similarities
technique in obtaining data. The subjects had to complete
a sorting task by placing the traits into categories on
the basis that the categories represented people they knew.
The subjects could use as many categories as necessary,
but they could not put a trait in more than one category.
A measure of disassociation, like a distance measure, was
obtained for each pair of adjectives by obtaining a dis
agreement score between the number of subjects who assigned
traits to different persons. The resulting three dimensions
from the scaling analysis were labeled good-bad, hard-soft
and active-passive. In addition to finding that the
adjectives used in impression formation studies do fall in
more than one dimension, they obtained additional ratings
on each of the traits with respect to the three different
properties or dimensions. Then multiple regression
techniques were used to locate the axis in the trait space
which corresponded to each of the rated properties. This
provided quantitative estimates of the degree to which
each external rating does interpret a dimension in the
trait space. Rosenberg et. al. confirmed the early
findings of Asch. They found six traits common to both of
Asch's lists to define the dimensions. Using some of the
adjectives in a subsequent study and scaling analysis* a
two-dimensional configuration was determined. The scaling
showed that a trait is influential in a particular list if
it brings another dimension into consideration.
Rosenberg and Sedlak (1972) essentially used the
same technique of a disassociation measure in a multi
dimensional scaling of traits* but they employed it with
trait adjectives which were generated by subjects them
selves. The reason for this procedure was that preselected
adjectives cannot eliminate possible biases in the initial
selection of the trait adjectives. The results indicated
that the configuration was not comparable to earlier
studies. It was unclear as to the reason for the differ
ence between the previous study and more naturalistic
stimulus materials.
Multidimensional scaling studies using different
references. The following studies relate directly to the
present investigation. They are concerned with the multi
dimensional scaling of' trait adjectives when they are
applied in different context conditions. They also show
some quantitative relationship between the multidimensional
configuration and some external rating or judgment.
15
Cliff and Young (1968) applied multidimensional
scaling to trait-descriptive adjectives. With a basic
viewpoint that a person has an internal organization of
external stimuli, this configuration may be represented by
a multidimensional space. Although this representation
may vary under certain circumstances, these representations
should be meaningfully related to each other and to other
unidimensional ratings. Similarity judgments were made on
all pairs of the adjectives under a standard context of a
"ship's captain." Favorableness ratings on single
adjectives were made under "ship's captain" and "tech
nician" contexts. To show that these were simply related
to the similarities data, multiple regression techniques
were applied with the dimensions as predictors. The two
favorableness vectors only differed by a rotation of the
vector's direction.
Also in the area of impression formation,
specifically conjoint measurement and multidimensional
scaling, Cliff (1972) studied the applicability of the
general additive model. Subjects made favorableness rat
ings on pairs of adjectives, as well as on single ones
when applied to two different reference persons, an
"executive" and a "skilled technician." Using multiple
regression techniques to predict the mean ratings of
combinations from the projections of two adjectives on two
dimensions, the results indicated that these ratings were
16
very consistent. This study extended and verified earlier
work by Cliff and Young (1968) in that simple, quantitative
relations between the stimuli in a multidimensional space
and other judgments do exist. This study also extended the
relationship of a single adjective to a particular judgment
to pairs of adjectives.
Hanno and Jones (1973) attempted to test the
generality of multidimensional solutions using two differ
ent stimulus persons conditions, "family doctor" and
"nationally-known politician." The subjects judged the
pairs of descriptive adjectives according to the likelihood
that both traits might occur simultaneously in the same
person. Unidimensional ratings were also obtained for both
likability and favorability for each profession for each of
the traits.- Hanno and Jones claim that the resulting
multidimensional scaling configurations in two dimensions
for the two reference persons were different. Also, when
the vectors were obtained from the multiple regression
analysis relating the "doctor" and "politician,!' the
direction of the two vectors relating favorableness of the
two references were markedly different. The favorableness
vector in the "politician" configuration was close to the
likableness vector for that reference person, but the
favorableness vector for the "doctor" was removed from its
corresponding likableness rating. Hanno and Jones (1973)
concluded that the reference person does have an influence
17
on the multidimensional structure obtained. It appeared to
these authors that the internal cognitive organization of
the subject changed by changing the reference person.
This was also evident in the unidimensional judgments.
The main difficulty with this study seemed to be
with the evidence that the two configurations, "doctor"
and "politician" are substantially different. It would
seem that a transformation might be found which would
relate the "politician" space to the "doctor" space in a
simple quantitative manner. In any event, this would be
an appropriate followup study.
The above studies by Cliff and Young; Cliff; and
Hanno and Jones all are applications of multidimensional
scaling to trait adjectives in different contexts. Cliff
and Young showed that the unidimensional judgments are
quantitatively related to the multidimensional judgments,
when the unidimensional ratings were obtained under two
different reference contexts. Hanno and Jones’ results
indicated that the configurations under two reference
persons conditions were different* with the likableness
unidimensional rating also being different under to two
conditions. The basic distinction between these two
studies is whether the multidimensional configurations do
change under different references, or whether it may be
demonstrated that the configurations do, in fact, represent
the same organizational structure, with apparent changes
18
simply reflecting systematic transformations of basically
the same space. It is this latter question which the
present study will investigate.
Purpose
The purpose of this present investigation was to
extend the work on scaling trait adjectives using different
reference persons conditions. Since the previous studies
used references which might he considered neutral or
positive, this study was to extend the work scaling trait
adjectives using both positive and negative references
under which the adjectives would be multidimensionally
scaled. The major question was "what are the differences,
if any, between the obtained configurations from the
different positive and negative reference conditions".
There were several secondary purposes of this
investigation. Preliminary studies showed very little
effect of using different reference persons. Therefore,
one purpose included manipulating the salience of the
reference person through the use of two different adjective
presentation methods. A condition which required greater
attention to the task on the part of the subject was
included in order to enhance the possibility of obtaining
differences, if any, between the references on the adjec
tive scalings. Another purpose was to clarify the contra
dictory results between the Hanno and Jones study and the
Cliff and Young study by replicating the Hanno and Jones
study. In addition, this present study was to provide
additional support for the validity of the multidimensional
scaling structures through the use of unidimensional
vectors and comparing them to the multidimensional
structures.
Approach. For the major question of this study,
similarity data were obtained for the adjectives under two
positive reference conditions and under two negative
reference conditions in order to determine if there is a
difference in the four configurations from the analysis.
An attempt was made to find a transformation of some of the
configurations in the hope of finding that the configura
tions were indeed basically the same. Two forms of stimu
lus presentations were used, an ordered reference condi
tion where all the similarity judgments for one reference
person were presented together, and a mixed reference
condition, in which the similarity judgments were obtained
for randomly ordered reference persons. In the latter
condition greater attention was necessary on the part of
the subject due to the changing reference person. In
addition, the Hanno and Jones (1973) study was replicated,
using a greater number of subjects. In order to compare
unidimensional vectors to the multidimensional structures,
favorability ratings were obtained on each of the adjec
tives separately in all conditions.
20
There are three main hypotheses for both presenta
tion methods of the main study, as well as the Hanno and
Jones replication. Firstly, when the multidimensional
scaling solutions are compared for the different reference
persons, they will be indeed similar. Secondly, when
systematic changes in the shapes of the multidimensional
scaling configurations are explored, these changes will be
meaningfully related to the specific reference person, and
thirdly, the unidimensional vectors will support the
validity of the multidimensional structures by being
meaningfully related to these structures.
CHAPTER II
METHOD
Construction of Stimulus Materials
Selection of stimuli. The trait adjectives used in
this study were selected by an elimination procedure so
that the remaining set would possess several specific
characteristics. Selection was made from the lists of
adjectives used in several prior studies of scaling or
impression formation, as well as in a pilot study
(Cliff & Young, 1968; Gollob, 1968; Rosenberg, Nelson &
Vivekananthan, 1968; Walters & Jackson, 1966). Twelve
adjectives were finally selected from these lists and are
presented in Table 1. They were chosen so that they would
represent approximately three dimensions and would seem to
fall around clusters of capable, inadequate, bold, timid,
kind and cruel.
Since the subjects were going to make similarity
judgments between these adjectives traits when they were
used to describe a particular kind of a person, it was
necessary that all combinations make reasonable sense.
For example, using extremely negative adjectives with a
positive stimulus person would make the judgment very
21
TABLE 1
Trait Adjectives Used as Stimuli
Pour References: Astronaut, Hypocrite, Killer, Surgeon
1. Bold
2. Kind
3.
Inadequate
4. Ruthless
5.
Aggressive
6. Intelligent
7.
Cautious
8. Cruel
9-
Ineffective
10. Sympathetic
11. Timid
12. Capable
23
difficult for the subject.
The four reference person conditions chosen were
"astronaut", "hypocrite", "killer" and "surgeon!,’ V They
were selected so that two of them were very positive
persons and two were very negative. Prior exploratory
work had shown that when using less polarized references,
there was very little difference in the scaling of the
adjectives with respect to the four reference conditions.
Four references and 12 adjectives were settled upon, with
each adjective being paired with every other adjective
(n(n-l)/2 pairs), yielding 66 pairs of judgments under
each reference condition. This total of 264 judgments
plus a number of repeated items appeared to be about the
maximum number of judgments a subject could reasonably
complete during a one hour session.
Ordered references booklet. Two distinct types .of
booklets were constructed. The first organized the pairs
of judgments for each reference person ("astronaut," etc.)
separately. In this way all the pairs of adjectives
describing one reference condition were presented first,
then all the judgments of another reference person were
presented, with the third and fourth following in
succession.
The adjectives were first put in random order,
then each of the possible pairs was randomly positioned
with the restriction that an adjective may only be repeated
24
no closer than three items away. The within-pair adjective
presentation was also randomized to balance out order
effects. Nineteen stimulus pairs seemed to fit well on a
page. Pour pages were needed to present all 66 pairs,
with room left for 10 additional pairs on the fourth page.
This provided a convenient number of items to be repeated.
Ten pairs of trait adjectives were then selected at
random from the preceding presentations and were repeated
as the last ten items for a particular reference condition.
The within-pair order was reversed from its initial pre
sentation. In this way, there were 40 repeated items all
together, 10 for each reference person. The similarity
judgments were presented on the first 16 pages of the
booklet.
Some examples of items as presented to the sub
jects in the ordered references condition follow. In this
condition four continuous pages of the to-be-judged pairs
for a single reference person were presented, the
reference person being repeated for each item,
kind astronaut ruthless astronaut
123456789
sympathetic astronaut cautious astronaut
123456789
intelligent astronaut ruthless astronaut
123456789
bold astronaut cruel astronaut 123456789
25
In addition to these similarity judgments, favor-
ability judgments were obtained under each of the reference
person conditions. The adjectives were presented singly,
the 12 adjectives being presented in random order on a
separate sheet of paper for each reference person. The
favorability judgments followed the similarity judgments,
delegating them to the last four pages of the booklet. In
all presentation modes, the references under which these
unidimensional judgments were made were in the same order
as they had been in the similarity judgment section. For
each booklet, then, there was a total of 20 pages.
The order effect of presenting all the pairs of
trait adjectives under a particular reference person
condition was balanced by compiling four kinds of booklets,
whereby the first presented reference condition was
different for each of these groups of booklets.
