1vu1‘ “31!; "3'1 . ’.

   
  
 
  
  

  
  
  
 
    

, ' ,
1
l ‘7
- *2

 

'arzh 1
n I 1313,13.

'11..
,11m911Mp311MH}
K343115311 ' 1'"

.1, 1111' 11131331111111: 5.1.3:: 11 -

  

nm‘

     
  
    
 
 
 
 
  
  
    

 
  
   

' an in m. 3. v
£1211 '4‘“! lift» .11 2.1"

   

 

.1.

1
'313’.
.' . ’1 U '
. 3 £11?“
. fig‘:
!

    

.. -
s
1

._"_¢.-—-
lob-“H
_' w

     
  
 
   

    
   
 
 
   
  
   

.-I u- ’
.._
.1.

,‘1
“"53
5m¥3**lwa§
.4 .L 5135;:
. '13-'31-

2%

+1; M» 7

     

       
 
 
 
   
  
  

 

  

A
. . I 8 O r
:‘2 . .3 : fl- ‘ .' .;
.1fig 1“? ._:.
1 1,7":111‘7111' ’ 3 - -
r“ 21%Mmg 3 «a .
‘3 ~ éu 1 hi . ~
“131:3 «13.1 3533;": "must 37' 3‘33:
' t; _".1.\(

   

. W},
)[l'
{'9' ‘

J
21K

‘-
.WE'

   
    
 
  
  

   
 
    

         
         
    

 

 

3 V]
j: “2. 3" "I".
.31}, .331 :, "' "1"."
2 'fiflgfiim 3N13”’
1; s 1 ""'-"1';'13'1...”3133'.
.WW. W£|JEM“”“’JPM3HR%QR' 3'
.l‘w '1 | '31 13323'3‘33'1391-"335'1M 11 "" " 1“”
3.23311, 2111,1111. '3 '321; 3‘" .1.1,;""j-'1:1_-11'
. 1 3.1.13.1, 11, M15 "1 '13". ' '-
I'HPVM 1'w”31‘fl
’l ‘ “‘1 .' H“ "|Ji“2"'231'331'33,"‘:l"2 I 1 . 1'
'II '13: "111,331: 333' 311333
'I
‘ 1.1'1'I1' I1" 1 II'I'1111

1311111131

 

HHW1'11

W LIBRARY

w. Michigan State
' University

 

 

This is to certify that the

thesis entitled
Behaviors, attitudes, and traits as information

units in person perception
presented by

James W. Gard

has been accepted towards fulfillment
of the requirements for

Ph.D. Psychology

degree in

 

 

yam? z’i 20W

Major professor

Date 53/31/77

0-7 639

 

BEHAVIORS, ATTITUDES AND TRAITS AS INFORMATION
UNITS IN PERSON PERCEPTION

BY

James William Card

A DISSERTATION

Submitted to
Michigan State University
in partial fulfillment of the requirements
for the degree of

DOCTOR OF PHILOSOPHY

Department of Psychology

1977

ABSTRACT

BEHAVIORS, ATTITUDBS AND TRAITS AS INFORMATION
UNITS IN PERSON PERCEPTION

BY

James William Gard

Person perception research, following seminal work by
Heider (1958) and Asch (1946), can be roughly divided into
two basic approaches: person attribution and attribute
organization research. Lines of investigation in each of
these approaches have proceeded in independent but paral-
lel directions. A model is proposed which views person
’perception as an inference process between three distinct
units of information available to the perceiver: the
actor's behaviors, attitudes, and traits. It is postu-
lated that each of these units has some probability of
mediating information presented to the perceiver in
behavioral, attitudinal, or trait form. Person attribu-
tion and attribute organization studies are seen as
focusing on different types of inferences with the same
perceptual objective.

Attribute organization studies have focused on the
structure of trait inferences mediated by other traits.
The implication of the proposed model is that the structure
of these inferences would differ if the inferences were
mediated by attitudes or behaviors. Person attribution
studies imply that these units differ with respect to the

specificity of information provided to the perceiver and

James William Gard

that the inference process goes more naturally from behav-
iors to attitudes to traits rather than vice-versa. Two
experiments were designed to explore these arguments.

In Experiment One, 60 upper-division psychology stu-
dents made 290 likelihood judgments that a particular unit
was to be inferred from another. Examination of the factor
structures of these inferences indicated that structural
representations of trait inferences were different when the
inferences were mediated by attitudes and behaviors.

In Experiment Two, 48 upper-division psychology stu-
dents were divided into twelve groups corresponding to the
nine basic inferences proposed and three no-mediator control
conditions. Each group responded to discrimination learning
tasks which required subjects to learn a particular behav-
ior, attitude, or trait given stimuli in one of these
three forms. More correct answers were made on the behav-
ior-to-attitude, behavior-to-trait, and attitude-to-trait
inferences supporting the argument that these units differ
in the property of specificity of information. Implications
of these findings for person perception research are dis-

cussed.

ACKNOWLEDGMENTS

I wish to express appreciation to several persons who
assisted me in the successful completion of this project. I
am deeply grateful to Dr. James Phillips, who, as committee
chairman, contributed time, expertise, and emotional support
when these could have been rightfully directed at more pres-
sing professional matters. He provided propitious doses of
levity and gravity, together with well placed "awks" and
"uncs". I am also grateful to Dr. Eileen Thompson, who pro—
vided careful guidance and subjected herself to the role of
sounding board for my ideas. I wish to thank Drs. Lawrence
Messé’and Raymond Frankmann for insightful comments and
questions. Finally, I owe special debts of gratitude to
Arline Jennex, for typing earlier drafts; to Peter Pirozzo
and Bill Walsh for programming assistance; and to the 490

students for important legwork.

ii

TABLE OF CONTENTS

List of Tables ........................................ iv
Introduction .......................................... 1
Historical perspective ............................. 1
Person perception: An information analysis........ 6
Mediation processes ................................ 17
Implications for a mediational model of person
perception ....................................... 19
Methods for investigating inferential processes
in person perception .............................. 26
Experiment One ........................................ 33
Method ............................................. 33
Overview of experimental procedures ............. 33
Selection of stimuli for the judgment task ...... 33
Administration of the judgment task ............. 37
Results ............................................ 39
Discussion ......................................... 43
Experiment Two ........................................ 45
Method ............................................. 45
Overview of experimental procedures ............. 45
Selection of stimuli for the discrimination
learning task ............ .. ................. 47
Discrimination learning task. ................... 54
Results..... ...... ....... . ........................ 57
Discussion................. ........................ 61
Implications of results ......................... 67
Reference notes ........ . .............................. 72
Footnotes.. ........................................... 73
Bibliography .......................................... 74

iii

LIST OF TABLES

Table 1
An information unit model of person perception ........ 23
Table 2
Behaviors, attitudes, and traits selected for
the judgment task.. ............. . .................... 38
Table 3

Factor loadings for two-factor solutions of
intertrait correlation matrices computed
across behaviors, attitudes, and traits .............. 41

Table 4
Stimulus sets for each selected mediator, the
mean probability ratings and the standard

error of the mean ratings ......................... 51-52
Table 5
Stimulus sets for the no-mediator control condition...53
Table 6
Mean total errors for each between-subject cell
and marginal means ................. . ................. 59
Table 7
Mean errors for each trial ............................ 60

iv

INTRODUCTION

This study examines the similarities between attribu-
tion theory and "implicit personality" approaches to person
perception. The major objective of the study is to propose
a process model for integrating seemingly unrelated find-
ings in these areas of research. In addition, basic assump-
tions of attribution and "implicit personality" approaches
to person perception research are tested with the framework
of the model. Prior to the presentation of the model, it
would be useful to trace the development of person percep-
tion research, noting the importance of the original work
in laying the foundations of the model.

Historical perspective

With the publication of "Forming impressions of person-
ality" by Solomon Asch in 1946, the study of person percep-
tion became prominent in social psychology. Prior investi-
gations of personality variables seemed to assume that a
person's attributes such as kindness or generosity were
somehow inherent to that person and perceptible to the
sophisticated observer. Asch, operating from a Gestalt tra-
dition, asserted that such trait information depends greatly
on the context or the entire stimulus set of traits the ob-
server is given to interpret. His finding that minor
changes in the traits ascribed to a hypothetical stimulus

person produced radically different personality impressions

strongly suggested that the observer or perceiver may be as
legitimate a focus of research as the person observed.

It would be misleading to contend, however, that per-
ceiver variables had previously been entirely ignored.
Tagiuri (1969) indicated that investigations in such research
areas as the perception of emotions in facial expressions
(e.g., Woodward, 1938) and "empathic" judgment (e.g., Estes,
1938) were numerous. These studies were often concerned
with identifying the variables associated with "accurate"
judges. While identification of such variables was germane
to person perception, the basic arguments in such studies im-
plied the existence of objectively measurable stimulus refer-
ents discernible to the accurate, i.e., sophisticated per-
ceiver.

The early studies did not seem to recognize that all ob-
servers of persons might use similar processes in integrating
person stimuli, independent of the measurability of those
stimuli. In addition, studies of the accurate perceiver were
confronted with many problems. The results were often inco-
herent and ungeneralizable from one study to another (Tagiuri,
1969). This state of affairs was related to a-number of issues
surrounding the methods and assumptions of these studies.

Bruner and Tagiuri (1954) noted that systematic biases
in judgment may have confounded many of the accuracy studies.
Among such errors were the "halo effect" (Thorndike, 1920),
which reflects a tendency to rate persons in several cate-
gories in terms of a global impression of goodness and bad-

ness. Another bias, the "logical error" (Newcomb, 1931), is

3

a judge's preconception of what traits go with other traits.
Bruner and Tagiuri have argued that the accuracy of a parti-
cular judge may merely reflect his use of the bias used most
often by other judges. Bender and Hastorf (1953) contended
that what passes for "social sensitivity" or accuracy might
simply be similarity between the judge and the person judged.
Cronbach (1955) also called into question the reliability of
accuracy measures. Thus, perceiver variables involving
biases and similarity may have been confounded with "accu-
racy", and it was uncertain that accuracy could be measured.

The goal of identifying accurate observers obscured the
fact that all persons perceive others. With respect to that
fact, the investigation of those processes, including error
tendencies, which characterize both the sephisticated and
the unsophisticated observer seemed to be a better, more
general approach to understanding person perception. The
work of Fritz Heider (1958) was historically important in
establishing the notion that the naive observer of others
proceeds to interpret interpersonal events in the same way
as the more saphisticated observer.

Heider based his observations on the premist that there
are formal similarities between the perception of persons
and the perceptions of objects. Borrowing from the "lens
model" of Brunswik (1956), Heider argued that the pr0per-
ties of the distal stimulus, the object to be perceived,
are only perceptible through the proximal stimulus. The
proximal stimulus is an aggregate of stimulus characteris-

tics of the object and other stimuli from the

4

environment. Assuming that stable representations of the
properties of the object are sought, the perceiver sorts

the proximal stimuli to arrive at some notion of the constant
preperties of the distal object.

Experimental verification for the analogy between object
and person perception was provided in related experiments by
Heider (1944) and Michotte (reported in 1963). They were
able to demonstrate that naive subjects were quite willing to
infer properties such as "bullying" or "following" from the
size and movement of triangle figures and blackened dots in
animated drawings. It therefore seemed that naive observers
responded to non-person stimuli as if they were persons. The
attribution of stable characteristics to the triangles and
dots was also implied by these responses. It seemed that the
goal of the perceivers was to achieve stability and meaning
in what were, at face value, ambiguous stimuli.

The scientific psychologist was chosen by Heider as the
prototypic sophisticated observer of persons. Presumably,
the psychologist proceeds in much the same fashion as that
described by the lens model. Faced with a diverse and often
confusing stimulus set, the psychologist arrives at constan-
cies in the person or events observed. More specifically,
the psychologist observes human behavior, seeks to determine
the motives or intentions behind that behavior and attri-
butes those motives to some underlying disposition in the
person. The underlying disposition is then presumed to have
been the antecedent of the elicited behavior.