Mixed references booklet. The second type of
booklet differed from the first primarily in the order of
presentation of the reference conditions. Since the
subject might adapt to the particular reference person and
not make his judgments with respect to that particular
stimulus person when all the stimulus pairs referring to
one person were presented together, a booklet with mixed
references was constructed. In this condition, 264 pairs
of adjectives (four references times 66 pairs) were
randomized over 16 pages of the booklet, with the same
26
criterion for the repetition of adjectives as before. Each
pair of adjectives was preceded by the specific reference
person, which changed between the pairs. The reference
remained the same for both adjectives of the pair to be
judged. The last 10 positions of every fourth page were
used for repeated items. The four reference persons were
then assigned randomly over the four presentations for each
stimulus pair.
Some examples of the to-be-judged pairs in the
mixed reference condition follow. The difference between
the ordered and mixed conditions is the changing reference
person for each pair of adjectives in the mixed references
condition, while the reference person remains constant in
the ordered condition until all pairs have been presented,
cruel hypocrite sympathetic hypocrite
123^56789
bold surgeon aggressive surgeon 123^56789
timid astronaut capable astronaut 123^56789
ruthless killer intelligent killer 123^56789
Only one type of mixed references booklet was
constructed for this condition. Repeated items were again
randomly selected from the preceding 66 pairs, with the
stipulation that the 40 pairs consist of 10 pairs of
traits under each of the reference conditions.
Subjects
Students from the introductory psychology courses
27
at the University of Southern California volunteered to
participate in this particular experiment in order to
partially fulfill a course requirement. Complete data were
obtained from a total of 100 subjects. Each subject was
randomly assigned to one of seven conditions (four
ordered reference conditions, one mixed reference condition
and two orders of the Hanno and Jones replication study).
The random assignment continued until each condition was
filled with the number of subjects stipulated earlier.
Forty subjects each completed the ordered reference
booklets, ten for each of the four orders. Using the
mixed reference booklet, 40 subjects responded with
complete data. In the Hanno and Jones replication study,
20 subjects participated, with 10 being randomly assigned
to one of two orders.
Administration
The booklets were administered individually during
a six-week interval. A subject was able to fill out a
booklet at his convenience, since there were many times
made available for this during the week. Some subjects
elected to complete the task at the same time as others,
but they each worked in a private experimental room.
The cover pages of the booklet contained the
instructions for the task, as well as examples and
practice trials. The directions for the ordered references
28 ■
condition, as well as for mixed references, (in the
appendix) stated that the subject was to make a Judgment
of the similarity between two adjectives which are used to
describe the same kind of person. The subject's task was
to think of the kind of person the first adjective
describes and then think of the kind of person the second
adjective describes. His Judgment was how similar these
people would be using a scale from one to nine. One
example was "kind friend versus considerate friend".
The subject read the instructions on his own and
then rated the eight practice pairs, which involved all
12 adjectives, and all four references. Then the subject
was given the opportunity to ask any questions regarding
the task. Very few subjects did not understand the
directions initially. The subjects were then able to work
through the booklet at their own speed, recording their
ratings to the right of each pair in the booklet. The
subjects were not required to identify themselves on the
booklet and their responses were kept anonymous. Most
subjects finished the task within an hour.
Constructing Hanno and Jones Replication
i t .
The two booklets for this replication were
constructed by following the method section in the Hanno
and Jones (1973) study. Sixteen adjectives were used to
make 120 pairs of trait adjectives. The adjectives,
29
presented in Table 2, were supposed to fall into four
clusters of four words each around the stimulus words
calm3 enthusiastic, reliable and unreliable. The two
reference conditions under which subjects were to rate the
pairs of adjectives were "family doctor" and "nationally-
known politician."
All 120 possible pairs were formed and presented in
random order, with an attempt to counterbalance the number
of times a stimulus was presented on the right or left.
Adjectives were distributed fairly evenly over the entire
list. Twenty-four pairs were chosen at random for a
reliability measure and these were presented on the last
page of the booklet. The booklet was made up of eight
pages of 18 items each for each reference condition. In
addition, the sixteen stimuli were randomly presented
individually and the subjects were to rate them on a nine-
point scale under two different kinds of directions. The
first unidimensional judgment was of how favorable the
adjective was in relation to the occupation of the
reference person, while the second judgment was of how
likable the stimulus person would be as a person when
described by that adjective.
The two booklets for this study differed only in
which reference person, "family doctor" or "politician,"
was presented first. In both booklets, all three types
of judgments were first made for one stimulus person, then
TABLE 2
Trait Adjectives in Hanno and Jones Replication
Two References: Family Doctor* Politician
1. Cautious 9.
Self-disciplined
2. Nonconforming 10. Aggressive
3.
Level-headed 11. Quick-witted
4. Inventive 12. Methodical
5.
Responsible
13.
Competitive
6. Unreliable 14. Reliable
7.
Enthusiastic
15.
Consistent
8. Calm 16. Unpredictable
31
for the other. In other words, the reference remained
constant until all the items had heen presented. A few
examples of the items would he:
family doctor: calm aggressive 123^56789
family doctor: quick-witted cautious 123^56789
family doctor: unreliable consistent 123^56789
The administration of the "doctor-politician"
replication was essentially the same as the four refer
ences conditions. Twenty subjects participated, with 10
being randomly assigned to each order. The instructions
(in Appendix) stated that the subject's task was to rate
the similarity between two adjectives on a scale from one
to nine while considering the similarity of the two
adjectives as being equivalent to the likelihood that
both of them apply to either a "doctor" or a "politician,"
depending upon which booklet the subject was assigned.
Practice trials utilizing all the adjectives were also
included and the subjects were provided with an opportunity
to ask questions concerning the procedure. These subjects
also apparently had little difficulty with the directions.
They finished in approximately three-fourths of an hour.
Data Analyses
The data collected in the main study consisted of
similarity judgments for 80 subjects, all of whom had
rated 66 pairs of adjectives under four reference condi
32
tions with 40 repeated items. A total of 48 favorability
judgments were also obtained.
The basic analyses carried out with these data
concerned basically separate, but parallel analyses on
each stimulus presentation methods. The analyses
concerned reliability judgments, group differences in
similarity judgments, multidimensional scaling analyses
relating groups of subjects and reference conditions, and
relating favorability judgments to the multidimensional
scaling results.
The data from repeated items were used to obtain
information regarding the consistency of subjects. This
would give a general picture as to the stability of the
data.
Secondly, analyses of variance were carried out,
testing for differences between the presentation orders of
the ordered references condition on the original
similarity judgments. If there were no significant
differences, this would result in determining whether the
data in the four subgroups of the ordered references
condition could be collapsed into one group with 40
subjects, making it comparable to the mixed references
condition with 40 subjects. An analysis of variance
procedure on the data from the ordered references condi
tion and the mixed references condition was used to show
whether there were any differences due to the specific
33 |
procedures used for each subgroup.
It was assumed that the similarity ratings of four
different orders of booklets used in the ordered refer
ences condition would not be significantly different from
each other in each of the four references taken separately.
Multiple within-subjects analyses of variance (subjects by
ordered references condition) were carried out separately
for the ratings on the adjective pairs for "astronaut*"
"hypocrite," "killer" and "surgeon." The results of these
analyses are presented in Table A1 (in the Appendix). As
is shown in the table there were only three significant
F-ratios out of 66 for "astronaut*" five significant
ratios for "hypocrite," four for "killer" and three
significant P-ratios for "surgeon." Since there were
fewer significant ratios than one would expect by chance,
there seemed to be no significant differences between the
similarity ratings due to order effects of the presenta
tions of the ordered references. The similarity data were
collapsed over these orders and mean similarity judgments
were computed for 40 subjects, all further analyses were
carried out using these collapsed mean similarity judgments
These similarity ratings were then compared more system
atically with the similarity ratings for the 40 subjects
in the mixed references condition.
The major analysis involved a multidimensional
scaling analysis of the similarity ratings for these
3^
resulting groups. Analyses were performed separately for
each reference condition using TORSCA (Young and
Torgerson, 1967)5 applying it in both a metric and non
metric manner. This program is based on a rationale
proposed by Torgerson (1952) and Shepard (1962a, 1962b)
and incorporates Kruskal’s (1964) stress index. This
resulted in eight sets of dimension loadings for the 12
adjectives in the main study.
These eight sets were then compared (in groups of
4) to determine the effects of the reference person upon
the configurations. First the configuration for one
reference person was treated as a target matrix and then
the other matrices were orthogonally rotated in order to
determine the best least-squares fit to it (Cliff, 1966).
Thus, stimulus configurations are made as similar as
possible without changing the shape of the space (Cliff,
1965; Schonemann and Carroll, 1970). Then, systematic
changes in the space were explored by using Procrustean
techniques (Schonemann, 1966). Each of the rotated
dimensions was predicted from corresponding original
dimensions of a different reference in order to determine
how close the rotated configurations matched the original
ones, with the original dimensions now free to be both
rotated and differentially stretched. Following this,
the regression weights (stretching or shrinking factors)
associated with the unrotated matrix of one reference
35
condition were used to transform an original set of ;
dimensions for a different reference person. The shape of '
the configuration of the rotated matrix in the first case
and the shape of this transformed matrix were compared.
Thus, the dimensions of the transformed configuration were
stretched or shrunk in the same way as a different matrix
was in the regression analysis and the congruence of these
configurations may he compared.
Additional support for the congruence of the
obtained spatial representations included the correlations
between the dimensions, using the projections of the
stimuli (adjectives) on the dimensions as data in order to
determine relationships between dimensions of different
reference persons.
A supplementary analysis concerned the differences,
if any, between the ordered references data and the mixed
references data. Analyses of variance on the difference
between groups on the original similarity data were
performed to yield perhaps a general idea on the differ
ence between ratings on the individual pairs.
Another important aspect of the data analysis was
the relationship of the unidimensional favorability ratings
or the relationship with the respective multidimensional
spaces. The favorability judgments were used as dependent
measures in a multiple regression analysis with the
stimulus coordinates for the three dimensions as the
36
independent variables for each of the reference conditions. 1
The direction of the favorability vectors were also
plotted in the multidimensional space, relating these
vectors to each other. These analyses and their inter
pretations follow in the next chapter.
Analyses for Hanno and Jones Replication
Twenty subjects gave 240 similarity Judgments,
with 48 repeated judgments. Thirty-two favorability and
32 likability judgments were part of this replication.
The data analyses were, in general, parallel to the
analyses on the four references conditions. The basic
analysis was a multidimensional scaling analysis on the
"doctor" and "politician" separately, utilizing the
rotational procedure, as well as the methods of finding
systematic changes in the space.
Additional analyses using the judgments between
the two orders of booklets in "doctor-politician" were
also used to show differences, if any, in the two groups
due to order effects. Similar within-subjects analyses of
variance were carried out for the two orders of presenta
tion for the "family-doctor" and the "politician." The
results from the 120 analyses of variance for each of the
two references are shown in Table A2 (in the Appendix).
There were four significant P-ratios (£< .05) for the
"doctor" condition and eight significant P-ratios for the
"politician." Since the total analyses were also non
significant, overall, the similarity data for "doctor"
were collapsed over the two presentations. The same was
done for "politician." Mean similarity data for each of
these references were then used in all additional analyses
CHAPTER III
RESULTS
Repeated Measurements
There were 40 repeated items in the four reference
persons conditions, 10 under each reference person. The
test-retest correlation coefficients were obtained over
the 40 items for each individual subject. They are
presented in Table A3 (in the Appendix), along with other
descriptive information. The subjects are arranged in the
conditions in which they served. The correlations for
reliability were, for the most part, fairly high. The
median reliability for all 80 subjects was .70 with a
range of .97*
Multidimensional Scaling Analyses
The eight different similarity ratings for all 66
pairs, four references and two kinds of stimulus presen
tations, were used each as input data to TORSCA. Both
metric and nonmetric solutions were computed in two to four
dimensions.
The first major question concerns the correct di
mensionality for the configurations. Kruskal's (1964)
stress for each of 2, 3 and 4 dimensions for each of the
38................... '
39
configurations is presented in Table 3. Kruskal claims
that a stress of 5$ is considered a "good" fit from the
data, while 10$ is considered a "fair" approximation. As
the table illustrates, there is a substantial difference
in the stress values between the metric and nonmetric
solutions, with the nonmetric solution yielding much
lower stress values. This would seem reasonable when
considering how a nonmetric solution would change the
points in the space with iterations. Since the stress
values from the nonmetric solutions were substantially
lower than those from the metric solutions, all further
analyses and discussion employed the nonmetric solutions.
First considering the four references conditions, in three
dimensions, the nonmetric analysis yielded stress values
of .05 or better. Increasing the dimensions to four did
not substantially decrease the stress values. There was a
greater drop in the stress values, however, from two to
three dimensions, than from three to four dimensions. The
number of dimensions would seem to be three, depending
upon the psychological interpretability of the adjectives
with respect to the dimensions. Table 4 indicates that
the size of the projections of the stimuli on the third
dimension were somewhat smaller than those in two
dimensions. However, there are several stimuli with pro
jections of .30 or greater.
The two-dimensional representations under the
TABLE 3
Stress Values
Group 2
Ordered Astronaut .19
Ordered Hypocrite .29
Ordered Killer .29
Ordered Surgeon .19
Mixed Astronaut .15
Mixed Hypocrite .16
Mixed Killer .14
Mixed Surgeon .19
Dimensionality
Metric Nonmetric
3
4 2
3
4
.14
.13
.08 .03
.02
.19
t
H
CO
.14
.05 .03
.19 .19 • 09
.06 m
o
.15
.14 .07 .02 .02
• 09 .07 .11
CO
o
•
.02
.11 .10
.09
.04 .03
.11 .10 .12 .04 .02
•
H
r o
.14
.09
.04 .03
TABLE 4
Coordinate Matrices
Ordered References
Astronaut Hypocrite Killer Surgeon
Stim. 1 2
3
1 2
3
1 2
3
1 2
3
1
-17
-46
-33 -51
-10
-29
-38 -34 -32 -18
-47 -32
2 62 -o4 -26 61
-19 -29 59
-16
05 58
03
-28
3
-18
65
-21 21 56 00
05
64
-33 -37 56 -26
4 -72 -01 01 -62 04
03 -67
-06 -14
-70
-19 17
5 -32 -43 -33
-54 -04 -21 -50
-29
-20 -26 -44 -40
6 22
-39 29 05 -37 36 12 -46
31 30 -36 31
7 29
00 52 20 -04 56 34
13
46
49 13 25
8 -76 00
15
-61 10 11 -70
17
16 -70 -06
33
9 -23 59
-26 24 52 -21 14
59 -05 -32
57
-28
10 66 01
-19 55
-l4 -29 60
-07
-18 56 09 -26
11 38 56
37 58 24 24 46 4o 10 24
57 45
12 22 -48
25
-18
-59
-02 -05 -56
15
34 -44
29
TABLE 4— (Continued)
Mixed References
Stim. 1
Astronaut
2
3
Hypocrite
1 2
3
1
Killer
2
3
1
Surgeon
2
1
-35
-26 4o -64 -07
35 -58 01
15 07
-48
2 52 -04 38 51 -38 29 58 -11 48 46 30
3 -08 67 05
46 52 24
31 58
09
-6l 36
4 -68
03 -25 -54 23
-25
-62 10 -01 -52 -42
5 -43 -19 44 -66 10
05 -59
01
15
-01
-55
6 14 -54
-17
-20 -52
03 -09
-46
-19
48 -24
7
28 -24
-33 38 -21
-39 17 -17 -55
41
29
8 -60 08
-43 -52 35 -18
-63
10 -01
-57 -38
9
-10
67
06 32 58 01
33 49
02 -6o 34
10 52 04
47 51 -13 35 62
-13 33 37 39
11 62
33
-44 66 01
-32 68 12 -42
03
66
12 16 -54 -16 -27 -47
-19 -19
' -54
-03 51 -27
32
25
08
■16
36
-22
-40
-21
11
32
■32
-12
Note.— Decimal points have been omitted.
ro
43 ;
different reference conditions seem to be the most inter
pretable. The first dimension seems to be defined by
"ruthless" and "cruel" on the negative end and "sympathet
ic" and "kind" at the positive end. The second dimension
seems to be represented by "capable" and "intelligent"
with the adjectives "inadequate" and "ineffective" at the
opposite end. More importantly, however, the adjectives
tended to fall in clusters in a circular fashion as
predicted for each reference condition. The clusters in
general seemed to be "bold-aggressive," "ruthless-cruel,"
"inadequate-ineffective," "capable-intelligent" and "kind-
sympathetic" for essentially all conditions. "Timid" and
"cautious" did not seem to form a cluster as did the
others. Two dimensions seemed to represent the data fairly
well. However, three dimensions were used in further
analyses for these data in the four reference conditions
because the stress values did appear to drop significantly
from two to three dimensions and because the earlier study
by Cliff and Young (1968) similar to this study used three
dimensions. It might be possible then to make some
comparisons between these results.
Rotation. The primary purpose of this study was
to compare the configurations for each of the reference
conditions. On first consideration it appeared that there
were some differences among the configurations for the four
reference conditions, but since these differences might be
solely due to rotational differences, which are irrelevant
to multidimensional scaling, Cliff's (1966) orthogonal
least-squares Procrustes procedure was applied to the data.
The "hypocrite" and "killer" configurations were very
similar without rotation, with the "astronaut" and
"surgeon" stimulus points appearing somewhat different.
It was decided to use an average matrix of stimulus pro
jections derived from the "hypocrite" and "killer" matrices
as the target and rotate the other two reference matrices
to it. Pour separate rotations were made for the stimulus
projections and two modes of stimulus presentation. This
procedure permitted the four references to be compared
with the specific rotation. The matrices from these
rotations are presented in Table 5* along with the
averaged "hypocrite" and "killer" matrix.
In general, it may be seen in Table 5 that the
matrices for each specific reference person seem fairly
consistent within a condition. The following sections
will describe and evaluate these consistencies for the
different configurations.
Dimensions. When the first two dimensions were
plotted against each other for the rotated "astronaut"
and "surgeon" references and the original "hypocrite" and
"killer" references, the configurations seemed qualita^
tively the same. The distances between the stimulus
points for the different references appeared similar. In
45
TABLE 5
Rotated Coordinates
Nonmetric Coordinates Ordered References
Hypocrite/Killer Rotated Astronaut Rotated Surgeon
Stim. 1 2
3
1 2
3
1 2
3
1 -44 -22 -30
-33 -35 -34 -34
-31
-38
2 60
-17
-12 56 -26 -28
55
-26 -21
3 13
60 -16 06 68
-19
00
67 -25
4 -64 -01 -05
-68
25
02
-71
18
07
5 -52 -16 -21 -45 -27 -34
-39
-24 -46
6
09
-41 34
07 -45 28
03 -47 31
7 27 05 51
28
-13 52 45
-14
31
8
-65 13 13 -71 27 17 -67 30 24
9 19 55 -13
-01 64 -24
05
66
-27
10 58 -11 -24 6l -22 -20 56 -21 -18
n 52 32 17
56 38
37
42 36
53
12 -12
-57
08
03 -53 23 03 -55 29
Nonmetric Coordinates Mixed References
Hypocrite/Killer Rotated Astronaut Rotated Surgeon
Stim. 1 2
3
1 2
3
1 2
3
1 -61
-03 25 -43
-10
39 -45 -15
34
2 54 -25 39 45 -29 36 4o -36 28
3 39
54 18
23 63 09
21 68 02
4 -58 16
-13 -59
34 -23 -53
40 -20
5
-62 06 10
-47
-00 44 -54
-07 37
6
-15 -49 -08 -12
-53
-20
-13 -53 -17
7 27
-18
-47
14 -32 -35 37 -37 -37
8
-57
22
-09 -50 36 -41 -50
45 -25
9 33 53
01 21
63 09
20 67 05
10 56 -13
34 48 -22
45
46 -24 34
11 67
06
-37
70 04 -44
65 09 -33
12
-23
-50 -11 -10 -54
-19 -15 -57 -07
46
the ordered references condition, the dimensions seemed
somewhat consistent across all references. In general,
the first dimension was represented by "timid,"
"sympathetic" and "kind" on the positive end of the first
dimension, with "cruel," "ruthless/1 "bold" and
"aggressive" on the negative end. The second dimension
was represented by "intelligent" and "capable" on the
positive end, with "ineffective" and "inadequate" on the
negative end. The labeling of the dimensions was not
readily apparent. Since the clusters and shapes of the
configurations were the most important aspects of the
study, and not the dimensions, a discussion of the
clusters follows.
Adjective clusters. The most important aspect of
the configurations appears to be that the adjectives seemed
to fall in clusters of two or more.adjectives regardless
of whether the configuration represented ordered or mixed
presentations. The solutions yielded configurations
which were fairly circular, with the adjective clusters
clearly displayed.
For the ordered condition, five clusters of
adjectives were apparent. These clusters for the differ
ent references may be seen in Figures 3 and 5 (see pages
62 and 64). It appears that "ruthless" and "cruel" form
one cluster with "aggressive" and "bold" as another
cluster in close proximity, "ineffective" and "inadequate"
47
are at one end of the second dimension with "capable"
and "intelligent" at the other pole, also in a cluster.
"Sympathetic" and "kind" appear together, with high load
ings on the first dimension. Two stimulus adjectives
which did not fit well into this clustering were "timid"
and "cautious." The stimulus points for "timid" lay on the
outer edge of the circle, while the points for "cautious"
were closer to the origin.
The solution for the mixed references gives the
clearest picture of the clusters, with the within-refer-
ence distances being the smallest in this condition. The
clusters are essentially the same as the previous solution,
and may be seen in Figures 4 and 6 (see pages 63 and 65).
As the figures indicate "timid" is now closer to "kind"
and "sympathetic" and its projections for each of the
four references are extremely close. "Cautious" is still
alone, and the stimuli are spaced further apart than any
of the other stimuli for the different references. This
configuration gives the clearest picture of how very close
the different solutions for each reference are.
Congruence of Configurations
Correlations of dimensions. One method of
describing the relationships between the dimensions, the
similarities and dissimilarities, was to compute the
correlations between the corresponding dimensions.
48
Correlations of the stimulus projections for the twelve
stimuli for each of the dimensions were carried out for
three dimensions for the four reference conditions,
ordered and mixed presentations. The results are shown in
Table 6. Many of these correlations are very high. The
correlations between the corresponding dimensions of the
four different references are above .93 for the first two
dimensions, and only somewhat lower, down to .81 in one
case, for the third. These very high correlations
between corresponding dimensions indicated that there was
a strong relationship between the dimensions for each
specific reference person. The dimensions are essentially
independent of the target person.