The Asch experiment and Heider's generalizations have

5
greatly influenced both the volume and the direction of per-
son perception research in the past two decades. Studies
emanating from the work of Asch and from the work of Heider
have proceeded in independent but parallel directions.

Asch's work spawned a wealth of investigations into the
impression formation process and into implicit personality
theories adopted by perceivers. The term "naive, implicit
personality theory" was originally used by Bruner and Tagiuri
to describe the perceivers' assumptions about interrelation-
ships among personality traits. As it is more broadly con-
ceived, implicit personality refers to the manner in which
perceivers systematically interrelate attributes of other
persons in an attempt to understand why they behave as they
do (Schneider, 1973). Impression formation research, like-
wise, has focused on the process of assigning and integrating
attributes in making judgments about persons. Both research
areas can thus be seen to fall into the general class of
investigations of attribute organization.

Heider's notions regarding the observer as naive psy-
chologist fostered a large volume of scientific research
under the general rubric of attribution theory. This work
has focused on the assignment of underlying dispositions
and motives to explain social events. More recently, this
research has been expanded to include the attribution of
casuation in general and a number of studies have focused
on the assignment of dispositions to the self. That area
of attribution research which deals with the assignment of

dispositions to other people is most clearly tied to

6
concepts involved in attribute organization. For that
reason person attribution seems to be a more precise
heading for studies considered in the present analysis.

Person perception: an information analysis

 

In spite of their common roots in Gestalt tradition
(Deutsch and Krauss, 1965) and the corollary interest in
orderliness and stability in person perception, person at-
tribution and attribute organization studies have not been
systematically related. For example, Heider (1958) did
not attempt to integrate the findings of attribute organi-
zation research in his work, and Hastorf, Schneider and
Polefka (1973) did not relate the two approaches in their
well-known work on person perception.

Recently, studies which borrow from both the attribute
organization and person attribution areas in person percep-
tion have appeared (Jones and Goethals, 1972; Kanouse and
Hanson, 1972; Himmelfarb, 1975). While these studies sug-
gest specific ways in which these approaches are compatible
they do not seem to add to an integrated understanding of
person perception research.

Perhaps a beginning in the direction of relating person
attribution and attribute organization research lies in the
view of man as an information processor. This view is
common to both approaches. Specifically, both areas focus
on the perceiver's ability to "explain" information pre-
sented about others. Explanation is used here to represent
the tendency to "make sense of,", i.e., infer stable pro-

perties in, a variety of stimuli.

7

The two approaches differ with respect to the nature of
the stimulus infdrmation and the particular kind of inference
required. For attribute organization studies the stimuli are
most often descriptive adjectives or traits and the observer
is asked to infer other traits (Schneider, 1973). In person
attribution research, the person observed (actor) typically
performs some behavior from which the perceiver infers the
actor's intentions and attitudes. If the intentions are
judged not to have resulted from some external circumstance,
the perceiver may infer a basic underlying trait which is
thought to have provoked the behavior (Jones and Davis, 1965).

In the context of the present analysis, behaviors are
simply defined as something a person does, or, an act that
a person commits. Attitudes are defined as a person's affec-
tive response toward some object or concept. Traits are
descriptive words ascribed to a person.

Person attribution and attribute organization studies
thus emphasize cognitive inference processes. They contrast
principally in the unit of information given as data -
behaviors, attitudes, traits - and the unit which is inferred
--attitudes and traits. A more detailed description of
theory and research on attitude organization and person attri-
bution may provide a better understanding of the distinc-
tions and interconnections between these units.

As stated above, attribute organization research has
focused almost exclusively on traits. In a comprehensive
review of the implicit personality theory literature,

Schneider (1973) asserts that concern with the trait

interrelationships:

"may be largely an historical accident resulting

from the importance of traits in personality

theory and personality assessment research, and

from the fact that much of the early research in

the field was stimulated by Asch's impression

formation research where traits had been used

merely as a convenience."

In point of fact, Schneider contends, we have little empiri-
cal basis for considering traits as the "natural" unit of
person cognition. We presently have no empirical justifica-
tion for assuming that person perceivers "feel more comfor-
table with traits as opposed to more behavioral or typo-
logical units."

In fact, however, the ubiquity of traits in impression
formation and implicit personality research (c.f., Messick
and Kogan, 1966; Warr and Knapper, 1968; Rosenberg, Nelson
and Vivekananthan, 1968), seems too great to be accidental.
It may be the case that the importance attached to traits in
implicit personality theory follows from some special infor—
mational value which traits have for person perceivers.
Traits have an abstract quality and capacity for generality
not usually associated with behaviors and attitudes. That
generalizability makes them ideal as encoders of experience.
Assuming after Heider that perceivers seek stable, constant
properties in the stimulus environment, traits may provide
inferential stability and encoding organization to the
person perceiver.

Insight into the information value of traits is found

in George Kelly's (1955) theory of personal constructs, an

important impetus for attribute organization research. The

fundamental postulate of this theory is that "a person's pro-
cesses are psychologically channelized by the ways in which
he anticipates events." The person anticipates events by
"construing", or, placing an interpretation on replica-

tions of those events. For Kelly, the construct is an ab-
straction employed to make sense out of events, and the
examples he presents are most often trait adjectives, par-
ticularly when the situation is interpersonal.

While the question of whether traits are the natural
unit of cognition may be moot, Kelly suggests that traits
are used for encoding and mediating the experience that a
person encounters. The "interpretation" placed on future
events, where it pertains to person perception, is clearly
a mediating response which allows the person observer to
order and encode current data for the purpose of predicting
future interaction with others. Thus, if traits are proto-
typic "constructs", Kelly's arguments place them in a posi-
tion of great theoretical importance as basic encoders of
experience in implicit personality research.

While traits as cognitive units appear to be effective
encoders, their informational value as stimuli presented to
the perceiver is suspect. The very generality which enhances
traits as "conclusions" in the inference process limits their
utility as decoders or as basic data. Rodin (1972) found
that trait information was not as useful as behavioral and
attitudinal information in revealing identities among pro-
fessiorial colleagues. Mischel, Jeffrey, and Patterson

(1974) found that subjects in a prediction task preferred

10

trait information over behavioral information only when pre-
dicting to situations which were dissimilar to those for
which behavioral information was available. Traits as
stimuli therefore seem to lack the clarity and specificity
that other information units such as behaviors and attitudes
may provide.

Schneider, after Bruner and Tagiuri (1954) strongly
suggests that attribute organization research would be im-
proved by analysis of which cognitive units perceivers use
in which circumstances. The implication of the preceding
arguments is that perceivers use traits often as encoders
of person stimuli, far less often as stimulus information.
In that sense, the emphasis in attribute organization
research on traits as encoders may have been justified,
while the emphasis on traits as stimuli may have omitted
other important information units such as behaviors and
attitudes.

Behaviors and attitudes are the focus of research in
person attribution. As previously noted, the inference
generally required in attribution research proceeds from a
behavior and its effect to the actor's attitude. While the
orientation of person attribution is derived from Heider
(1944; 1958), a more formal presentation of the inference
process is given by Jones and Davis (1965).

Jones and Davis view the perceiver as proceeding in an
intent-act-effect sequence as follows: The perceiver ob-
serves the effects of an actor's behavior, determines that

the actor had both the knowledge of the effects and the

ll
ability to perform the behavior, and concludes that the
actor had some intention or attitude in performing the be-
havior. In addition, the perceiver may choose to attri-
bute some "underlying disposition" to the actor.

An example of this process would be as follows: An
observer reads in the newspaper that a freshman college
athlete has chosen not to remain in college and to sign
with a professional team. The effect of the athlete's sign-
ing is perceived to be financial security. The observer
assesses that the athlete was motivated for this outcome
and concludes that he likes or values money, at least more
than he likes the benefits associated with a college educa-
tion. The perceiver might also decide that the athlete is
financially indisposed, or alternatively, greedy and self-
seeking, depending on other circumstantial information.

While this example tends to oversimplify the process,
the emphasis placed on the behavior-to-attitude link is
clear (cf., Jones and Harris, 1967; Snyder and Jones, 1974).
In some detail, Jones and Davis discuss the conditions under"
which the observer makes "correspondent" inferences, i.e.,
attitudinal inferences which maximize the utility of be-
havioral information. In a recent revision of the theory,
Jones and McGillis (1976) assert that the correspondent in-
ference model of attribution theory has been almost exclu-
sively concerned with attitudes. There is a strong sense
in which attitudes in the attribution process perform the
same encoding function as traits in implicit personality

theory. An actor's likes and dislikes are viewed as

12
assisting the perceiver in explaining and interpreting the
actor's behavior.

There is some unfortunate ambiguity, however, regarding
the definition of an "underlying disposition" or "attribute"
in Heider's original exposition and in the Jones and Davis
model. The question concerns whether an intention is suffi-
ciently stable to be considered a diSposition or whether
such stability requires some more permanent information unit,
such as a personality trait. Jones and Davis, while clearly
focusing on the intentional inferences, cite traits, e.g.,
"dominant" and "aggresive" as examples of dispositions. It
would seem important to establish that traits may be invol-
ved in the attribution model of person perception so that
the connection between this model and the model implied by
implicit personality theory studies would be more apparent.~

There is some suggestion that dispositional attribu-
tions take the form of an attitude-to-trait inference in
attribution theory. -Jones and McGillis concede that inten-
tional inferences may not be the final step in the person
inference process, hinting that there is some more stable
inference which represents the underlying disposition. They
further note that the perceiver's decision that a given
behavior was caused by circumstances internal to the actor
is a necessary precursor to dispositional attributions. The
notion of "internal cause" requires some consideration of
Kelley's (1967, 1973) version of attribution theory.

Kelley posits that internal causes will be attributed

when the actor performs a behavior in the face of several

13

different controlling stimuli (low distinctiveness), when
other persons do not perform the behavior (low consensus),
and when the behavior is performed many times in many dif-
ferent settings (high consistency). MacArthur (1972)

found empirical support for this argument and, in addition,
found that the distinctiveness variable was most important
in determining internal causation.

Kruglanski (1975) in a major explication of the inter-
nality-externality issue, argues that the basic determinant
of the internality of an act is the inferred intentionality.
That is, if the behavior is perceived as being enjoyed for
its own sake, the perceiver will infer that the actor inten-
ded the act and therefore likes or dislikes elements associ-
ated with the act. If, on the other hand, the behavior is
seen as being performed for the attainment of some other
objective, the act is externally motivated. These arguments
can be integrated with the results of MacArthur's study. If
a behavior is performed in the face of many controlling stim-
uli, these stimuli are seen as insufficient to have caused
the act and the actor is inferred to feel some positive
affect toward elements associated with the act. Once the
attitudinal inference is made, "something" internal to the
actor may be inferred as prefacing or provoking the attitude.
The argument of the present paper is that the "something" is
very much like a trait. This trait attribution process
depicts traits as inferred from attitudes.

Unfortunately, the issue of which traits may be infer-

red from which attitudes has not been explored in attribution

14

research. In fact, it is unclear in the theory that the
perceiver does any more than make the attitudinal inference,
decide that the behavior was internally caused, and terminant
the inference process. Whether a more general cause is sought,
in the form of a trait attribution, is open to question.

In any event, the very important notion of causality
serves to elucidate the encoding relationship between the
behaviors, attitudes, and perhaps traits in attribution
theory. Heider (1944, 1958) proposed that perceivers invoke
causation as a way of explaining behavior. Thus, some atti-
tude or trait in the actor is viewed by the perceiver as
causing or evoking behavior. Jones and Davis' (1965) intent-
act-effect sequence orders intent first to represent the
fact that the attributed intention or attitude is ultimately
viewed by the perceiver as having caused the act. To return
to the previous example, the athlete's affinity for money,
or his greed, may be seen as the cause of his signing with
a professional team.’