Stretching or shrinking the dimensions. The high
correlations indicated that the dimensions obtained under
the different target conditions match quite well. However,
this does not necessarily mean that the spaces are
identical, since they could have different shapes, i.e.
the dimensions could have different sizes under the
different conditions. The degree to which this was the
case was now tested. Using Procrustes techniques, the
"hypocrite" projections were now permitted to be
differentially stretched or shrunk, as well as freely
rotated, in order to match the projections of the
"astronaut" and "surgeon" data. First, the amount of
difference in the configurations was measured. The sums
TABLE 6
Correlations Between Corresponding Dimensions
for Four References
Astronaut Hypocrite Killer Surgeon
1
--
.96
.99 .99
Astronaut 2
---
• 95 • 94
.99
3
--
.92 .84
• 95
1 .96
—
.98 .94
Hypocrite 2
Co
CO
—
.94 • 96
3
CO
CO
—
.78 .81
l
.99 .98
—
• 99
Killer 2
• 95 • 93
—
.95
3
.81
.83
—
.78
1 .98 Co
00
.98
—
Surgeon 2
.99
.98 .93
—
3 Co
00
.87 .85
— — —
Note.— Correlations for the ordered references condition
are in the upper half of the matrix; correlations
for the mixed references condition are in the
lower half.
50
of squared differences between the "hypocrite" and the
rotated "astronaut," and between it and the rotated
"surgeon" projections (i.e. in Table 5) were computed as
a measure of difference between the spaces obtained from
the different reference conditions.
Then, the rotated dimensions for "astronaut" and
"surgeon" were used in turn as the targets of an unre
stricted Procrustes transformation (a multiple regression
program was used to perform the calculations).
"Hypocrite" was chosen as being representative of the two
negative reference conditions. The multiple correlation
coefficients, as well as the regression weights, the
stretching or shrinking factors, are given in Table 7.
As the table indicates, the multiple correlations are .98
or greater for the first two dimensions, dropping to no
less than .85 for the third dimension. These multiple
correlations are, however, only slightly higher than the
ones between single corresponding dimensions.
The weights obtained in the Procrustes analysis
may be thought of as elements of a 3 x 3 transformation
matrix, which were then applied to the original "hypocrite"
dimension loadings in order to determine how the differ
ential stretching and shrinking of the axes affect the
shape of the configuration. The transformation matrix
elements may be seen in Table 7- The elements indicate
how the axes have changed from reference person to
Multiple R
Transfor- 1
mation
matrix by 2
dimensions
3
TABLE 7
Procrustes Transformations for "Astronaut"
and "Surgeon"
Predicted Dimensions
Ordered Mixed
astronaut surgeon astronaut surgeon
1 2
3
1 2
3
1 2
3
1 2
3
Co
• 95 .9^ • 97 • 96 .85 .98 Co
00
.88
• 99 .98 .89
• 97
-.02
.05 .92 -.02
.09
.81
• 03
1
•
0
.83 .03 -.09
OJ
on
•
i
1.22 -.18 -.32 1.23 -.23 -.15
1.10 .00 -.16 1.20 .02
i
•
o
r o
.15 .99 .03
.10
• 93
.10 •
0
LO
1.15
1
•
0
cr*
-.04
.92
ui
H
52
reference person. For corresponding dimensions, the
specific elements across the "astronaut" and "surgeon"
references and across conditions are very comparable. The
results for the first dimension (.97» -92, .81 and .83)
indicates that the corresponding dimension for "hypocrite"
would be slightly diminished by the transformation,
while the second dimension (1.22, 1.23, 1.10 and 1.20)
may be seen as being stretched or elongated. The effect
on the third dimension is less consistent. The presence
of appreciable off-diagonal terms (e.g. -.32) indicates
that some of the change is oblique to the axes in their
current positions.
An important question is whether the unrestricted
transformations result in an appreciably better fit than
that provided by the orthogonal one. Does the orthogonal
rotation provide about as good a fit to the original
configuration, or does the differential stretching or
shrinking of the axes allow the points to be moved around
so that they correspond more closely to the original
configuration? If the points in the configurations after
orthogonal rotation are only random displacements, then
an attempt to move these points and change the shape of
the space would have little effect on the fit of the
configurations (Cliff, 1965; Schonemann and Carroll,
1970). However, if the unrestricted transformation of one
set of dimension loadings corresponds well to the
53
dimension loadings from another reference person, this is
also evidence that the unrestricted transformation
provides a better fit.
In order to test for the fit of these two
different procedures, a variance ratio test was devised.
This was to test the difference between the orthogonal fit
and the fit of the matrix with a changed shape. The
difference used was between the sum of squared deviations
of the fit of the "hypocrite" matrix and the orthogonally
rotated "astronaut" or "surgeon" matrix, and the sum of
squared deviations between the fit of the "hypocrite"
matrix, transformed by the unrestricted transformation,
to the "astronaut" or "surgeon" when the "hypocrite"
points were allowed to move freely in the space. Table 8
gives these sums of squares of deviations. It can be seen
there that the points are moved appreciably closer
together by the unrestricted transformation than they
were by the orthogonal transformation. Since some
improvement must necessarily take place, the question of
statistical significance arises.
The degrees of freedom associated with the fit of
the entire matrix was 36 (three dimensions by 12 stimuli).
Subtracting three degrees of freedom for the requirement
that the projections on each dimension sum to zero and
three degrees of freedom for the orientation and overall
stretching factor for the orthogonal rotation yields 30
ordered astronaut
ordered surgeon
mixed astronaut
mixed surgeon
TABLE 8
Fit of Transformed References
Sums of Squared Deviations
orthogonal
.8955
1.2250
.6286
.5368
unrestricted
.3674
.6014
.4470
.316.0
difference
.5281
.6236
.1816
.2208
F
34.52
24.84
9.76
16.73
All F-tests are significant p^. .01, df = 33 24
u i
55 i
degrees of freedom for an orthogonal fit of one matrix to
the other. For the fit of the matrix with a changed
shape, 24 degrees of freedom were used because nine
parameters, regression weights, were estimated in the
regression analyses.
Table 8 shows the results of this ratio test. In
the first column are the sum of squared deviations of
the specific reference condition, when rotated orthogonally,
from the original "hypocrite" projections on three
dimensions. This, then, represents a measure of how well
the orthogonal rotation fits the original data. The
second column gives these deviations for an unrestricted
match, the case where the shape of the configuration is
free to vary. The sum of squared deviations was obtained
from the difference between the transformed "hypocrite"
matrix of stimulus projections and the orthogonally
rotated "astronaut," when the regression coefficients
were used as a 3 x 3 transformation matrix. The differ
ence between these two measures of fit is presented, as
well as the F-ratios for this difference divided by the
error mean square from the unrestricted match.
Table 8 indicates that the F-ratios in all condi
tions are extremely significant, indicating that the
unrestricted match greatly improves the fit of the con
figurations over the orthogonal rotation. Thus, the
targets change the shape of the spaces.
56
In order to show how the shapes of the configura
tions were changed, idealized, smoothed out configurations
were used. In Figures 1 and 2 the original "hypocrite"
plot of the first two dimensions is presented as a
smoothed-out ellipse. The lengths of the two original
axes were used to plot an ideal version of the "hypocrite"
configuration. A smoothed line was used to draw the
elliptical shape of the configuration, although this line
does not correspond specifically to the stimulus points
involved.
The first dimension is definitely more dominant
than the second. The second plot illustrates an
idealized transformed "hypocrite" space. Using the points
circumscribed by the ellipse in the first case, they were
transformed by the 2 x 2 matrix of weights (for the first
two dimensions) from the regression analysis appropriate
to the reference person and condition. The differences
between these two plots illustrates how the "hypocrite"
configuration was changed. The transformed plot
indicates that, while the first dimension did not change
very much in overall length or orientation, the second
dimension became more prominent in relationship to the
first. The ellipse has been changed also by rotation of
the second dimension obliquely to the first dimension.
The shape of the configuration is now rotated and is
squeezed so that there is a better fit of the configura-
II
57
Hypocrlte
Transformed
Hypocrite
(Astronaut stimulus points)
Transformed
Hypocrite
(surgeon stimulus points)
Fig. I. Transformations on Idealized hypocrite, ordered condition,
56
Hypocrl +e
Transformed N.
Hypocrite
(Astronaut stimulus points)
II
Transformed N
Hypocrlte
(Surgeon stimulus points)
Fig. 2. Transformations on Idealized hypocrite, mixed condition.
f
I
........... 59
tion. This change is also evident in the other plots of
ordered and mixed conditions. It may he seen that the
plots of the mixed condition are very similar to the
ordered condition. The second dimension has become more
dominant in these plots, as well. The specific stimulus
points from the "astronaut" and "surgeon" references are
also presented in these plots. As may be seen, these
stimulus points correspond well to the changed ellipse.
In general, the elliptical shapes of the spaces have
changed so that the second axis, defined by "capable" and
"inadequate," is as dominant as the first and is
obliquely rotated by a small degree.
Unidimensional Favorability Ratings
Average favorability ratings were obtained for
each trait adjective for each reference person under both
ordered and mixed conditions. These are given in Table A4
(in the Appendix). The main purpose of obtaining these
ratings was to determine how a unidimensional favorability
vector would map into the multidimensional space of trait
adjectives and how these vectors might change with
respect to the reference person or presentation condition.
If the favorability ratings were considered to be
weighted combinations of the coordinates of the multi
dimensional scaling analysis, then these ratings might be
plotted in the multidimensional space in a meaningful way
60 :
with the stimulus points. A standard multiple regression
analysis was used in order to determine these weights.
Each of eight sets of favorability ratings, four refer
ences and ordered and mixed conditions was used in turn
as the dependent variables in a multiple regression
analysis. The corresponding dimension loadings for three
dimensions were used as independent variables. The
projections used were the original matrices for
"hypocrite" and "killer" and the rotated matrices for
"astronaut" and "surgeon." The resulting multiple correla
tion coefficients and the weights for the three dimensions
are given in Table 9* The correlations are high, ranging
from .88 to .98. In all but one reference condition, the
weights for the third dimension are substantially lower
than for the second dimension. Some systematic differ
ences may be noted. The second dimension is relatively
more important in the positive reference conditions of
"astronaut" and "surgeon," while in the negative reference
conditions, the weights are more nearly equal for the
first two dimensions. Between the ordered and mixed
conditions, there appears to be a difference in the
weights for the first two dimensions, with the weights
being more negative in the latter condition.
These results are also presented graphically in
Figures 3 through 6. These figures indicate the projec
tions of points on the first two dimensions from the eight
6l !
i
TABLE 9
Prediction of Favorableness Ratings for
Three Dimensions
Ordered References
Astronaut Hypocrite Killer Surgeon
Multiple R • 96 .89 .94 .97
Relative 1. .21 -1.05 -1.90 2.65
Weights 2.
-5.^8
-1.30
-1.51 -4.57
3. • 75
.28 2.10
1.73
Mixed References
Astronaut Hypocrite Killer Surgeon
Multiple R
.97
.88 .98 • 98
Relative 1. -1.60 -1.63 -3.33 -.51
Weights 2.
-5.97 -.67 -1.57
-5.81
3.
1.38 .14 -.72 1.55
9 3 1nadequate
62
Fig.
cruel *¥-
aggressive s
Astronaut
a'O sympathetic
vi. capable
V favorability
Inadequate
3. 9
cruel 8
4
II
aggressive 5
Surgeon
sympathetic
capable
, favorabl11ty
3. Location, of points In two dimensions, ordered astronaut and
surgeon with direction of favorableness Indicated.
cru el
in e f f e c t iv e
93
a g g ressiv e 5
10
2
11
kind
Astronaut
(612 capable
in e f f e c t iv e
3
,8 cru el
ag g ressiv e
*2
capable
Surgeon
U
10
1 2 kind
63
tim id
tim id
4. Location of points in two dimensions for mixed astronaut and
surgeon with directions of favorableness Indicated.
3
9 in e f f e c t iv e
11 tim id
cru el 8
a g g ressiv e 5
12 capable
10
'2
kind
H ypocrite
9 in e f f e c t iv e
1 * 1 tim id
cru el 8
a g g ressiv e
10
kind
6
capable
K ille r
Fig. 5. Location of points in two dimensions for ordered hypocrite
and killer with directions of favorableness indicated.
65
g i in e f f e c t iv e i
i 1
3 i
f
c r u e l 8
4
a g g ressiv e -5
12 capable
6
H ypocrite
I
’11 tim id
• 1 0
7
2. kind
3.
9 in e f f e c t iv e
c ru el i *1 W e s s i v e
11 tim id
Z10 kind
capable
K ille r
Fig. 6. Location of points in two dimensions for mixed hypocrite and
killer with directions of favorableness indicated.
66
multidimensional scaling analyses, using the projections
for "astronaut" and "surgeon" after rotation. The lik-
ableness vectors drawn from the origin represent the
directions of maximum favorableness of the ratings with
respect to the stimulus points.
In Figures 3 and 4, the plots of the positive
references are given. In both the ordered and mixed
conditions, the direction of maximum favorableness for
"astronaut" corresponds with "capable" and "intelligent"
trait adjectives. Unfavorableness in an "astronaut"
would correspond to "inadequate" and "ineffective.An
"astronaut," then, is seen as being most favorable when he
is described as being capable. The "surgeon" configura
tions indicate that surgeons are more favorable when they
are described by other adjectives of "sympathetic," "kind"
and "cautious," as well as "capable" and "intelligent."
In the ordered condition, the favorableness vector is
rotated somewhat towards the "sympathetic" group of traits
This may be seen to a lesser extent in the mixed condition
Thus, "surgeons" are most favorable when described by
intelligent, as well as kind traits.
The negative references in Figures 5 and 6
indicate that the favorableness vector has been rotated
towards "bold,1 . 1 "aggressive,"ruthless" and "cruell" The
configurations for both "hypocrite" and "killer" appear
very similar. In the ordered "killer" condition, the
67 :
direction of maximum favorableness is congruent with
"hold" and "aggressive," while in the other configurations
the favorableness vector approaches the locations of these
adjectives. These configurations indicate, then, that a
negative person, a "hypocrite" or "killer" is seen as most
favorable when described by negative traits such as
"aggressive" and "ruthless." In other words, a "killer"
who is a good "killer" is one who is "aggressive" and
"cruel."
Hanno and Jones Replication
Repeated Measures. The correlations for the sub
jects in the two references condition, "family doctor" and
"politician" were based on a total of 48 repeated measures.
The median reliability was .77, with a range of .88. Most
subjects responded fairly reliably. These reliabilities
are in Table A5 (in the Appendix).
Multidimensional scaling analyses. The similarity
data for 120 pairs of judgments under two reference
conditions were used as input to TORSCA, (Young and
Torgerson, 1967) with both metric and nonmetric solutions.
In this replication study the stress values for the metric
solution were very poor, with stress being .20 or larger.
Nonmetric stress values are very comparable to the
results obtained by Hanno and Jones (1973)• In two
dimensions, the stress values are "fair," (.09 for
68 |
"doctor" and .12 for "politician") while in three dimen
sions they are substantially better (.06 for "doctor"
and .08 for "politician"). The projections of the stimuli
on the dimensions indicate that the second dimension is
not very strong. As Table 10 Indicates, there are only
two projections greater than .30 for "politician" on the
second dimension, while there are also two projections
larger than .30 on the third dimension. For "family
doctor" there are six stimuli whose projections are
greater than .30, but they are all less than .40.
It appears that only one dimension is really
interpretable. This includes "methodical" and "consistent"
at one end of the dimension and "unreliable" and "unpre
dictable" at the other end. The interpretability of the
configuration does not add much information in the deter
mination of the number of dimensions. Since the second
and third dimensions are both weak, and for comparison to
the other analyses, it was decided to maintain three
dimensions for further analyses.
The original configurations for "family doctor"
and "politician" did not appear to be too disparate. An
orthogonal rotation was carried out as in the four
references conditions. The stimulus projections for
"politician" were used to rotate to the target "family
doctor." When considering the adjectives in two dimensions
the configurations appeared extremely one-dimensional.
TABLE 10
Coordinates for Doctor and Politician
Doctor Politician Rotated Politician
Stim. 1 2
3 l 2
3
1 2
3
1
-47 -32 01 -59 -34
-13
-56 -38
-15
2
43 33
44 32 08
53 31
06 54
3
-44 -14 -04 -44
-03
14
-43 -08 14
4
29 12
-03 29 -20 06
31 17
o4
5 -43
-01 -08 -46 14 16 ' -46 08
17
6 112
-51 -19 105 -11 -18 -
105 02
-19
7
20 21 -18 27 20
-27 25 25 -25
8
-51 -15 13 -43 -25 26 -4o -32 24
9 -43
-04 10 -48
-07
01
-47 -12 00
10
37 11
-33 38 33 -15 34 38 -11
11 16 4o 24 33
07
-14
31 11
-13
12 -58 s 0 6 11 -50 -00
-23
-50
-03 -23
13
l6
33 -20 25
03 -31
24
09 -31
14
-45 00
-05 -47 20 11 -48
13
14
15- -51 -06
-13 -50 17 -09 -51 12
-07
16 109 33
21 98
-23
22 100
-15 19
Note.— Decimal points have Been omitted.
70
The configurations for the reference persons coincided
quite well, primarily because the first dimension was so
large. "Unpredictable" and "unreliable" are at the
negative end of this first dimension, while "consistent,"
"methodical" and "reliable" are at the positive end. It
makes it very difficult to identify clusters when the
entire configuration appears so one-dimensional.
Congruence of configurations. The correlations
between corresponding dimensions of "doctor" and "politi
cian" indicate that the first dimension corresponds very
well, .99j while the second and third dimensions
correlate poorly, .33 and .62, which are nonsignificant
correlation coefficients. These correlations are con
sistent with the previous results, since the first dimen
sion was strong and the second and third dimensions were
weak. Because of these puzzling unidimensional results,
further analyses exploring the shapes of the spaces were
not carried out.
Favorability and likability ratings. The average
favorability and likability ratings were obtained for
both "doctor" and "politician" and are presented in Table
A6 (in the Appendix). These were used as dependent
variables in four separate multiple regression analyses
with the corresponding stimulus projections for three
dimensions as independent variables. The resulting
multiple correlations and the relative weights are given
71
in Table 11. The correlations are fairly substantial, with
the highest being the favorability for "doctor!"
The plots for the first two dimensions for
"doctor" and "politician" are presented in Figure 7. Also
located in these configurations are the favorability and
likability unidimensional vectors. These vectors are
drawn from the origin indicating the maximum favorableness
and likableness. The directions of these vectors is
interesting in that for of both the references, they were
both as "likable as people" when described by the same
adjectives. "Doctor" and "politician" were most liked
when they were described by a combination of adjectives of
"reliable" and "responsible" and "quick-witted" and
"competitive." Looking at the favorability vectors,
there is a difference in the directions of these vectors
for the references. A "doctor" is most "favorable" when
he is described by traits such as "methodical," "reliable"
and "responsible," while a "politician" is most "favor
able" when described also by adjectives such as "quick
witted" and "competitive."
These results concerning liking generally confirm
those of Hanno and Jones. They found that evaluation of a
"politician" on both judgments was the same, the vector
being located midway between "enthusiastic" and "reliable:V
The favorability changed towards "methodical" when
considering "doctor" as a profession, however. In this
Multiple R
Relative 1.
Weights 2.
3.
TABLE 11
Prediction of Favorableness and Likableness Ratings
Doctor
Favorableness Likableness
• 96
-3-73
1.50
- 1.10
.83
-2.29
2.96
.77
Politician
Favorableness Likableness
.78
-2.09
3.15
-.79
• 70
- 2.02
2.12
.78
^3
r o
73:
ii
13 com petitive 16
7
10 a ggressive
8 3 level-headed
cau tiou s 1 2
u n relia b le 6
m ethodical 12,.5
15 9
Family Doctor
Fav.
Lik
m ethodical 12
9 level-h ead ed
10 ag g ressiv e
7
Ij. 16
13 2 com p etitive g .
u n relia b le
4
1 6
cautious
P o litic ia n
F ig. 7. Location o f p oin ts in two dim ensions fo r fam ily doctor and
p o lit ic ia n w ith d ir e c tio n s of favorableness and lik a b len ess
in d ic a te d .
present study, then, when the profession is considered,
the favorability may change, but the reference persons may
be "likable as people" when described by the same traits.
CHAPTER IV
DISCUSSION
Multidimensional Scaling Results
The main purpose of this study was to determine
whether the multidimensional scaling solutions obtained
■under the conditions of different positive and negative
reference persons were representations of the same under
lying configuration. The multidimensional scaling analyses
indicated that there were three dimensions, with two of
them being readily interpretable. After the rotational
procedures employed, it seemed that the representations
were indeed very similar. However, simply because these
multidimensional spaces appear similar, this is not strong
evidence in itself that there are essentially no differences
in the configurations between reference persons.
Additional, more impressive support was found for
the proposal that these solutions do in fact represent the
same underlying configuration. The relationship between
the multidimensional spaces was uncovered by systematically
manipulating the shapes of the spaces by using Procrustes
techniques on the matrices of stimulus projections. The
important aspect of this result was that the transformation
75
76;
of the "hypocrite" matrix caused the second dimension, a
dimension defined by "capable" on the positive end and
"inadequate" on the negative end, to be stretched in order
to more closely correspond to the "astronaut" and "surgeon"
configurations. This would be intuitively reasonable when
considering that being "capable" is of lesser importance
when obtaining judgments with respect to a "hypocrite."
In other words, not only did the Procrustean transforma
tion provide a better fit of the data, but the stretching
factor changed the shape of the space so that the length
of the axes became more meaningful in terms of the
specific reference.
These results are consistent with the findings of
Cliff (1965) in which differences in the multidimensional
scaling solutions between groups in which experimental
manipulation of the presentation orders of the stimuli
occurred. These differences were explored using Procrustes
techniques, finding the changed shapes to be interpretable
in terms of the experimental conditions. The present
results extend this work by using a very different
experimental manipulation.
It is apparent, then, that when judging adjectives
used to describe different reference persons, the shape of
the resulting representations are similar, but not
identical, but more importantly they are systematically
related to each other so that induced changes in one
77
representation corresponded in a meaningful way to the
other representations.
Relation of Favorability Ratings to the
Multidimensional Structures
Another purpose of this study was to determine how
the unidimensional favorability judgments would change
with respect to the different reference persons. The
favorability vectors do relate to the multidimensional
spaces as the previously mentioned figures have indicated.
This provides additional support for Cliff and Young's
(1968) point of view. In this study, the multidimensional
spaces were similar, yet the unidimensional favorability
judgments changed with respect to the reference person in
a meaningful way. The change in favorability from
"competent" for "astronaut" towards "cruel" for "killer"
occurred in both types of stimulus presentations. The
direction of favorability for "astronaut" matching the
"capable" and "intelligent" adjectives also corresponds to
the direction of the favorability vector for "ship's
captain" (Cliff and Young, 1968).