Causation, when invoked in this fashion by the perceiver,
is clearly a symbolic response which functions to connect or
encode diverse information about the actor. Heider describes
this process in the naive observer as mediation. The attri-
bution of an attitude or a trait is therefore seen as a
mediating response between the data available to the per-
ceiver providing for a unified, stable impression of the
actor.

Mediation processes are also suggested in the inferences

required of subjects in attribute organization research. In

15

impression formation studies, subjects are typically pre-
sented with a list of stimulus traits characteristic of a
hypothetical person. They are then asked to assign to that
stimulus person values on other trait continua, e.g., good-
bad, likable-dislikable. These studies frequently examine
the effects on those ratings of the stimulus context vari-
ables such as stimulus set-size, stimulus order, and charac-
teristics of various traits in the list, e.g., centrality.
Impression formation studies thus investigate the extent

to which traits are inferred from other traits.

In implicit personality studies, mediation processes
are suggested by the methods used to obtain trait inter-
relationships. One general technique (Osgood, Suci, and
Tannenbaum, 1957) require subjects to rate stimulus persons,
e.g., physician or intelligent person, on several trait con-
tinua. From ratings across stimulus elements, the investi-
gator obtains a matrix of trait intercorrelations, which are
similarity measures. In this procedure, trait judgments are
mediated by specific person elements.

A second procedure, developed by Rosenberg (Rosenberg,
Nelson and Vivekananthan, 1968; Rosenberg and Olshan, 1970)
asks subjects to group traits which would describe a dis-
tinct unspecified person. Similarity measures are calcu-
lated based on the differences of trait groups for different
subjects. In this procedure, measures of the similarity
between traits are mediated by other traits.

Direct distance measures comprise a third general pro-

cedure. In this technique, subjects simply judge the

16

similarity in meaning of trait A and trait B. Here the medi-
ation process used by the subject is unclear. It is possible
that subjects respond to the number of behaviors which both
traits imply in common, or, perhaps, to the number of rela-
ted traits each trait has in common with the other.

Thus, for the most part, the research in attribute or-
ganization has been concerned with trait-to-trait inferences.
The concern in impression formation studies with context
effects and the centrality of various traits in trait lists
indicates an emphasis on the use of traits as mediators in
these judgments as well. In the implicit personality studies
which use correlations across stimulus persons to establish
trait inferences, these stimulus persons would appear to be
used as mediators. However, such generalized stimulus per-
sons are often themselves described using traits, i.e., "an
intelligent person." While in some direct co-occurrence
studies the mediating unit is left to the imagination of the
subject there is some indication that the subjects may be
using traits as mediators. Todd and Rappoport (1964) com-
pared direct co-occurrence methods with correlational tech-
niques and found a high correlation between trait associa—
tion measures generated from these procedures.

In summary, mediation processes are suggested in both
the person attribution and attribute organization approaches
to person perception, and seem to be likely possibilities
for the relation between the units of person cognition. In
person attribution studies, attitudes and, to some extent,

traits, mediate behavioral observation. The implication of

17

attribute organization research is that traits are similarly
invoked as mediators of trait information. A brief examina-
tion of mediation as it is more commonly studied in concept

learning may be helpful in further establishing similarities
in the two approaches to person perception.

Mediation processes

 

S-R mediational theories of concept learning (Kendler
and Kendler, 1968) propose that a mediator is a common in-
direct response to a set of stimulus instances which link the
instances to an overt response. The members of the stimulus
set need not have a common element or dimension but are
viewed as related by a single symbolic response made to all.
The process can be illustrated by a simple example. A banker,
an accountant, a lawyer, an insurance salesman, and an invest-
ment counsellor all belong to the same conceptual class -
namely, people who handle one's financial affairs -- not
because of common dimensions, though these are present, but
because one makes common responses to all of them.

The most notable use of a mediation construct in psy-
chology is Charles Osgood's (1952) formulation of meaning.
For Osgood, a "disposition" is a fractional anticipatory
goal response (after Hull, 1943) which links stimuli in the
environment to stimuli produced by that implicit response,
which in turn elicit the overt response. This representa-
tion thus divides the usual S-R paradigm into two phases.

The first is an association of signs with the mediator, or
an "interpretation". The second stage is the association

of mediated self-stimulation with overt instrumental acts,

18

i.e., "the expression of ideas" (Osgood, Suci and Tannenbaum,
(1957).

An example might serve to illustrate the application of
this process to person perception. I may meet a man who
dresses in expensive clothes, buys lunch for a large group,
and drives an expensive automobile. My implicit response is
that this is an affluent person, and I associate the above
stimuli with other "signs", e.g., he might be able to pro-
vide employment opportunities: I might therefore consider
asking him for a job.

In this illustration, "affluent" is a trait inference
or attribution based on behavioral observation. The attri-
bution of "affluent" to "ability to provide employment
opportunities" is also an inference not necessarily given
by the original behavior. It should be noted that in both
cases the mediating inferences are probabilistic and not
unique. For instance, my implicit response to the behav-
ioral stimuli could as easily have been "ostentatious", and
my overt reaction in that instance would have been to with-
draw from his company.

The major implication of the foregoing arguments is
that units of information, i.e., behaviors, attitudes, and
traits, are related through their roles as stimuli and as
mediators in person perception. Systematic relationships
between attribute organization and person attribution
approaches to person perception are clearly suggested in
terms of the mediation processes hypothesized to be used by

the perceiver.

19

Implications for a mediational model of persongperception.

 

The assumption that behaviors, attitudes and traits are
central units of information in person inference processes
is so commonly made that their identification as mediators
seems almost trite. Nevertheless, such an identification pro-
vides a basis for integrating major person perception areas.
To that end, the exposition of the units of information in
person inference and the implications of the mediation pro-
cess are crucial to a basic understanding of social cognition.

To be sure, the selection of behaviors, attitudes and
traits as distinct information units may be arbitrary.
Recently, however, some examples of that distinction have
been evident in the person perception literature. Thompson,
Phillips, and Gard (1977)1 for example, have investigated
the differences between traits and relations (attitudes and
behaviors) as stimuli in several basic impression formation
studies. Marcus (1977), in studies of self-schemata, and
Abelson (1977), in a presentation of script theory, point
to qualitative differences in social inference between know-
ing a persons' traits and a person's attitudes. Kanouse and
Abelson (1967) and MacArthur (1972), separated objective
behaviors and "subjective" behaviors, which were osten-
sibly attitudes. The attribution research and theory cited
previously though not making the distinction explicit,
certainly suggests basic differences in informational value
between behaviors and attitudes.

Certain rather obvious differences in the properties

of these units lends support to treating the units as

20
informationally distinct. Behaviors, for example, are ob-
servable, while attitudes and traits are unobservable.
Attitudes and traits must either be inferred or reviewed
in communication from another observer. In addition,
attitudes and traits seem to have an abstract property
suited for mediated associations that the specificity of
behaviors does not allow. Conversely, traits presented as
stimuli, may not provide the observer with specific infor-
mation as do behaviors and attitudes.

If behaviors, attitudes, and traits are viewed as dis-
tinct types of information, we may postulate that they are
related by the extent to which one unit mediates another.
While previous arguments have suggested that some units are
more likely to have served as a mediator in particular
research paradigms, there is no reason to expect that any
unit may always be used or that other units are never used.
It seems more reasonable instead to propose that a particu-
lar unit has some probability of "bringing to mind" or
implying another. In that case, a continuous probability
model of inference, similar to that developed by Wyer (1974),
is suited for examining mediation processes. The provision
of continuity would allow any unit to be a mediator for
any other unit while also allowing some units to have greater
chances of being inferred than others. The probabilistic
nature of stimuli-to-mediator inferences is in keeping with
Osgood's representation of a mediator summarized above.

In fact, aside from the significance imputed to particu-

lar inferences in the person attribution and attribute

21

organization paradigms, there is no reason to assume that any
unit cannot serve as a mediator for another. It seems en-
tirely possible, for example that an observer might be given
a trait judgment about a person and want to infer the actor's
attitude, or, an observer may be informed by another that a
person has some attitude, and may wish to predict that per-
son's behavior, i.e., infer what the actor would do. Indeed,
there are reasons to suspect that such inferences are quite
common. Millions of advertising and political campaign
dollars are spent inculcating positive attitudes toward pro-
ducts and candidates with the prediction that the public will
behave in a particular way, i.e., buy the product, vote for
the candidate.

In less pedestrian terms, the social psychological lit-
erature abounds with examples of hypothesized inferential
processes between traits, attitudes, and behaviors. Con-
tributions to the attitude change literature can be viewed
as an attitude-to-attitude mediation as, for example, in
congruity theory, where the attitude toward a source and
target of a message are mediated by the evaluative content
of the message. Fishbein's and Ajzen's (1975) "intention
to act" model is clearly an attitude-to-behavior inferential
model. These models obviously involve many additional
assumptions about the perceiver as actor but are entirely
consistent with an information unit analysis.

If, as has been argued, each unit stands in some infer-
ential probabilistic relation to each other unit, then the

mediation processes between behaviors, attitudes, and traits

22

can be represented as a nine-cell matrix of conditional prob-
abilities presented in Table 1. Each cell of the matrix rep-
resents the probability of using one unit as a mediator
given one unit as a stimulus. The columns of the matrix
correspond to the type of unit presented as stimuli to the
perceiver and the rows correspond to the mediating unit.
Thus, the pr (T/B) cell corresponds to the likelihood of
using traits as mediators when given behavioral stimuli.

It is clear upon examination of Table 1 that traditional
person perception research has been concerned with a small
portion of this matrix. Person attribution research has
focused primarily on the pr(A/B), and the pr(T/A) and the
pr(T/B) cells, attribute organization studies on the pr(T/T)
cell. The composition of the matrix implies that, depending
on which unit of information we receive and which unit of
understanding we require, all of the cells are potentially
involved in person perception.

This is not to say that the concentration of traditional
person perception areas on the above cells is misplaced.

This research has been based upon particular theoretical
conceptions concerning the directions of trait, attitudinal,
and behavioral inferences. The clear argument of attribu-
tion theory, in terms of the present information unit analy-
sis, is that person perception is an inductive, "upward"
process. Induction here refers to the perceiver's tendency
to use more general information units to mediate person
stimuli. Thus, in person attribution studies, attitudes

or intentions mediate behaviors and stable trait attributes

Z3

 

 

 

ne\eoca fi<\evca nm\evua menace
fie\<oca A<\<ena fim\<opa maespaau< wouauemz
fih\mvpm m<\mvnm fim\mvpm nofl>mnom

 

 

 

 

 

muwmhb mowzuwup< muow>mcom

one: m3H55a3m

cowuaoUHom comnom mo Howoe awn: :ofiumEHomnfi :<

H mam<9

24

mediate attitudes. This argument implies that the level of
generality of each unit increases from behaviors to attitudes
to traits. Stated another way, the capacity of an attitude
to summarize information is greater than that of a behavior
and a trait may have a greater summarizing capacity than an
attitude. Conversely, it should be more difficult to medi-
ate information using a specific unit since a more general
unit encodes more than a single instance of that specific
unit. That is, trait-to-attitude, trait-to-behavior and
attitude-to-behavior inferences should be difficult for per-
ceivers.

The information specificity-generality issue does not
seem to have been explicitly examined (cf. Kanouse, 1971).
However, the argument that person perception is characterized
more by induction than deduction may aid in resolving some
apparent contradictions in the results of two research areas
with similar perspectives. Fischoff (1976), has noted the
apparent inability of subjects to make sound probabilistic
judgments in standard decision theory paradigms contrasted
with subjects' facility in attribution judgments. He con-
cludes that attribution theory is primarily concerned with
explanation, decision theory, with prediction. Stated in
terms of the information unit analysis, attribution research
asks subjects to mediate behaviors with attitudes and traits.
Judgment research seems to require subjects to infer beha-
viors from behaviors (Fischoff, 1975) or behaviors and
traits from other traits (Kahneman and Tversky, 1973).