A second way to look at the relationship between
the multidimensional structures and the favorability
ratings involves comparing the regression weights from the
analysis predicting the favorableness of the adjectives
for the different references and the elements of the
transformation matrix used to transform "hypocrite" to
match the other structures. It is evident that the weight
for the second dimension in predicting the favorability
ratings is less important for negative references. This
corresponds to the earlier Procrustean transformation in
which this capability dimension is stretched after the
transformation. In predicting favorability of a
"hypocrite" and a "killer," then, how "adequate" or
"capable" they are is of less importance. More important
is the first dimension, defined by "cruel" and "kind,"
an evaluative kind of dimension. Likewise, when consider
ing the positive references, "astronaut" and "surgeon" the
first dimension, the "cruel," "kind" dimension is signifi
cantly smaller than the overwhelming capability dimension.
The main conclusions from this study are that when
the multidimensional scaling solutions are compared for
the different references, they appear to be very similar.
However, systematic changes may be induced, resulting in a
change in the relative length of the capability dimensions
for one of the negative references. These changes in the
spaces are meaningfully related to the reference person
involved. The change in the unidimensional vectors of
favorability are also systematically related to the refer
ence person. These results were also found under two
distinct types of stimulus presentation conditions.
79
Hanno and Jones Replication
The multidimensional scaling analyses of "family
doctor" and "politician," in general, did not replicate
the results obtained from Hanno and Jones (1973)* The
resulting configurations were essentially one-dimensional,
whereas the data from the Hanno and Jones study yielded a
two-dimensional structure. These results were very
puzzling, especially since the subjects used in this study
were not expected to be different from the subjects in the
main study using four reference persons. Subjects were
randomly assigned to both the study in which they par
ticipated and the specific condition in which they served.
The differences between the four references study
with more clearly defined dimensions and this replication
study were the use of different stimuli and the instruc
tions to the subjects. The 16 trait adjective stimuli
would seem to lie in more than one dimension since they
clearly fall in four clusters and had been previously
scaled in two dimensions by Hanno and Jones.
It is difficult to make any conclusions regarding
the similarities of the multidimensional scaling solutions
from the "family doctor" and "politician" because of the
unidimensional nature of the solutions. The only outcome
is that the spaces appear similar, as the figures indicate.
Even though the multidimensional scaling results were
inconclusive, the directions of the unidimensional
80
favorability and likability vectors were consistent with
part of the Hanno and Jones study. They found that there
was essentially no difference between likableness and
favorableness vectors when considering a "politician^but
that there was a difference between these vectors and the
favorableness vector of the "family doctor." The position
of the favorableness vector now was changed in direction
so that it lay closer to the locations of adjectives such
as "methodical." The explanation given by these authors
is that likability of a "politician" as a person is an
important asset and part of a "politician's" success in
his career. Intuitively, these would not be expected to
change. Hanno and Jones also utilized the resulting
differences in these unidimensional vectors to provide
support for their position that the trait adjective
configurations do change with a change in the reference
person.
The results in this present study indicate that the
favorability vectors do change as a function of the
reference person. It would appear quite reasonable to
assume that the favorableness and likableness of a
"politician" and "family doctor" could be distinguished
when considering the occupations and the persons them
selves separately. It seems that a "doctor" would be more
favorable when he was "reliable" and "responsible," while
a "politician" would be more favorable when described by
8i
"competitive" and "enthusiastic.!1
These favorability and likability results do not,
however, contradict the Cliff and Young results, and they
are completely consistent with the results of the favor
ability judgments from the main study of this investiga
tion utilizing four reference conditions. The vectors for
favorability and likability have been shown to be related
to the multidimensional space. These unidimensional
vectors may be meaningfully related to the multidimensional
scaling analyses because changes in the experimental situa
tion led to changes in these unidimensional vectors. The
likability and favorability vectors may be rotated to be
congruent with one another, the rotation indicating the
effect of the different references and whether they are to
be judged with respect to their favorability in their
occupation or their likability as people.
Unfortunately, these present data are inconclusive
so that they cannot be used as direct evidence for
rejecting the Hanno and Jones assertion that a person’s
internal organization does change when there is a change
in the reference. Refinements of the techniques employed
in this study need to be used in further research.
The results from the multidimensional scaling
analyses are inconclusive since they are one-dimensional.
However, the solutions for the two references do seem
similar. The main conclusion from this study comes from
82 ;
the results of the favorability and likability unidimen-
i
sional vectors in their relationship to the multidimen
sional space. These vectors are all distinct and seem to
be reasonably related to these structures. It is evident
that the adjectives were judged differently for "family
doctor" and "politician," as well as whether the occupation
was judged or the person himself.
CHAPTER V
SUMMARY
The main purpose of this study was to apply
multidimensional scaling analyses to trait adjectives,
extending the work of Cliff and Young (1968) in which the
basic hypothesis was that an individual has an internal
organization of a set of stimuli which may be revealed
by multidimensional scaling. Furthermore, different
multidimensional structures as well as other responses
would be systematically related to each other.
Forty subjects, in each of two stimulus presenta
tion conditions, judged the similarity of 12 adjectives,
each adjective paired with every other adjective (66
pairs). The pairs of adjectives were judged for
similarity when they were used to describe each of four
reference persons, "astronaut," "hypocrite," "killer" and
"surgeon." The two stimulus presentations were an ordered
references condition, whereby all the pairs of judgments
about a particular reference person were presented
separately, and a mixed references condition, in which all
pairs of adjectives for all references were entirely
randomized.
84
Eight separate multidimensional scaling analyses
were computed, using the TORSCA (Young & Torgerson, 1967)
program, including two stimulus presentation conditions
and four references. Three dimensions were indicated and
used in further analyses. The dimension loadings of the
references were orthogonally rotated to match one refer
ence condition (Cliff, 1966) in order to make the stimulus
configurations as similar as possible without changing the
shapes of the stimulus spaces.
Since the spaces could have different shapes,
systematic differences in the shapes were explored by
using Procrustes techniques. The stimulus loadings for
the "hypocrite" reference were transformed, using an
oblique transformation matrix, the elements of which were
the stretching and shrinking factors associated with
"astronaut." "Hypocrite" was changed so that a capability
dimension was lengthened, indicating that an important
dimension when Judging an "astronaut" was now stretched
in the "hypocrite" transformation. This transformation
showed how the shapes of the spaces are related in a
meaningful way. A variance-ratio test was also used to
show that this oblique fit was a better fit than the
orthogonal one.
Favorability Judgments for all four references were
obtained on the trait adjectives separately. These were
used in multiple regression analyses which related these
85
unidimensional Judgments to the multidimensional configura
tions. The differences in the favorability vectors for
the different references were meaningfully related to the
particular reference in all conditions. These results
supported the earlier Cliff and Young (1968) study.
A replication study of Hanno and Jones (1973) was
carried out using 16 adjectives and two references,
"family doctor" and "politician." Twenty subjects judged
the liklihood of co-occurrence of these traits when each
adjective was presented with every other adjective for a
total of 120 judgments for two references.
The two multidimensional scaling analyses yielded
one-dimensional solutions. These puzzling results did not
support the Hanno and Jones findings. Further analyses
were not performed on these data. Two additional judg
ments were obtained, favorability of the person as part of
his occupation and likability of the person. These were
used in multiple regression analyses as indicated
previously. While these one-dimensional results are
inconclusive, the change in the directions of the uni
dimensional vectors seem to show a meaningful relationship
to the particular reference person and whether he was being
judged in his occupation or as a person.
REFERENCES
86
REFERENCES
Anderson, N. H. Application of an additive model to
impression formation. Science, 1962, 138, 817-818.
Anderson, N. H. Note of weighted Siam and linear operator
models. Psychonomic Science, 1964, 1, 189-190.
Anderson, N. H. Averaging model analysis of set-size
effect in impression formation. Journal of
Experimental Psychology, 19655 753 156-165.
Anderson, N. H. Application of a linear-serial model to a
personality-impression task using serial presentation.
Journal of Personality and Social Psychology, 1968,
1^7 "354-362.
Anderson, N. H. Integration theory and attitude change.
Psychological Review, 1971* 78, 171-205.
Anderson, N. H. & Barrios, A. A. Primacy effects in
personality impression formation. Journal of
Abnormal and Social Psychology, 1961, b3, 346-350.
Anderson, N. H. & Jackson, A. Effect of stimulus incon
sistency and discounting instructions in personality
impression formation. Journal of Personality and
Social Psychology, 1965, 2, 531-539*
Asch, S. E. Forming impressions of personality. Journal
of Abnormal and Social Psychology, 1946, 2, 531-539-
Asch, S. E. Social Psychology. Prentice-Hall Inc.
Englewood Cliffs, New Jersey, 1952.
Boyd, J. E. & Jackson, D. N. The perceived structure of
social attitudes and personality: a multidimensional
scaling approach. Multivariate Behavioral Research,
1967, 2, 281-297.
Cliff, N. Multidimensional scaling and cognition: I.
Relations among judgments obtained under varying
conditions. Technical Report No. 1, November, 1965*
University of Southern California, Los Angeles.
87
88
Cliff, N. Orthogonal rotation to congruence. Psychomet
rika, 1966, 31, 33-42.
Cliff, N. Consistencies among judgments of adjective
combinations. In Multidimensional scaling, vol. II,
applications, Eds., A. K ! . Romney, R. N. Shepard and
S. B. Nerlove, 163-182. New York. Seminar Press,
1972.
Cliff, N. & Young, P. ¥. On the relation between uni
dimensional judgments and multidimensional scaling.
Organizational Behavior and Human Performance, 1968,
3a 269-85.-------------------------------------
Fishbein, M. & Hunter, R. Summation versus balance in
attitude organization and change. Journal of
Abnormal and Social Psychology, 1964, 69j 505-510*
Friendly, M. L. & Glucksberg, S. On the description of
subcultural lexicons: a multidimensional approach.
Journal of Personality and Social Psychology, 1970,
14, 55-05.
Gerjuoy, H. & Aaronson, B. S. Multidimensional scaling of
terms used to describe personality, Psychological
Reports, 1970, 26, 3-8.
Gollob, H. F. Impression formation and word combination
in sentences. Journal of Personality and Social
Psychology, 196b, 10, 341-353.
Guttman, L. A general nonmetric technique for finding the
smallest coordinate space for a configuration of
points. Psychometrika, 1968, 33, 469-506.
Hanno, M. S. & Jones, L. E. Effects of change in reference
person on the multidimensional structure and evalua
tions of trait adjectives. Journal of Personality
and Social Psychology, 1973a in press.
Kruskal, J. B. Multidimensional scaling by optimizing
goodness-of-fit to a nonmetric hypothesis.
Psychometrika, 1964, 29, 1-28. (a)
Kruskal, J. B. Nonmetric multidimensional scaling: a
numerical method. Psychometrika, 1964, 29a 115-129*
0>) ~
89
McGee, V. E. The multidimensional analysis of "elastic"
distances. The British Journal of Mathematical and
Statistical Psychology, 19bb} 19, 181-195.
Partington, J. T. & Jackson, D. N. A multidimensional
scaling evaluation of alternative models in person
ality impression formation. Psychonomic Science,
1968, 10, 331-332.
Rosenberg, S., Nelson, C. & Yivekananthan, P. S. A
multidimensional approach to the structure of
personality impressions. Journal of Personality and
Social Psychology, 1968, 9* 283-294.
Rosenberg, S. & Sedlak, A. Structural representations of
perceived personality trait relationships. In R. N.