As previously suggested, the mediation of trait stimuli

25
by attitudes and behaviors should be difficult for subjects

in person inference studies. Implicit personality research
has not focused upon the pattern of attribute organization
which might result from examining trait similarity based on
behavioral or attitudinal mediation. Schneider, in calling
for "situational" mediators, has in part suggested that such
investigation has been lacking. In fact, the stability of
attribute structures found across attribute organization
studies may have been due to the fact that all of these
studies examined trait mediation processes. Furthermore,
Mischel (1973) has argued that many of the traditionally
accepted findings in the area of personality psychology
simply represent a picture of psychologists' beliefs about
which traits "go together" in people. These beliefs are
often held without actual observation of subjects' behaviors.
The implication of these arguments is that trait structures
resulting from mediation by behaviors might be very differ-
ent from those mediated by other traits.

Attribute organization studies have also observed simi-
lar dimensional structures in trait inferences. Rosenberg
and his colleagues (Rosenberg, Nelson and Vivekananthan,
1968; Rosenberg and Sedlak, 1972), for example, consistently
find that trait judgments cluster around two basic dimen-
sions: "social good-bad" and "intellectual good-bad."

While these dimensions may indeed be general to trait usage,
the present paper has argued that traits have informational
properties, e.g., abstraction, that are distinct from those

of attitudes and behaviors. In keeping with the previous

26
arguments, it seems reasonable to ask whether similar struc-

tures would be found if the inferences were of attitudeseto-
other-attitudes or of behaviors-to-other-behaviors. In
addition, basic dimensions in inter-unit inferences, e.g.,
attitudes-to-traits, might also be examined to assess the
generality of the dimensions of trait usage found in implicit
personality studies.

A final note should be made regarding the issue of des-
criptive and evaluative judgments in person perception. The
model presented thus far has made no claim about the valence
of the mediators. However, attribution literature (Kanouse
and Hanson, 1972) and attribute organization (Peabody, 1967;
Felipe, 1968) often focus on evaluative aspects of the per-
son inference process. While the model could easily be
applied to both positive and negative stimuli and mediators,
it would seem more important to demonstrate that mediation
processes are at work in positive or neutral stimuli before
more complex questions are examined. Thus, the present
study will concentrate on more descriptive inferences made
in person perception, holding evaluative inferences con-
stant.

Methods for investigptipg_inferential processes in person

perception

 

The methods used to study inference processes in tra-
ditional social psychology can be divided into basically
two varieties: mutliple-stimulus item judgments or labora-
tory experiments.

The first commonly used procedure is to ask subjects

27

to make a large number of judgments over several combinations
of stimuli of interest. In attribute organization studies,
these judgments usually take the form of trait redundancy
(Wyer, 1968) trait similarity and distance (Rosenberg and
Sedlak, 1972), or semantic differential ratings (Warr and
Knapper, 1968). The results of these judgments are typically
subjected to clustering and scaling procedures or to factor
analytic methods. These procedures result in the identifi-
cation of basic dimensions of inference. A typical criticism
of this methodology is aimed at the subjectivity involved
both in choosing the stimulus items and in identifying the
dimensional structure. The experimenter may arbitrarily
limit the kinds of inferences made or may misinterpret the
structure of those inferences. The subjectivity criticism
does not seem to be wholly justified when one considers the
previous observation that many of these studies find highly
similar patterns in the structure of the inferences. The
procedure remains widely used in person perception studies.
Another variation is exemplified by Gollob's (1974)
S—V-O model of social cognition. The experimental subject
is required to make many judgments but the structure is
provided through manipulation of stimulus arrangements
according to an analysis of variance design. While
studies using these methods eliminate much of the subjec-
tivity of the previous procedure, they are often Open to
the claim that subjects may be responding to the demands of
the experimental situation.

Some explication of the demand characteristics argument

28
is in order here. That subjects will work assiduously to

fulfill what they perceive to be requirements of an experi-
ment is a widely observed phenomenon (cf. Orne, 1962).
Experimental compliance may be translated by the subject into
behavior which conforms to the eXperimental hypotheses of
interest, to the extent that the subject detects the hypo-
theses. Evidence that experimenters are able to communicate
the hypotheses, however subtly, is well documented (Rosen-
thal, 1966). Repeated measure designs, where subjects res-
pond to items in practically all of the conditions of the
experiment, especially where the items are inferences to be
made or problems to be solve, often provide the subject with
ample opportunities to detect or infer the experimental
predictions. As a result one is never certain that predicted
findings in such an experiment reflect genuine responses to
the independent variables of interest of compliant responses
in the experimental situation.

A final variation of the multiple-item procedure, is
the construction of an artificial language (cf. Kanouse,
1971). Often fictitious persons or nonsense syllables are
given in the stimulus sentences. The subject is asked to
judge many items but is expected to be less suspicious about
the intent of the experiment since he or she spends much of
the time concentrating on the artificial language. One
could, however, argue that the provision of an artificial
language increases the suspicion of the subject and in many
respects makes the results more open to the demand charac-

teristics argument.

29

A second widely used and more traditional procedure for
investigating attribute inferences is that of the laboratory
experiment. In this type of experiment a subject is typi-
cally presented with a single, relatively detailed descrip-
tion of a behavioral situation and is asked to make a set of
inferences. The descriptions used in the study represent
factorial combinations of the variables being investigated,
and a completely between-subjects design is used. For
example, in a study by Jones, Davis and Gergen (1961) sub-
jects were asked to listen to one of four job interviews.
These interviews were with an applicant who was either
inner-or other-directed and who either conformed to role
expectations or did not.

The between-subjects design of such laboratory studies
permits the assumption that a given subject is unaware of
the other conditions in the experiment. Thus, the demand
characteristics argument is less persuasive. The number of
variables investigated is greatly limited, since larger
numbers of subjects are required than in repeated-measure
designs. Consequently, simple 2x2 experiments like the
one described, are quite common.

The inferences that the subjects are asked to make in
such an experiment may require consideration of several
other important variables which are extraneous to the hypo-
theses of interest. In Jones, et a1.'s job interview situ-
ation, variables such as the candidate's qualifications and
manner of presentation would seem to differentially affect

subjects' judgments. Typically, as is the case in this

30

study, these variables are held constant. Often this results
in highly restricted inferences of limited generality.

In addition, Schneider (1973) has suggested that the
typical tasks in attribution experiments are somewhat artifi-
cial. In the Jones, Davis and Gergen experiment, for
example, the subjects may not have been convinced that they
were listening to real job interviews but rather to a con-
trived experimental manipulation. Thus, subjects may be
more suspicious about the experimental hypotheses in experi-
ments which do not preserve a sense of task reality. In
that case, while they are unaware of the tasks performed by
other groups in the design, subjects may still conform to
expectations about their behavior.

An alternative to these methods which may remove many
of these difficulties is suggested by a procedure used in
experimental psychology. In the present study, it has been
postulated that mediation processes are used in person in-
ferences. These processes were originally hypothesized to
explain results found in concept learning tasks. One of the
most widely used concept formation tasks is discrimination
learning. The subject is presented with two stimuli, one
which is an instance of the concept and the other which is
not. Either through knowledge of results or direct
reinforcement the subject is informed of the correct member
of the pair. Choosing between a number of such pairs, the
subject learns the concept. Evidence for a mediation pro-
cess is provided if the concept is learned quicker than

would be expected by rote memorization.

31

Discrimination learning clearly eliminates the problem
of demand characteristics since the ostensible purpose of
the experiment is to make correct answers. Subjects have
little time or need for suspicion or for the detection of
experimental hypotheses. Furthermore, while some subjec—
tivity may be involved in selecting stimuli between condi-
tions, the learning task provides a straightforward, objec-
tive measure of performance in the number of correct answers
or the number of errors.

Recently, other social psychological studies have used
learning paradigms to investigate the use of rules or schema
in social cognition (Cottrell, 1976; Thompson, Gard and
Phillips, 1977)2. The discrimination learning task seems
especially suited for the demonstration of mediation pro-
cesses in person perception. In terms of the present study,
it would be possible to establish that an information unit
like an attitude functions as a better mediator than a
behavior. Such a demonstration would require that more
correct answers are made in discriminating stimuli which
differentially imply an attitude than in discriminating
those which differentially imply a behavior.

More specific to the previous arguments, the sequence
of person inference proceeding from behaviors—to-attitudes-
to-traits implied by attribution theory suggests that per-
formance on the learning task will be greater where be-
haviors are mediated by attitudes, behaviors are mediated
by traits, and attitudes are mediated by traits than where

attitudes are mediated by behaviors, etc. 'This argument

32

is equivalent to the statement that the matrix in Table l
is asymmetric, the lower triangular cells will be charac-
terized by mediation processes relative to the upper trian-
gular cells.

In summary, the present study has been divided into
two experiments. It was previously noted that attribute
organization studies seemed to identify uniform dimensions
in trait association inferences. Experiment One will examine
differences in structures of inferences for each of the nine
judgment types proposed to be involved in person perception
(See Table 1). More germane to the attribute organization
findings, Experiment One will focus specifically on similar-
ities and differences in the structure of trait associations
when the mediators vary from traits to attitudes and behav-
iors. Such an examination is expected to lend support to,
or raise doubt about, the generality of attribute structures
found in inference studies.

In Experiment Two it is proposed that if behaviors,
attitudes, and traits are viewed as information units in
the person inference process, then the units are related by
the extent to which they mediate one another in that process.
The arguments advanced and theoretical positions noted above
suggest that attitudes and especially traits will be more
likely to function as mediators in person perception. More
importantly, mediation processes, in terms of performance
on a discrimination learning task, will be more evident in
inferences which proceed from behaviors-to-attitudes-to-

traits than in a reverse direction, supporting an inductive

facility in person inferences.

Experiment One
Method

Overview of experimental procedures. A judgment task was
created for two purposes: 1) to obtain a structural repre-
sentation of inferences made between and within each type of
information unit; and 2) to select stimuli for the discrimi-
nation learning task in Experiment Two. The judgment task
was elaborately constructed to accommodate the more detailed
requirements of the discrimination learning task. For
example, it was important that the stimuli for the judgment
task be subject-generated to protect against the imposition
of a prior mediational structure on the stimuli ultimately
chosen for the discrimination learning task. Thus, while
the stimulus selection procedures described in detail here
may seem excessive the reader should be aware of the dual
selection purposes. With that in mind, the procedures are
divided into subheadings corresponding to the selection of
stimuli and the administration of the judgment task respec-
tively.
Selection of stimuli for the judgment task. The objective
of the stimulus selection for the judgment task was to ob-
tain from subject protocols ten behaviors, ten attitudes,
and ten traits which would eventually be used in making
likelihood judgments. The selection procedure incorporated
two steps for this purpose. First, one group of subjects
generated a population of behaviors, attitudes, and traits
associated with college students. A second group of subjects

supplied more behaviors, attitudes, and traits which were

33

34

implied by single units sampled from the first set. The
selection of stimulus units for the judgment task could
thus be made to ensure that an example of a given unit,
e.g., a behavior, was mentioned equally often in response
to all three units.

Twenty-four introductory psychology students partici-
pating for courSe credit were asked to write ten of each
cognitive unit--behaviors, attitudes, and traits--common to
college students. Each set was written on a separate sheet
of paper which recounted the instructions for that unit.
The following is an example of the "behavior" instructions:

"Everyone interacts with people and different people
are seen in different ways.

On the blank space (or the backside) of this page,
we would like you to write at least ten behaviors
that you might observe in college students with
whom you interact. You should begin each sentence
with 'A college student...' and describe something
that he or she does.

 

Try to think of behaviors which might be involved in
various phases of college life (dorm or apartment
living, dating, classwork, etc.) Use your imagi-
nation, but avoid writing obscure examples.

Sample: 'A college student writes on bathroom
walls.'"

The "attitude" instructions were identical to the above
except that the subjects were requested to begin the sen-
tence with "A college student likes or dislikes...," "feels"
was substituted for "does" and the word "attitudes" sup-
planted "behaviors". The example was "A college student
likes classical music."

In the trait instructions, "traits which might be

attributed to college students" were requested. The

35
sentences were to begin with "A college student 15...", and

the example was "A college student is kind."