Shepard, A. K. Romney and S. B. Nerlove (Eds.),
Multidimensional scaling, vol. II, applications. New
York, New York: Seminar Press, Inc., 1972.
Schonemann, P. H. A generalized solution of the ortho
gonal Procrustes problem. Psychometrika, 1966, 31a
1-10.
Schonemann, P. H. & Carroll, R. M. Fitting one matrix to
another under choice of a central dilation and a
rigid motion. Psychometrika, 1970, 35, 245-255.
Shepard, R. N. The analysis of proximities: multidimen
sional scaling with an unknown distance function. I.
Psychometrika, 1962, 27» 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)
Torgerson, ¥. S. Theory and methods of scaling. New
York: Wiley, 1958.
Tucker, L. R. & Messick, S. An individual differences
model for multidimensional scaling. Psychometrika,
1963, 28, 333-367.
Walters, H. A. & Jackson, D. N. Group and individual
regularities in trait inference: a multidimensional
scaling analysis. Multivariate Behavioral Research,
1966, 1, 145-163.
90
Wyer, R. S. A quantitative comparison of three models of
impression formation. Journal of Experimental
Research in Personality, 19&9» 4, 29-41.
Young, F. W. & Torgerson, W. S. TORSCA, a fortran IV
program for Shepard-Kruskal multidimensional scaling
analysis. Behavioral Science, 19^7, 12, 498.
APPENDIX
91
1
2
3
4
5
6
7
8
9
10
ll
12
TABLE A1
Analyses of Variance for Astronaut (upper half) and Hypocrite (lower half)
1 2
3
4
5
6
7
8
9
10 11 12
— 1.21 .30
.95
.24
.55 • 71 .27 • 33
1.62 .16 4.44**
.54 — .60
1.91
.36 .81 .67 1.39
.72 1.12 ro
.41
.42 1.64 — l.4l .47 • 90 • 73
2.80
2.55
.12
1.53
.14
2.81
2.35
i.o4 — l.4o 1.25 .38 2.85
.24 .81 .92 .51
2.52 1.37
5.41**1.40 — .06
• 71 .50 .29 .25 1.69 .90
.60 1.64
.57 .03 .83
— 2.34 .02 3.23* .25 .67 3.32*
.45 2.79
1.34 l.4o
.89 2.77
— 1.70 .84
• 95
2.54 1.60
.88 2.06
.37 4.03* 2.10 1.07 1.76 — .48
1.39 1.17 .05
.42
00
0
•
.64 .56 1.54 .89 .56 1.63 — .03 • 74 .52
.65
.18 4.83** .78 1.50 1.23 1.17
.41 1.02 —
.67 1.35
.49
1.14
1.79 1.58 .34 • 95 4.57* • 52 .34 .41 —
.17
1.20 2.87* 1.07 3.33*
00
00
OJ
.46 .36
2.57
1.16
1.37
1.78
--
TABLE Al— Continued
Analyses of Variance for Killer (upper half) and Surgeon (lower half)
1 2
3
4
5
6
7
8
9
10 11 12
1 — 3-10*
1.19 1.83 1.45 1.32 .75 .36
.91 .27 .59
1.82
2 .78 — .16 .82
.29 3.63* l.4o 2.40
.37 .81- .40 l.4l
3 • 71 .19
—
1.73 .19
.80 .36
00
*X>
•
1.11 .86 .04
2.09
4 .61 1.38 1.68 —
2.77 .35
2.11 .76 .49
.18 2.30
• 73
5
.04 .88
1.27 1.19
--
1.45 .51
.84
.03 1.69 1.81
.31
6 2.46 .86
.29 .33
2.12 — .52 1.07 .34 2.91* 1.20 .82
7 1.73 .77 .19 3.65* 1.70 2.53
—
.93
.84 1.88
.65 .67
8
1.17
.66
.87 1.94
1.53
.68 .20 — .50 3.25*
.86
• 31
9
.82
.09
3.21*
.73 .43 1.57
i.o4 .60 — .46
.83
2.01
10 .63 1.08
• 57
1.36 1.03 2.03 .37
.20 1.02 —
.65 .63
11 .41
.35
1.22
.23 .57 .63 1.32 1.05 2.46 1.06 — .21
12 2.94* 1.28
.79 • 98 1.18 1.84
1.19
1.20 1.21 .72 2.28
*pz. .05; **p< .01
TABLE A2
Analyses.of Variance for Doctor (upper half) and Politician (lower half)
1 2
3
4
5
6
7
8
9
10 11 12
13
14
15
1 .02 2.62 .02 1.54 .04 1.78 5.74* .09
.01 1.66 1.28 .56 .03 .87 •
2 2.24 —
.33 .05 1.97 1.07 .79 .07 .49 .17 .05 .01
2.71 .01 .08 1.
3
.21
.59
— 1.01 .41 • 34 .01
1.99
.21 4.47* 3.24 l.ll 3.54 .32 3.04 1.
4 4.74*3.29 2.16 —
.07 .33 • 70 .28 1.72
1.75
1.12 .01 2.24
• 92 .26 •
5
.68 .16
.07
.42
—
.24 .08 .92 .07 .68 .60 .92 .41 .06
1.83 •
6 .18 -
4.53*
.04 .41 .01 - -
.07 .01 .01
.63 1.60 1.74 .02
1.71
2.49 •
7 3-51
.22 2.28 .01 .02
2.79
----
4.35 • 91 8.68** .37 .06
• 95
1.66 .10 1.
8 .01 : 1.14 .60
• 51
.62
1.19 • 38
- -
.49 8.36**1.04 .18
.09 .01 1.04 1.
9 .43 1 6.4l* .38 .48 0.00
7.95* .32 .05 .78 1.45 .29 .58 1.70 .01 2.
10
1.19 .21 .78 3.43 1.87 .91 .6014.70**.00
—
.04 1.00 .02 .22
.71
0.
11 .20 .58 .47
.18 1.36 .06 .00 1.08
• 39
.00 .01 .18
.13 • 34 2.
12
.29 .27 .58 2.98 .62 .00 .01
.33
.00 .00
1.37
— —
.00 .48 .00 •
13
.02 .01 .02 1.01 .00 4.8l* .04
.49 .31 .03 .02 • 32
_ m m
.07 1.15 •
14 1.52 .00 .00 .46 .42 1.64 l
.57 • 37 2.24 8.58**1.56 .63 .06
.55
•
15
.54 : 3.26 .24 4.47* .28 1.21 .58 .36 .08
.17 1.95
.36 .02
1.33
1.
16
.59 : 1.30 3.34
.25
1.46
• 53 .65 .01 .30 3.05 1.46 1.46 3.58 1.29
1.28
*p ^ . 05
**p< .01
16
57
52
36
05
72
85
83
36
11
00
24
06
67
83
01
vo
4 = -
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
TABLE A3
Descriptive Statistics for Subjects
Ordered References
Mean Mean
Simil S.D. Favor S.D.
4.43 2.55
4.92 3.21
4.97
2.12
5.27 2.87
4.94
2.59
4.60 3.12
4.69 2.43 5.17 2.65
4.84 2.44 4.48 2.98
4.oo 2.66 5.40 3.46
4.21 2.19 4.75 3.15
3.98 2.46
4.77 2.77
4.4i
2.89
4.68
2.59
4.31 2.37 5.17
3.06
4.58 1.28
5.33
2.24
5.55
2.16 5.56 2.60
1.71 1.59
4.60 2.62
3.11
l.4o 5.21 3.12
3.93
2.21 4.92 2.74
4.47
2.49 5.08 2.34
5.02 2.18 5.42 2.94
4.38 2.26 4.94 3.26
4.64
2.77 4.67 3.21
4.25 2.77 4.67
3.41
3.80 2.30 4.71 2.59
4.57
2.50 4.98 3.38
4.o4 1.85 4.81 2.26.
5.18 2.48 4.56 3.05
3.02 2.22 4.50 2.92
4.52 2.61
4.77 3.09
4.68 2.14 5.19
2.40
5.36 2.69 5.o4 2.01
3.97 1.53 3.92 1.68
4.49 2.25
4.48 2.47
4.01 2.65 5.25
2.94
4.72 1.80 4.60 3.03
4.01 2.95
5.00 2.88
4.74 2.25
5.22 2.68
4.44 2.18
5.19 2.51
3.71
3.18 5.23
2.86
4.98 2.85 5.46 2.58
4.6l 2.50 5.63
3.01
4.99
2.56 4.88 2.56
TABLE A3— Continued
VO
C T\
n J
cm
i H
i n
O
CO
n
w
H
•H
a
< 1) - H
a w
S'
a
o
03
P
0 )
Ah
M3
03
in
00
C M
00 O CO COM3 00 H O O ^tV O IVVO C'-03vt-VO CO 03 CO C M CM-^--=d-CO O H CO COM3 H - = t v f O ^ t C M C--00 03 C M O
0 0 00M3 IVtvVOCO C O in t'-tv H tV O V O V O t^-M3 IVM3 t^-=t in tv IMX> VOCO-d-VO C^-inVOVOVOVO C —
f H
P o in
3 > vo
0 c t i
a
a
MO
on
C M
oo
M3
o
03
a )
o
P
0
P
0
< HH
0
a:
' d
0
s
a
CO in O 00 M3 H tv COM3 in ^ t D^-M3 H [S-00M 3^t tv C M C 33V O Ln^f- IS-CM COOO O M 300 CMLnCM-^-V0H03CM
cm co H c — oo^t- cm c-ao o CMM3M3 h ivcoovcm cm^-m3 on c m o 030-3- 03 is -in 030 0 3 V 0 ino o i n o in 0 3
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • « • • • • « •
COCMOOCMCMOOOOCMCMOOCMCMCMPOCMCMCMOOOOOOCMOOCMOOCMOOCMHr-ICMCMOOCMCMCMCMCMPOCMCM
ts-cocooo o c o i s i n i s L n i n i n o o o i o ^ - i n L n c o o H i o c M i n H o o o o t^-m ^t- h co H -=j-vo cm co
03 C M M3 CM^t^tCO C --tv H COM3 M3 C M 03H -=i- O 3 t '- t v C '- O 0 0 in in t V C M a 3 O incMM3 OCMCMCM-=tinCJ3tV
• • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • • a
-= !■ in^t- m ^ - in-=t ^ in in in-=t in ^ j- in in - ^ -=j- ^t- in-=t -=t in ^ f in ^ t- in in in in in in -cf -=t
OVd- O C M VO in C M VO 03-3- C M P O C h C M lO ^ -^ -O H H O COOO H CT\H O C M i n H 4 - U ) O lO S -C O H COO
O C O -= t^ t CO H Lnco^}- O 0 3 0 0 0 00 C M O COtvVOCO 03M3 C M IVtS-VO C M inoO COH COvf CO C M C0-=t 00-3" in
a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a a
CMCMCMCMCMOOCMCMCMCMHCMHHCMCMCMCMCMCMHCMCM-3-CMCMCMCMHCMOOCMCMCMCMCMCMOOCMCM
C M 00 CO tv O O C M D^-M30000 tVO 3 0 O i n H in in c O tV C M V 0 L n 0 O 0 0 H 0 0 C M tV t'-tV in O M 3 H -3 -in 0 0 0000
H -3 - H C M VO in H 00-3 -CO incMM3M3 O C M 00 COinoO C O C O in ^ -0 3 0 M 3 v i-M 3 [V-=t H tV C ^ O C M t--03VO H
• • • • • • • • • • • • • • • • • • • • • ■ • • • • • • • • • • • • • • • • • a
-3- i n ^ - ^ t co -3-^ t cm in 0 0 ^ - 3- o o o o in -3 --3 --3 -^ t c*-pM--3--3- cocmvI--3--3- coco-3-
H C M CO-3" inM3 IS-00 03 O H C M CO-3- UOVO [VOO 03 O H C M CO-3- LnVO IV 00 03 O H C M CO-3" LOVO N-00 03 o
............................................... in in in in in in in in in in v o vo vo vo vo vo vo vo vo vo t v t v is - t s - t '- t v is - t v iv t '- o o
0
a
0
■p
•H
Tt
0
-p
0
0
Ph
0
p
o
^3-
S M
o
r a
a
o
•H
+5
3
iH
0
u
S - H
o
0
•p
03
0
•p
0
p
1
•p
ra
0
EH
0
TABLE A4
Likability (Favorability) Ratings for Adjectives
and Reference Persons
Four Reference Persons5 Condition: Favorability
Adjectives Astronaut Hypocrite Killer Surgeon
Ord. Mixed Ord. Mixed Ord. Mixed Ord. Mixed
1. Bold 7.4
7.3 5.7
6.0 6.0
6.9
6.3
6.3
2. Kind
5.9
6.2 4.1 4.1
3.5 2.9
7.4
7.1
3. Inadequate 2.0 1.6
3-9
4.6 3.4
3.3
2.0 1.6
4. Ruthless
3-3 2.9 5.1
5.2 5.6
6.3 2.3 2.3
5. Aggressive
6.5
6.6 5.6 6.4 6.2
7.1
5.4 6.1
6. Intelligent 8.1 8.4 5.8 6.0 6.0
5.7
8.6 8.8
7. Cautious
7.1 6.7 4.9 4.9 5.9
5.6 7.2
6.7
8. Cruel 2.4 2.1
4.7
4.8
5.9
6.8 2.2 2.1
9. Ineffective
1.9 1.7 4.3
4.0 3.6
3.1
2.1 1.8
10. Sympathetic 5.4 4.8 4.0 3.6
3.5
2.8 7.0 6.8
11. Timid
3.1 3.7 3.5 3.5
2.4
3.3
2.6
12. Capable 8.4
8.5 5.9
5.6 6.1
6.9 8.5
8.6
TABLE A5
Descriptive Statistics for Subjects Doctor and Politician References
Mean Mean
Person Simil S.D. Favor S.D.