In short, behaviors were operationally defined as some-
thing a person does, attitudes as something a person likes
or dislikes, and traits as a descriptive word. To ensure
that the order of the units requested did not systematically
effect the responses, the pages each containing behavior,
attitude, or trait instructions were presented in the six
possible orders. In all, 107 distinct behaviors, 153 atti-
tudes, and 141 traits were generated. From this population
of responses, 20 traits, 20 attitudes, and 20 behaviors were
selected at random.

In the second part of the experiment, subjects were
given a booklet containing the 20 selected units in a single
category and were asked to supply either behaviors, attitudes
or traits implied by each unit presented. The booklets in-
structed subjects to write four examples of one of the three
units --behaviors, attitudes or traits --which were likely
to be implied by the given 20 units, each on a separate page
of the booklet. The cover sheet described which of the
units the subject was to infer for the entire booklet. In
all, there were nine such booklets, one requesting four
inferred behaviors for each of the 20 attitudes, another
requesting four traits for each of the 20 behaviors, etc.
The following is an example of the cover sheet instructions
for the attitudes-to-traits-booklet.

"On the following pages are a number of attitudes

that are typical of college students. We would
like you to write four traits (usually descriptive

36

adjectives) which, in your Opinion, would be
likely to be attributed to a person who holds
each of these attitudes. The traits should
begin with 'be... .'

Sample: A college student who dislikes authority
might be likely to:

1) be anti-authoritarian.

2) be violent.

3) be funny.

4) be dissatisfied."

Thirty-six introductory psychology students partici-
pating for course credit were randomly assigned to one of
the nine booklet conditions. Booklets were administered in
several group sessions with varying numbers of subjects.
Approximately 120 traits and attitudes and 80 behaviors
were so generated. Of these, ten behaviors, ten attitudes,
and ten traits were selected as stimuli for the judgment
task.

For the most part, the ten of each unit which were
selected were mentioned often in response to each unit
category. For example, an attitude would be selected if
it was frequently implied in each of the sets of behavior,
attitude and trait stimuli. It was not difficult to select
ten traits and ten attitudes according to this criterion.
The implied behaviors, however, did not always meet the
criterion, so the experimenter was required to select three
or four which, in his judgment, could potentially be equally
implied by behaviors, attitudes and traits.

In any event, the selected traits, attitudes and

behaviors seemed generally to be characteristic of a normal

college student's experience. More specifically, each

37

selected unit seemed roughly equally likely to characterize
or not characterize any given student. Table 2 presents the
ten selected behaviors, attitudes and traits.

Administration of the judgment task. In this phase of the

 

experiment, the objective was to obtain judgments of the
likelihood that each chosen behavior, attitude, or trait
implied every other behavior, attitude or trait. The 30
selected behaviors, attitudes, and traits form the rows and
columns of a judgment matrix. The matrix is composed of
nine sub-matrices corresponding to the cells in the diagram
in Table 1, e.g., pr(B/B), pr(A/T), etc. A total of 900
separate conditional probability judgments comprise the
matrix, 300 in any row or column and 100 in each sub-matrix.

In an effort to reduce the judgments made by a subject
to an administratively feasible number, each subject was
asked to respond to one row or one column of the matrix
eliminating items which asked for the implication of iden-
tical units. Thus, each subject was asked to make all
inferences to one particular category (column) or was given
one particular category as stimuli (row), for a total of 290
probability judgments.

The items were presented on viewer terminals inter-
faced with a Hewlett-Packard 2000 computer. Items were pre-
sented one at a time after each judgment was made and no
more than three items could be seen on the screen at any
given time.

The order of presentation was randomized within each

submatrix and the order in which the submatrices were

38

cowuma
-wcmMHO can nacho moumfiuopnmm

mxoon mcflpmoyouafi moxwa
onoom maxed

waw>fla Enow mxowco
:ofiuwpomEoo moxwamfip
ovspfiHOm moumfluopmmm
:ofiucmpum moxwa
meowuocsm wauom moxwa
Hoogom whamco

 

movfipfipu/x

mcwowuso
msovaum
oumhowwmcoo
wcfixho3wum:
know:
ucoucoaoccfi
Hanna»:
uHmEm
sawcofihm

muwmhe

mucosa

xocos can mean mpowwsn
mEoHnOhm .muonuo ow mcoumwa
mnaau some m=H0n

mucmwc wcoxooz no mofiwzpm
mocwnmmmsmzo: mummy

zaoufl: Soon moumhouow

coumo moxmaoh

oawu-pnmm mxuoz

 

mpofi>mnom

xmmu pooEmosm one pom wouoofiom mpwmhp wow movSpHpum .maofi>mnom

N m4m<H

39
presented was counterbalanced.

The format for each item was as follows:

If a student X, how likely is it that he or she Y?,
where X is a particular behavior, attitude, or trait, and Y
is some other behavior, attitude or trait from the list.
For example, if the attitude was "enjoys school" and the
behavior was "works part-time", the item read "If a student
enjoys school, how likely is it that he or she works part-
time?" The "if" unit was considered the stimulus input,
while the inferred unit was considered the mediator.

Subjects were asked to respond to the items using an
ll-point scale from "0" to "10" where "0" represented
extremely unlikely and "10" represented extremely likely.
In addition, subjects were instructed that the scale values
represented the number of chances or frequency of occur-
rence in ten. That is, if the subject could think of ten
persons who were characterized by the "if" unit, the scale
value chosen should correspond to the number of the ten
who could be characterized by the inferred unit.

Sixty students who had been enrolled in upper-division
psychology courses volunteered as subjects in response to
a telephone invitation.. Ten subjects responded to the
items in each row or column of the judgment matrix yielding
20 respondents in each submatrix.1 The average completion
time was approximately 55 minutes.

Results
Using a procedure for the analysis of trait similarity

judgments suggested by Bryson (1974), mean probability

40

judgments for stimulus units were correlated across the set
of mediating units resulting in nine correlation matrices
coresponding to behaviors mediated by behaviors, attitudes,
and traits, attitudes mediated by behaviors, attitudes and
traits and traits by behaviors, attitudes and traits. A
principal components analysis was performed on each matrix
and two factors were rotated using a varimax procedure. In
the nine matrices, the median per cent of variance accounted
for by two factors was 70%. Three factor solutions were
also examined but were eliminated from consideration since
often the eigenvalue of the third factor was less than 1.0.
In addition, three factors did not seem to increase inter-
pretability.

Since structural representation in attribute organi-
zation studies pertained to trait usage, factor structures
of the traits-mediated-by-traits, traits-mediated-by-atti-
tudes, and traits-mediated-by-behaviors matrices were
examined. Table 3 presents the factor loadings for two
factors for each of these matrices.

For the traits mediated by traits matrix, high load-
ings on Factor 1 reflected an extreme social good-bad fac-
tor exemplified best by an outgoing-quiet distinction.
Factor 2 suggests more of a warmth element in another social
good-bad factor. Intellectual good-bad seemed to be reflec-
ted, if at all, in a reversal of the signs of Factor 1.

Where traits were mediated by attitudes and behaviors,
the factor structures are far less interpretable. Factor 1

of the traits-mediated-by-attitudes matrix might have been

41

cowuwuou xmefihm> no“: coMuSHom monocomsou Hmaflocfium

 

vm.N ov.v
u¢.- NH.-
ow. mv.-
oe.- mm.
ow. mo.
Hv.- mu.
OH. mm.-
No. on..
mm. Hm.
ma. vw.
mo. mo.-
NuH Hm
mpofi>mnom
muwmpu

fiOHHwHOHHOU

 

.o um.H oa.m
.. 5H.- mm.
.- Hm.- mm.-
.- mm. mm.
. mm.- mo.-
. mm. Hm.-
. mo. on.
.- 05.- Hm.
. um. co.
. Hm.- em.-
. He. ow.
mm Hm
mpwmhh

mH.~ mH
mN. ea
NH. ea
me. am
m~.- am
we. ma
ma.- no
wfl.- mm
mm. om
HN. mm
mm. we
me an

WOUSHHHH<

“H ouoz

03Hm>aomwm

mnflomuso
msowwaum
oompoufimnou
mcfixhozvpm:
pofiso

Annex
ucoucomownH
Hsmnaom
uumEm
saenofipm

use .mopSpwuum .muow>mnon mmOhom wousmaoo mooflhpms
mcofipDHOm houuwm-ozp pom mmcwmeH nouomm

ufimhuuoucfl mo

H

m mgm<9

42

interpreted as the opposite of Factor 1 in the traits media-
ted by traits matrix, except for the surprisingly negative
loadings for "independent" and "studious". Factor 1 for

the traits mediated by behaviors suggests a "suffering
servant" dimension while Factor 2 seems more interpretable
as a competence, conscientiousness dimension.

In any case, the relationship between these structures
is weak. Correlations between factors computed across
loadings for each variable were generally low, ranging from
.10 to .56. The median correlation was .27. Thus, trait
association varied greatly with the information unit used to
mediate trait inferences.

The only interpretable indication of a social good-bad,
intellectual good-bad dimensional representation was found in
the factor structure of the attitudes-mediated-by-attitudes
matrix. The first factor was characterized positively by
"likes social functions" with a loading of .96 and negatively
by "appreciates solitude" with a loading of .-85. Factor 2
had as its highest loading variables "appreciates order and
organization", .94, "likes interesting books," .93, and
"enjoys TV documentaries," .89.

Finally suggestive evidence that traits and attitudes
were more useful in mediation was found in the process of
selecting stimuli for the judgment task. Subjects seemed
to find difficulty in generating behaviors implied by other
behaviors, attitudes, and traits. The number of behaviors
written was greatly lower than the number of attitudes and

traits at each stage of the stimulus selection process. In

43
many cases, the subjects responded with traits and attitudes
when behaviors were requested.
Discussion

The most important result of this study is that factor
structure of person inferences varied greatly with the type
of information unit given and inferred. The inclusion of
attitudes and behaviors in trait inferences altered trait
associations markedly. There is little evidence in the
present study to indicate that trait-to-trait inferences
are the only, or even the most basic, inferences in person
perception. These results seem to support Schneider's (1973)
arguments regarding the need to consider situational vari-
ables and other information units in implicit personality
theory studies. Clearly, investigations of attribute organi-
zation must examine the role of attitudinal and behavioral
inferences before advancing any notion of general dimensions
in person perception.

In keeping with this argument, the results also pro-
vided little support for the generality of the social good-
bad, intellectual good-bad dimensional structure found by
Rosenberg and his colleagues. Not only was this the case
when attitudes and behaviors functioned as stimulus and
inferred units, but it was also true of trait-to-trait
inferences, the inference nearly exclusively used in inves-
tigations of implicit personality theories. The structure
of the traits-mediated-by-traits matrix seemed to differen-
tiate two social good-bad dimensions; one which could be

loosely characterized as a "fraternity" factor, and the

44
other which emphasized the considerate but submissive
nature of persons. The factor structures of other matrices,
when interpretable, seemed also to differentiate two osten-
sibly social dimensions, though not with consistently similar
content. The failure to find an intellectual dimension is
particularly curious in this study, since college students
were the persons about whom the inferences were made.

It is not reasonable to argue that a restrictive
stimulus selection accounted for this result. Several
examples in each set of information units would seem intui-
tively to have been highly correlated with an intellectual
good-bad dimension. "Smart" and "studious" were in the
trait set, "enjoys school", "likes interesting books", and
"enjoys TV documentaries" were attitudinal units, and
"studies on weekend nights", "tutors", and "reads news-
magazines" were behavioral units.

In any case, the results of Experiment One have estab-
lished the existence of differences in the structure of
person inferences which involve different information units.
However, very little can be said about the nature of these
differences. In fact, at this stage, there is little more
than the suggestion that mediation processes were used by
the subjects. Furthermore, no specific differences between
the informational prOperties of behaviors, attitudes and
traits are evident. Experiment Two is designed to inves-
tigate the nature of mediation processes used by person

perceivers.

Experiment Two
Method

Overview of experimental procedures. In order to test the
hypotheses proposed in Experiment Two, it was necessary to
construct a discrimination learning task which would test
for differences in mediation processes when behaviors,
attitudes, and traits are used both as stimulus units and
mediators. Construction of this task required several
prior stimulus selection procedures, many of which were
detailed in Experiment One. Before discussing the remaining
methods the requirements for the discrimination learning task
and the method adopted to meet those requirements will be
outlined briefly.

The discrimination learning task required subjects to
combine and differentiate various stimuli about another
person. Mediation was assumed to be at work when the sub-
ject processed the information in such a way as to make
fewer errors than would be expected by rote learning. Thus,
in constructing this task, it was first important to estab-
lish a combinatory rule and to choose an appropriate stimu-
lus set size.

A transitive or hierarchical rule seemed to capture the
theoretically continuous nature of the stimulus units. It
seemed unwise to use a dichotomous, absolute instance-~non-
instance combining rule, since the conception of the person
inference process deve10ped in this study was based on con-
tinuous probabilistic implication. Thus, a person charac-

terized in the stimulus set could be often friendly, less

45

46

often, studious, and even less often independent rather than
always friendly and never independent. Use of the transi-
tive rule required each stimulus unit in a set to be paired
with every other unit in that set. In the discrimination
learning task, a trial would be defined by presentation of
all pairs to the subject.

Six units were chosen for a stimulus set. The number
was chosen to compare to Miller's (1962) seven-unit chunk.
However, it seemed that the population of stimuli necessary
to select seven units was cumbersome, requiring far too
many items for the judgment task. Furthermore, the 21 all-
possible-paired combinations of seven units were, in the
author's experience, too lengthy to comprise a single trial.
The selection of six units seemed to allow for easier selec-
tion of stimuli and more management administration.

The final requirement of the discrimination learning
task was that the stimulus set be scaled for probabilistic
implication. The scaling would be expected to produce sic
units, monotonically ordered in terms of implication of a
single other unit. For a given unit x, and six units a-f,
prCXIa) > pr(X/b) > pr(X/C) > pr(de) > pr(Xle) > pr(le).

The probability judgment task used in Experiment One
was suited for this task and was adopted as the scaling pro-
cedure. The judgment task, it will be recalled, required
subjects to estimate the probability that one behavior,
attitude, or trait is characteristic of a person, given
another behavior, attitude or trait characteristic of that

person. The mean probability judgment would be the scale

47

value for that pair of units. The subjects performing the
final discrimination learning task were thus "decoding" a
scaled implication structure established by the judgment
subjects.

As has been noted, the initial stimuli for the judgment
task were subject-generated. Furthermore, college living
was chosen as a reasonably unrestrictive cognitive domain
for selecting the judgment stimuli. Finally, the behaviors,
attitudes, and traits selected were positive or neutral in
evaluation.

The description of the data collection procedures for
Experiment Two has been divided into two phases: the
selection of stimuli for the discrimination learning task,
and the discrimination learning task.

Selection of stimuli for the discrimination learning_pask.
As noted in the description of the requirements for the dis-
crimination learning task, the judgment task was to provide
six "input" units in each submatrix which differentially
implied one other unit, the "mediator". In terms of the
wording of the judgment items the given unit, or, "if"
statement, was labelled the input unit and and the inferred
unit was the mediator.‘ It should be noted that this choice
is somewhat arbitrary, since all possible non-duplicate
pairs were judged. A cursory examination of the correlation
between the order of mean probability judgments of elements
presented as "if" statements with the order of those same
elements presented as the inferred unit indicated that the

items were acceptably symmetric.

48

For each mediating unit, the ten stimulus units were
ranked in order of their mean probability rating. Mediating
units in each submatrix were first eliminated if the dif-
ference between the highest and lowest mean stimulus ratings
was less than three to four scale units. For each mediating
unit meeting this criterion, an attempt was made to identify
four additional stimulus units whose mean probability ratings
were relatively equally dispersed between the highest and
lowest ratings. This resulted in the selection of six
stimulus units for each mediating unit for which the stimuli
met the above criteria. A further requirement applied was
that the standard errors of the probability ratings for
these six stimulus units be approximately equal. For approxi-
mately 70% of the mediating units, it was not possible to
select six stimulus units meeting the above criteria. These
mediating units were eliminated from further consideration.

The selection within each submatrix of one particular
mediator which had met the above criteria required attention
to additional factors. Three cells in the total matrix con-
tained stimulus units in the same unit category with behavior,
attitude, and trait mediators. Thus, it was possible that the
same order of stimulus units could be observed across the
behavior, trait, or attitude mediators. Single mediators
were chosen to represent each of these three units from
within the mediator-stimulus unit sets which met the criteria
described above. That mediator was selected so as to mini-
mize the degree of overlap between the stimuli in its set

and the stimuli of other mediating units. In instances

49

where identical stimuli were chosen across mediators the
relative position of these stimuli in the order was required
to be varied.

It must be noted that corrections for overlap in stimu-
lus order were only approximately satisfied. Where possible,
mediators of different categories which ordered a given
category of stimuli were eliminated from consideration.
However, in one case, no mediator could be found which sat-
isfied both the criterion of range and dispersion of mean
probability ratings and of dissimilar stimulus orders. In
that case, it seemed more important to select a mediator
according to the ratings criterion.

In any event, it could be said that, for the most part,
the stimulus orders were unique to a given mediator across
the sample of 30 behavior, attitude, and trait mediators
used in the judgment task. While the procedures in this
study may guarantee that the stimuli could not be so
ordered for any other mediator in the sample, they do not
rule out the possibility that the stimulus orders may elicit
some mediator in the population.

Finally, in order to provide a no-mediator control
condition for the discrimination learning task, six behav-
iors, six attitudes, and six traits were simply chosen and
ordered to be dissimilar to orders established for the
behavior, attitude, and trait mediators. The discrimination
learning instruments composed of these units defined a rote-
learning baseline against which the other mediation units

were compared. Table 4 presents the selected stimulus units

SO

mom.
5mm.
vow.
mvm.
Now.
mum.

can.
va.
Nam.
mom.
cam.
men.

Nmm.
own.
owm.
mom.
mam.
cum.

.m.m

mm.m
mm.v
ow.v
ov.m
mm.m
om.n

mm.H
wm.m
HH.¢
mm.v
oo.m
mm.o

ow.m
ov.m
mn.m
mo.o
mH.n
om.n

 

ucoEmczthufiHfianOAQ :moz

mcfixuozphmg
msovaHm
Hamgfloe
unmam
ucowcomoecfi
sumo:

 

ZMHmO mmx<qmm

mcw>fia show meHGo
onoom moxwa

cofipcopum moxfia

mxoon wcfiumonopafi moxwfl
Hoogom m>0nco
cowuwuomsou moxwamwu

 

m3m2<uummo mm>HA

:oumo moxmaon

mEoHnoum .mnonuo ou mcoumfia
mocwnwmme mzo: memoh
msmsmo-mmo mo>fla

munmwa uaoxooz co mowwnum
oswu-uuem mayo:

 

wmzoz Qz< mZHB memomam

Hmmcfiumh some may mo honho thwcmum 0:» wow

mmcwump zpfififinmnohm some one .AOpwfion wopooaom some how meow msazefium

v mqm<9

51

 

 

 

 

mm~. mo. unmam
mmm. mm.m mcfixhozvhm:
mwm. om.e same:
men. oe.m kapcoflam
OAN. om.m oumnowwmcoo
owe. me.o uoflsc

oneHemmzou mmxquHn
mom. mm.~ coauqoppm moxfia
Ham. ¢N.m mcfi>fiH Show whence
mme. Nw.m meowuoczw Hmwoom maxed
NNm. mm.m Hoocum kaMGo
mmm. nv.o coflumuwcmmho pom mocha mopmwoohmmm
mom. em.o mxoon wafiumououcw moxfla

mQDBqum mme<Hummmm<
Nme. mo.e magmao ecoxoos co mowwspm
“mm. om.v xocoe can mafia muomwsn
can. mm.m osfiu-upmm mxnoz
owe. m~.o mEoHnOHQ .mnonpo on mcoumfia
emu. mo.o same“: Soon no: no we: moumuooou
mmm. mm.n manau Some mafiOM
.m.m pemamese suafiananoaa new: onezmee< mmeHa

fi.e.p:ouo e mamae

52

owm.
5mm.
Hmm.
mum.
mmm.
mmm.

mmm.
com.
Nae.
Ham.
com.
me.

mmm.
emu.
mom.
mom.
new.
0mm.

.m.m

ommu nozoa me who Hfissfium .mhopuofi ommu-nomms cw ohm whoHMMwoz "H ouoz

 

mn.e uofisc
mH.m Homecomoucw
oo.m Hyman
oe.o mcwxuozwuen
mm.o wcwomuso
o~.m oumhopfimcoo
a

oo.v menuHHOm moumflooummm
m~.m :owumuficmmho can hoppo moonwooummm
mm.m cowufipomEoo moxflamwv
mm.o coaucouum moxwa
om.n wcfi>fifi Show mxonco
mc.w oaaooa moxfla
a

oo.v :oumo moxmaoh
mw.e mnzau some mcfion
mm.m madamu-mmo mo>flH
mc.o zocos paw asap muompsn
om.n mxmm 0:3 ocozcm mHOpSH
me.w munmfi: wcoxooz :o mowwSHm
pcoamwsm xuwfiwanOHm :moz msoHnDhm

fl.e.ueOuo e mam<e

53

unaccomowcfi
unmEm
mzoflwaum
some:
wcwomuso
wowsc

mHH<mH

manomm moxfia

coflpfiuomaoo moxfiamwe

mxoon mcfiumohoucw moxfia

mcofiuu:5m HmAUOm moxwfi

cowucouum moxwa

:owpmaficmwao use nacho mopmfluohmmm

 

meDHHBH<

mocfiummmsmzo: momma

Samoan Eoou no: no we: mopwhouow
mxmm on: eschew mucosa

mnzau same mcflon

mucmfi: wcoxoos no mofiwSHm

coumo mmxwaoh

 

mmOH><mmm

soapfiwcoo fichucoo hepmfiwos-oc one now meow moasswum

m m4m<8

54

their means, standard errors, and the mediator for each
experimental condition. Table 5 presents the selected
stimulus units for the no-mediator control conditions.

Discrimination learningtask. With the addition of the

 

three no-mediator control cells the matrix illustrated by
Table l was expanded to a 3 x 4 matrix with behaviors,
attitudes, and traits as input units, and behaviors, atti-
tudes, traits and no-mediator control as mediating units.
Within each cell, a programmed learning instrument was
created on a computer which paired the six input units for
a total of 15 discrimination items. Each set of 15 items,
designated as one trial, was presented four times in
succession. The items were presented in a different ran-
dom order for each trial for each subject. The item format
for behavior and attitude stimuli was as follows:

1. The college student 2213:

2. The college student pnip.

For trait stimuli, the phrase "is known to be" prefaced
the trait unit. The following was an example of an item
in the attitude stimuli conditions:

1. The college student likes social functions.

2. The college student enjoys school.

As noted earlier, the transitive rule was to be
learned. The stimulus unit with the highest mean probabil-
ity rating visfgfzis the mediating unit was correct in
every pair and the number associated with the correct answer
was random within each cell. The instruments were, in

addition, designed for interactive administration.

55

Items were presented on a viewer terminal screen for
ten seconds. When an answer was detected, or if no answer
was forthcoming in the ten second interval, the correct ans-
wer and its number were displayed underneath the options on
the viewer screen for another ten seconds. However, the
inter-item interval could be shortened by depression of the
carriage return key on the terminal console. Thus, the
length of time required to complete the instrument could be
adjusted to suit the individual subject.

In addition to the discrimination learning instruments,
a practice instrument was designed to familiarize subjects
with discrimination learning, the transitive rule, and the
computer terminal operation. In the practice instrument,
all possible pairs of the names of five animals varying in
size were presented on a viewer terminal according to the
following format:

1. Raccoon

2. Hippo
For this instrument, the correct answer was always the smal-
lest of the animals. Ten items composed the practice ins-
trument.