1
4.95 1.74 5.44 2.12
2 5.38 2.30
5.75 2.39
3
5.23 2.08 6.41 2.44
4
5.57 2.38
6.75 2.28
5 6.05 1.92 6.41 2.64
6 4.78
2.07 6.06 2.22
7 4.43 2.54
5.75 2.15
8 4.88
2.79
6.34
2.79
9 5.50 2.44 6.84
2.37
10
5.59 1.85 6.66 2.62
11
5.69 2.83 6.47
1.79
12
5.45
2.99 7-19
3.06
13 5.48 2.58
7.53
2.48
14
5.23 2.79 7.13 2.58
15 3.83 1.84 5.44
2.29
16 5.00 2.18
6.03
1.99
17
5.37 2.63 5.72 2.52
18 5.54
1.95
6.4l 2.00
19 5.65
2.73 6.63 2.46
20 5.21
1.79
6.00
1.98
Mean
Likab,
5.38
5.88
6.44
6.78
6.38
6.25
5.44
6.72
6.72
6.97
6.34
6.44
7.25
7.28
6.06
6.23
6.31
6.50
7.22
5.66
S.D.
2.10
1.90
1.20
1.87
1.82
2.05
1.94
2.24
3.01
2.64
1.59
2.99
2.02
2.25
2.49
2.33
1.96
2.08
2.93
1.73
a. Test-retest correlations for 48 repeated items.
99
TABLE A 6
Family Doctor and Politician: Favorability and
Likability
Doctor Politician
Fav. Lik. Fav. Lik.
1. Cautious 7.40 6.10 6.30
6.55
2. Nonconforming 4.15 5.60 4.80
5-75
3.
Level headed
7.75 7.45
7.60 7.65
4. Inventive
6.35 7.35 6.55 7.25
5.
Responsible 8.65 8.15 7.55
7.80
6. Unreliable 1.4o
1.75
2.60 2.00
7.
Enthusiastic 7.40 8.00 7.70 8.00
8. Calm 8.15 7.10 6.35
6.40
9-
Self-disciplined 8.00
7.55
7.00
7.55
10. Aggressive
5.25 5.10 7.25 5.85
11. Quick-witted
6.35
8.00 6.90 7.40
12. Methodical
7.15
5.60 5.80 5.50
13.
Competitive 5.85 6.25 7.30 6.20
14. Reliable 8.60 7.90 7.45 7.75
15.
Consistent 7.85 7.10 6.45 6.8o
16. Unpredictable
1.95 3.95 3.25
4.40
100
SUBJECTS' INSTRUCTIONS FOR FOUR
REFERENCES CONDITION
The general purpose of this research is to study
how people form impressions of others. In many situa
tions people form impressions by knowing only a limited
amount of information. This study attempts to look at
this problem by restricting the amount of information one
knows by using only a few personality traits.
On the following pages two different adjectives
are used to describe the same kind of person. These
adjectives may vary in many ways. Two examples are:
big businessman fat businessman 123^56789
sneaky thief friendly thief 123^-56789
Your task is to rate these pairs on a scale from
one to nine in the following way. Think of the kind of
person the first adjective describes. Then think of the
kind of person the second adjective describes and judge
the similarity between these two people. If the kinds of
people are similar, then use the high numbers in your
judgments. For example:
kind friend considerate friend 123^-56789
If the kinds of people are very dissimilar, then circle
one of the low numbers.
cruel murderer considerate murderer 123^56789
Use the middle numbers if the two people are of inter
mediate similarity. When you have decided on your judg
ment, circle the number to the right of the pair.
Try to work quickly, but thoroughly. We are
interested in your first response to the pair. Some of
the pairs of adjectives may be repeated. Please respond
to the pair as it is presented and don't try to recall
how you responded previously. It is also important not to
leave any pairs blank. Even if you are unsure of your
response, it is better to give an answer than to leave it
blank.
Several practice -trials follow. Remember to
judge how similar two of the same kind of people would be
if they were described by different adjectives. High
numbers represent greater similarity while low numbers
represent greater dissimilarity. Are there any questions?
Please judge the following pairs.
SUBJECT'S INSTRUCTIONS FOR HANNO AND
JONES REPLICATION
The general purpose of this research is to study
how people form impressions of others. In many situations
people form impressions by knowing only a limited amount
of information. This study attempts to study the problem
by restricting the amount of information one knows by
using only a few personality traits.
On the following pages there are pairs of
adjectives used to describe a FAMILY DOCTOR. These
adjectives may vary in many ways. Two examples would be:
FAMILY DOCTOR: helpful intelligent 123^56789
FAMILY DOCTOR: clumsy dull 123^56789
Your task is to rate the pair of adjectives on a scale
from 1 to 9 in the following way. If the pair of
adjectives is highly similar, then circle the number 9
to the right of the pair. If the pair of adjectives is
highly dissimilar, then circle the number 1 to the right
of the pair. Use the middle numbers if they are
moderately similar or dissimilar.
In your ratings you are to consider the
similarity of any two of the adjectives as being equi
valent to the liklihood that both of them apply to your
image of a FAMILY DOCTOR. In other words, given that the
DOCTOR is 1st adjective, how probable is it that he is
also 2nd adjective. IT it is very probable, then you
would say that the two adjectives are very similar. If
it is very unlikely, then you would say the two adjec
tives are very dissimilar. Intermediate degrees of
probability would indicate intermediate degrees of
similarity. Two examples would be:
FAMILY DOCTOR: kind considerate 123^56789
FAMILY DOCTOR: kind cruel ■ 1.2 3 ^ 5 6 7 8 9
Try to work quickly, but thoroughly. We are
interested in your first response to the pair. It is
important not to leave any pairs blank. Even if you are
unsure of your response it is better to give an answer
than to leave it blank. Also, try to keep the reference
person constantly in mind while judging the pairs. Some
of the pairs of adjectives may be repeated. Please
respond to each pair as it appears, since we are simply
interested in your first response to each pair of
adjectives as it comes up.
Several practice trials follow. Remember to
judge how likely it is that the two adjectives of the
pair apply to your image of a FAMILY DOCTOR. High numbers
represent a greater liklihood of going together in a
FAMILY DOCTOR, while lower numbers represent a lower
liklihood of going together in a FAMILY DOCTOR.
Are there any questions? Please judge the
following practice trials and then continue through the
rest of the booklet.
Linked assets
University of Southern California Dissertations and Theses
Conceptually similar
PDF
A Rotational Approach To Psychological Invariance
PDF
Personality Variables And Intellectual Abilities As Determinants Of Concept Learning
PDF
A Multidimensional Scaling Of Mood Expressions
PDF
A Critical Test Of The Dominant Motive Of High Need-Approval Persons
PDF
An Empirical Analysis Of The Relationship Between Risk Taking And Personal Probability Responding On Multiple-Choice Examinations
PDF
A Monte Carlo Evaluation Of Interactive Multidimensional Scaling
PDF
Hypnotic Susceptibility, Achievement Motivation, And The Treatment Of Obesity
PDF
Intellectual And Cognitive Factors In The Production Of Psychological Stress Reactions
PDF
The Effects Of Making Social Desirability Judgments On Personality Inventory Scores Of Schizophrenics
PDF
Effects Of Different Treatment Procedures On Reading Ability And Anxiety Level In Children With Learning Difficulties
PDF
The effects of an RNA polymerase in the improvement of verbal learning and memory
PDF
Role Playing Ability And Social Adjustment In Children
PDF
Measuring Thought Process As An Ego Function In Schizophrenic, Mentally Retarded And Normal Adolescents By Means Of The Rorschach
PDF
The Effect Of Subject Sophistication On Ratio And Discrimination Scales
PDF
A Factor Analysis Of The Semantic-Evaluation Abilities
PDF
Non-Specific Treatment Factors And Deconditioning In Fear Reduction
PDF
Dissonance and Self-Perception Analysis of "Forced Compliance": When Two Theories Make Competing Predictions
PDF
The effects of physical proximity and body boundary size on the self-disclosure interview
PDF
An Exploration Of Interpersonal Behavioral Possibilities And Probabilities
PDF
The Systematic Isolation And Validation Of Personality Determiners In The Handwriting Of School Children
Asset Metadata
Creator
Doherty, Linda Margit
(author)
Core Title
The Relationship Of Multidimensional Scaling Spaces Of Trait Adjectives For Different Reference Persons
Degree
Doctor of Philosophy
Degree Program
Psychology
Publisher
University of Southern California
(original),
University of Southern California. Libraries
(digital)
Tag
OAI-PMH Harvest,psychology, clinical
Format
dissertations
(aat)
Language
English
Contributor
Digitized by ProQuest
(provenance)
Advisor
Cliff, Norman (
committee chair
), Conolley, Edward S. (
committee member
), Fox, Frank H. (
committee member
)
Permanent Link (DOI)
https://doi.org/10.25549/usctheses-c18-873855
Unique identifier
UC11363431
Identifier
7400913.pdf (filename),usctheses-c18-873855 (legacy record id)
Legacy Identifier
7400913
Dmrecord
873855
Document Type
Dissertation
Format
dissertations (aat)
Rights
Doherty, Linda Margit
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
Tags
psychology, clinical