Forty-four students who had been enrolled in upper-
division psychology courses responding to telephone invita-
tions and four volunteers from a graduate level business
course served as subjects. Four subjects were randomly
assigned to each cell in the experiment.

Subjects, individually upon arrival, were seated at

viewer terminals and administered the practice instrument.

56

They were asked to guess the correct answer by simply typing
the number Of the selected option on the terminal console.
Having completed the practice instrument, subjects were
requested to guess the correct dimension (size) and the
positive instance (smallest) for the animal stimuli in the
task. The experimenter then informed the subject that the
main task was similar, except that the concept or dimension
pertained to persons.

Instructions presented on the viewing screen for the
main discrimination learning task informed the subject that
the experiment was designed to determine the extent to which
subjects could discover a fairly specific piece Of informa-
tion about a person. TO that end, two pieces of information
about a particular college student would be presented, one
which was correct and one which was incorrect for that pair
only. The subject was encouraged to guess the correct ans-
wer on each item and, as in the practice instrument, discern
the requested information from the pattern of correct ans-
wers. In addition, the instructions stated that the subject
would be asked to write the information following comple-
tion Of the instrument.

Upon completion Of the instrument, each subject was
asked to supply the information requested in the instruc-
tions. Each subject was allowed three such conjectures.2
The subject was then debriefed and excused.

The major measure Of the Operation of mediation pro-
cesses was the number Of errors per trial. The Open-ended

protocols also provided measures of the identification Of

57

the concept and Of the extent to which some information
units were supplied over others.
Results

The results indicated strong support for the argument
that person inference proceeds more confidently from
specific information units to general units than from gene-
ral to specific. A significant difference in mean errors
committed on the task was found when the attitudes-to-trait,
behaviors-to-trait, and behaviors-to-attitude cells were
compared with the attitudes-tO-behavior, traits-to-behavior,
and traits-to-attitude cells, t(36) = 3.20 p<.05. The mean
total errors for each of the twelve cells in the design and
the marginal means are presented in Table 6.

In order to assess the particular effects Of behaviors,
attitudes and traits as stimuli and as mediators on learning
in the discrimination paradigm, a 3x4x4 analysis-of—vari-
ance was performed on the number Of errors committed in the
task. There were two between-subject factors: input, with
levels corresponding to behaviors, attitudes and traits;
and mediator, with levels corresponding to behaviors, atti-
tudes, traits and no-mediator control. Trials, with four
levels,was a within-subject factor.

Results indicated a main-effect for trials F(3,108) =
54.53, p<.01. Mean errors per trial are presented in Table
7. Tests for the differences between mean errors per trial
using Dunn's procedure indicated that more errors were
committed in Trial 1 than on all subsequent trials and more

errors were committed on Trial 2 than on Trial 4. Thus,

58

subjects were, in general, able to learn the discrimination
task with some facility.

More pertinent to the hypotheses Of the study, a mar-
ginally significant main effect was found for the mediator
condition, F(3,36) = 2.22, p<.ll. Subjects in the traits
mediation condition committed fewer errors than those in
the no-mediator control condition, t(36) = 2.05, p<.05.
Attitude mediators also produced significantly fewer errors
than control, t(36) = 2.03, p<.05.3 While it must be
recognized that these tests are not independent, the indi-
cation is that attitudes and traits served as mediators
in the discrimination learning task and behaviors did not.
In addition, in keeping with the argument that trait stimu-
li are less useful in the inference process, there is some
nonsignificant indication that trait stimuli procedure
more errors than attitudinal or behavioral stimuli.

Few of the subjects were able to identify correctly
the specific mediator on the Open-ended responses. NO
relationship could be found between the response of a
particular information unit on the Open-ended protocols
and performance on the discrimination learning task. Since
the stimuli could have been mediated by a concept other than
the chosen unit, it is instructive to examine the percen-
tages of subjects who responded with each of the units. In
describing the person characterized by the pattern of
stimuli in the discrimination learning task, 78% of the
subjects responded with traits, either exclusively or in

part, 32% with attitudes, 15% with behaviors, and 12% with

59

mm.na

oo.~a

mo.~H

mm.oa

eaa.~H n om:

 

 

 

 

 

 

 

 

 

mcooe Hocwwuoe

wanna msaoswum

o mam<e

V"
ww.0H ww.NH vu.mH
mN.NN oo.mH mN.OH HOHHQOU
oo.mH mN.m mh.HH muflwhh
om.mH om.HH mN.HH movSHHHH<
mm.OH mh.mH om.cfi whow>wnom
WHHNHH WQUSHMHH< mHOM>M£0m

_m

"N Opoz

"H 0902

houmwooz

woo HHOO HOOnOOm-:oozuoo some How mnoaeo Hmpoo coo:

Trials

60

TABLE 7

Mean errors for each trial

 

6.23 3.33 2.79 2.21

 

61

categorical nouns. It seems that these percentages provide

ample evidence for considering the three information units

as pervasive in person perception processes. In addition,

attitudes and traits were inferred more Often than behaviors.
Discussion

In general, the results provide support for the dis-
tinct identification Of traits, attitudes, and behaviors as
information units in person perception. In each phase Of
the present study--in stimulus selection, and in the two
experiments--differences in the subjects' use of these
units were evident. The clear implication of the present
study is that person perception processes probably involve
each Of the three units and studies in the area should
attend to the differences between them. Several findings
in Experiment Two indicate rather specific differences in
behaviors, attitudes, and traits in making inferences about
persons.

Behaviors, for instance, do not appear to be effective
mediators Of information presented about others. Subjects
in the behavior mediation condition performed no better
than the no-mediator control groups on the discrimination
learning task. The simplest explanation for this failure
seems to be that behaviors, being rather specific as infor-
mation units, are infrequently inferred from any person
stimuli and are therefore less salient. In fact, on the
Open-ended descriptions of the person characterized by the
stimulus items in the discrimination learning task, subjects

in behavior mediation groups most Often wrote attitudes

62

and traits. Apparently, the rank order of the stimuli,
although Obtained in exactly the same fashion in behavior
mediation tasks as in the attitude and trait mediation
tasks, did not elicit an appreciably salient mediator Of
any sort.

The fact that behaviors are less salient mediators in
person inference may aid in understanding some rather
puzzling findings in social psychology. Fischoff (1975),
for example, found that subjects, when informed Of an out-
come Of some fictitious battle between nations, were per-
fectly able to"explain" the outcome by drawing from typo-
logical and behavioral data. Uninformed subjects found the
"prediction" of the outcome difficult despite having the
same information available to them. While Fischoff uses
the construct Of "hindsight" to interpret these data, one
could argue that the informed subjects found no difficulty
in making inferences from the behavioral outcome to other
more general information while the uninformed group found
the Opposite inference far more difficult.

In a more methodological vein, the variable support
shown for predicting behaviors from attitudes in many
studies (Wicker, 1969) may be confounded by an experimenter's
perception Of which behaviors are implied by attitudes in
the study. Subjects in these studies may potentially per-
form many behaviors implied by their attitudes. These
behaviors may or may not coincide, however, to those
chosen, i.e., inferred, by the experimenter.

In the present study, attitude and trait mediation

63

subjects made significantly fewer errors on the discrimina-
tion learning task than the nO-mediation control groups,
indicating that attitudes and traits serve as encoders of
person stimuli. It seems reasonable to argue that the
abstract property of both units allows for a more flexible
and therefore more frequent integration of information
about others. That flexibility makes attitudes and traits
salient and useful as mediators.

It seems curious that no differences in performance on
the discrimination learning task were found between atti-
tude mediator and trait mediator conditions. Traits, as
had been previously argued from the implications of impli-
cit personality research and attribution arguments, were
thought to serve more effectively as "underlying disposi-
tions" for the person perceiver than were attitudes. It
will be recalled, however, that when the distinction between
the specificity or generality values Of traits and attitudes
as information units is made (Thompson, Phillips, and Gard,
1977;1Abelson, 1977), it is made with respect to differences
in their stimulus properties, not in terms Of encoding or
mediating properties. The non-significant trend for more
discrimination errors in the trait stimuli conditions rela-
tive to attitude stimuli conditions suggests that there is
modest justification for the stimulus property distinction
here. That either traits or attitudes are more effective
as mediators cannot be unequivocally shown in this experi-
ment.

While the data indicate no differences between the

64

relative effectiveness traits and attitudes as encoders of
person stimuli, there is some indication that traits are more
Often used as mediators. Procedurally, in the present study,
there was no Objective way to determine the actual mediator
used by the subject. In that light, it is entirely possible
that the subject formed some impression of the person charac-
terized by the stimuli in the discrimination task in trait
terms, even though the order of the stimulus set implied a
selected attitude. For example, in the trait-to-attitude
cell, where the stimuli were ordered in terms of the probabil-
ity of implication of the attitude, "dislikes competition,"
one subject wrote the trait, "shy" on the open-ended person
description. The fact that traits were written on the Open-
ended forms more often than attitudes suggests that they

were used more frequently as mediators.

In addition, when the number of errors were controlled
for the time required to complete the task, only traits
differed significantly from control (see footnote 3). More-
over, less errors were made in the attitudes—to-traits
inference condition than in any other in the discrimination
learning task, a fact which attests to the facility of that
inference. Thus, there is a strong implication in this
study that, while an attitude is sufficient to explain
behavior, it is likely that the "underlying disposition".
is ultimately encoded and stored as a trait attribution.

An important observation should be made regarding the
unexpectedly good performance of subjects in the no-

mediator control groups on the discrimination learning

65

task. While the mean number of errors made by control
subjects on trial one was not appreciably less than the
number expected by chance, i - 7.5, the errors on trial
two were reduced by half. This learning rate far exceeded
the experimenter's expectation of rote performance. In
this light, there may be slight justification for the
claim that the discrimination task, using only six stimu-
lus units, was too easy to adequately separate the per-
formances of the control and experimental subjects.

There were instead some more persuasive intuitive
indications that the control tasks were not learned com-
pletely by rote memorization. Subjects found no difficulty
in writing responses-~most often traits--on the open-ended
description. Most of these subjects seemed quite sur-
prised when informed during the debriefing that there was
no ostensible mediator to the set of stimuli in the dis-
crimination task. Apparently, these subjects did use
mediators to aid their performance.

That the subjects construed mediators was under-
standable since in the no-mediation condition as in the
others subjects were told to look for a "fairly specific
piece of information" about the person described by the
stimuli. In that sense, they were responding to experi-
mental demand. That they may have found the mediators
useful, however, may simply attest to the skillful capacity
of person perceivers to find stability and organization in

diverse and ostensibly meaningless stimuli. On the other

66

hand, the control subjects found mediators less useful than
the attitude and trait mediation subjects in the discrimi-
nation task. Taken together, these findings support the
argument that while mediators will be Often elicited, some
are more salient than others.

More specific to the methodology, the good performance
of subjects in the control groups may have accounted to
some extent for the marginality in the mediation main
effect. That is, the requirements for a true control con-
dition may not have been met. Perhaps the exclusion of the
instruction requiring the subject to look for information
about the person in all groups would provide the necessary
control. No-mediation subjects might then be less prone
to generate mediators. It is important, in any case, that
a true rote memorization baseline be established and there
is some evidence that it was not so in the present study.

Perhaps the most fundamental result of Experiment Two
is the strong support shown for a behaviors-to-attitudes-
to-traits model of person inferences. The cells in which
behaviors were mediated by an attitude, behaviors mediated
by a trait, and attitudes mediated by a trait produced
significantly fewer errors than cells in Which these units
were reversed. While this finding is not altogether inde-
pendent of the previously discussed mediation main effect,
it provides for a slightly different perspective on the
person inference process.

Although Heider (1958) never quite stated his genera-

lizations in these terms, the strong implication of his

67

model of interpersonal perception is that person inferences
are basically inductive. Induction is especially suggested
in Heider's application of Brunswik's lens model noted pre-
viously. In this sense, the unit of information used to
stabilize and organize environmental stimuli has, of neces-
sity, a more general meaning than that provided by the
stimuli themselves. Thus, for example, noting that a person
who shakes others' hands and smiles is friendly or happy is
analogous to noting that a cube composed of a cardboard is

a box.

However, caution should be taken in generalizing from
these results. The reader must recall that a single unit
was chosen as the mediator in each cell. While the find-
ings are unlikely to have been due to variance associated
with particular units, that possibility cannot be ruled
out. Ideally, each Of the twelve cells in the design
should have been tested with more than a single mediator.
Subject availability considerations restricted the number
of effects tested in the study. Therefore, the reader
should note the limitation on the generality of the results.

Nevertheless, the collective results of Experiment One
and Experiment Two have a number of important practical and
theoretical implications for research in person perception.
The next section of the paper addresses these implications
and summary comments follow.

Implications of results. The present study, in Experiment

 

Two, has specifically demonstrated that perceivers are more

facile in making inductive rather than deductive inferences.

68

In that context, it appears that the attribution research
focus on inferences proceeding from behaviors to attitudes
and traits has perhaps justifiably emphasized a more "natu-
ral" perceptual process. Fischoff's (1976) previously noted
assertion that attribution research generally finds person
perceivers to be highly proficient is in line with this
notion.

Less understood, in terms of the results of both experi-
ments, is the focus of the attribute organization research
on trait-to-trait inferences. The poor performance of sub-
jects in the traits-mediated-by-trait cell on the discrimi-
nation learning task was quite surprising. The trait-to-
trait inference has been so prevalent in implicit-personality
research that it was predicted that subjects would perform
well on the task. Furthermore, with respect to the judg-
ment task in Experiment One, the factorial structure of the
traits-mediated-by-traits matrix was dissimilar to the
structures of matrices in which the judgments involved
other units.

It seems that attitudes do serve as effective media-
tors Of person stimuli and that behaviors and attitudes
provide useful and prObably frequent stimuli in person
inferences. The clear indication in both of the experi-
ments presented here is that many findings in implicit
personality and impression formation research, in failing
to include behavioral and attitudinal judgments, may be
far less general than has been assumed. Indeed, the argu-

ment of the present study is that any inveStigation into

69

person perception processes ought to clearly identify the
particular informational judgment(s) the subjects are
asked to make and, further, to generalize only to those
specific inferences.

An additional implication of this study is that both
person attribution and attitude organization studies have
focused on a small portion of the inferences actually made
by perceivers and have virtually ignored some frequently
encountered inferences in person perception. The finding
that upward, inductive inferences are easier and possibly
more common for person perceivers does not imply that other
inferences are infrequently made. At least, the nine in-
ference modes considered here (see Table 1) should be incor-
porated into research dealing with the person perception.

Consider, as an example, the traits-to-behavior and
attitudes-to-behavior inferences Often made in graduate
school letters-of-recommendation. Trait words such as
"hardworking", "studious", and "smart", or attitudes such
as "likes the class" are Often the only pieces of infor-
mation passed on about the applicant. The decision by the
admissions committee to accept or reject the applicant
will largely depend on the behavioral implications Of those
traits and attitudes. The results regarding behavioral
mediators and trait stimuli in this study suggest that the
admissions committee would find it difficult to make an
unambiguous decision. This ambiguity may perhaps explain
the notorious lack Of utility reference letters are reputed

to have in admission decisions, both from the perspectives

70

of the applicant and the decision maker. The example
clearly points to the importance of investigating inferences
other than those emphasized in attribution and implicit
personality research.

The reference letter example also suggests an impor-
tant limitation to the present study. Subjects were told
nothing regarding the source of the stimulus information.
It was simply assumed that the subjects would accept the
information as valid and credible and there were no indi-
cations that this assumption was not met. Reference letters
are received, however, from known and unknown, liked and
disliked sources. Preliminary results from Thompson,
Phillips, and Gard (1977)1 suggest that ambiguous descrip-
tive judgments force perceivers to consider other informa-
tion such as the evaluation of the source. Inability to
mediate the trait or attitudinal information with a behav-
ior therefore requires the committee to strongly consider
an evaluation of the writer prior to the decision.

The model also may seem to be limited in that it ex-
cludes other person stimuli. Nevertheless, while the model
proposed here has focused on behaviors, attitudes, and
traits an information units, it does not rule out the
inclusion of other important stimuli such as facial expres-
sions, eye contact, and physical attractiveness. For
example, the research of Ekman and his colleagues (Ekman,
Freison and Ellsworth, 1972) has examined the extent to
which perceivers are able to encode facial expressions

into attitudes and traits. These variables were excluded

71

more for economy than for reasons related to judgments of
importance. They are clearly significant contributors to
person perception research.

In fact, integration of all findings relevant to this
behavior, attitude and trait mediation model would go
beyond the scope of a single paper. In many ways, the
model was developed for heuristic purposes. The major import
of the study is that it provides a way of integrating areas
of person perception research which have proceeded indepen-
dently despite obvious similarities. More generally, the
prOposed model suggests that most research in person per-

ception can be viewed from a single perspective.

BIBLIOGRAPHY

72

Reference Notes

Thompson, E. 6., Phillips, J. L., and Gard, J. W.
Trait and relational information in the impres-
sion formation process (in preparation)

Thompson, E. G., Gard, J. W., and Phillips, J. L.
Experimental tests of the S-V-O model of social
cognition (in preparation)

73

Footnotes

Four subjects were inadvertently administered incorrect
submatrices. Thus, the judgments of only 17 subjects
were recorded in the attitudessto-attitudes sub-

matrix and the data from 19 subjects were recorded

in the traits-to-attitude submatrix.

Originally, half of the subjects in the experimental
conditions were told the exact information unit
requested, i.e., the mediator category for the
assigned cell. The other half were free to supply
any sentence about the person. Since the latter
"uninformed" group was expected to supply more
traits on the Open ended protocols, it had been
hypothesized that their performance on the discri-
mination learning task would exceed that of subjects
in the "informed" behavior and attitude mediator
groups. It became clear, however, that there were
no differences in the patterns of the Open-ended
responses between these groups. Furthermore, no
differences could be found in performance on the
discrimination learning task. Finally, limited
subject availability reduced the projected cell sizes
to two subjects per cell. For these reasons, the
informed-uninformed manipulation was ignored.

When the total errors were covaried with time required
to complete the task, only trait mediation conditions
were significantly different from controls.

Bibliography

Abelson, R. P. Script theory. In J. Carroll and J. Payne
(eds.) Cognition and social behavior, Hillsdale, New
Jersey: 7Lawrence Erlbaum ASsociates, 1977.

 

Asch, F. E. Forming impressions of personality. Journal of
Abnormal and Social Psychology, 1946, 4;, 258-290.

 

Bender, I. E. and Hastorf, A. H. On measuring generalized
empathic ability (social sensitivity). Journal of
Abnormal and Social Psychology, 1953, 48, 503-506

Bruner, J. S. and Tagiuri, R. Person perception. In G.
Lindzey (ed.) Handbook of Social Psychology, Vol. 2,
Reading, Massachusetts: Addison-wesley PUblishing
Company, 1954.

Brunswik, E. Perception and the representative design of
psychological experiments. 2nd ed}, rev. and enl.
Befkeley: The’University of California Press, 1956.

Bryson, J. B. Factor analyses of impression formation pro-
cesses. Journal of Personality and Social Psychology,
1974, 30,‘l34¥143.

Cottrell, N.B. Heider's structural balance principle as a
conceptual rule. Journal of Personality and Social
Psychology, 1975, 31, 7l3¥720.

Cronbach, L. J. Processes affecting scores on "understanding
of others" and "assumed similarity". Psychological
Bulletin, 1955, 52, 177-193.

 

Deutsch, M. and Krauss, R. M. Theories in social psychology,
New York: Basic Books, 1965}

Ekman, P., Frieson, W. V. and Ellsworth, P. Emotion in the

human face: Guidelines for research and an inte ration
offindingg. *Elmstrderew YOrk: Pergamon’Press, I972

 

Estes, S. G. Judging personality from expressive behavior.
Journal of Abnormal and Social Psychology, 1938, 33,
217-236. —

Felipe, A. I. Evaluative versus descriptive consistency in
trait inferences. Journal of Personality and Social
Psychology, 1970, 16, No. 4, 627-638}

74

75

Fischoff, B. The silly certainty of hindsight. Psychology
Today, April, 1975.

 

Fischoff, B. Attribution theory and judgment under uncer-
tainty. In J. H. Harvey, W. J. Ickes, and R. F. Kidd
(eds.) New Directions in Attribution Research, Vol. 1,
Hillsdale,INew Jersey: ‘Lawrence Erlbaum Aesociates,
1976.

 

Fishbein, M. and Ajzin, I. Belief, attitude, intention, and
behavior:~ An introduction to theory andfiresearch.
Reading, Massaéhusetts: AddisonIWESIey Publishing
Company, 1975.

Gilson, C. and Abelson, R. P. The subjective use of induc-
tive evidence. Journal of Personality and Social
Psychology, 1965, 2} 301-310.

 

Gollob, H. F. The subject-verb-object approach to social
cognition. Psychological Review, 1974, 81, 286-321.

 

Hastorf, A. H., Schneider, D. J. and Polefka, J. Person
Perce tion. Reading, Massachusetts: Addison-WesIey
PuEIishing Company, 1970.

Heider, F. Social perception and phenomenal causality.
Psychological Review, 1944, 5;, 358-374.

 

 

 

Heider, F. The Psycholo y_of Interpersonal Relations.
New York: John W1Iey and Sons, 1958.

Himmelfarb, S. Integration and attribution theories in
personality impression formation. Journal of Per-
sonality and Social Psycholpgy, 1975, 32, 309-313.

 

Hull, C. L. Principles of behavior: An introduction to
behavior theory. New York: AppletonTCentury-Crofts,
1943.

 

 

Jones E. E. and Davis, K. E. From acts to dispositions:
The attribution process in person perception. In
L. Berkowitz (ed.) Advances in experimental and
social psychology. Vol. 2. New York: AcademiE Press,
1965.

 

 

Jones, E. E., Davis, K. E., and Gergen, K. J. Role-playing
variations and their informational value for person
perception. Journal of Abnormal and Social Psychology,
1961, 63, 302-310.

 

Jones, E. E. and Goethals, G. R. Order effects in impres-
sion formation: Attribution context and the nature of
the entit . Morristown, New Jersey: General Learning
Press, 1971.

 

 

76

Jones, E. E. and Harris, V. A. The attribution of attitudes.
Journal of Experimental Social Psychology, 1967, 3,
1-24.

/

 

Jones, E. E. and McGillis, D. Correspondent inferences and
the attribution cube: A comparative reappraisal. In
J. H. Harvey, W. J. Ickes, and R. F. Kidd (eds.) EEK
directions in attribution research. Hillsdale, New J
Jersey: Lawrence Erlbaum Associates, 1976. '

 

Kahneman, D. and Tversky, A. On the psychology of predic-
tion. Psychological Review, 1973, 32, 237-251.

 

Kanouse, D. E. Language, labelingfiand attribution. Morris-
town, New Jersey: General'Iearning Press, 1971.

 

Kanouse, D. E. and Abelson, R. P. Language variables
affecting the persuasiveness of simple communications.
Journal of Personality and Social Psychology, 1967, 1,
158-163.

Kanouse, D. E. and Hanson, L. R., Jr., Negativity in evalu-

ations. Morristown, New Jersey: General Learning
Press, 1971.

 

Kelley, H. H. Attribution theory in social psychology. In
D. Levine (ed.) Nebraska Symposium on Motivation, 1967.
Vol. 15. Lincoln, Nebraska: ‘University of Nebraska
Press, 1967.

Kelley, H. H. The process of causal attribution. American
Psychologist, 1973, 33, 107-128.

 

Kelly, G. A. Theypsychology of pgrsonal constructs. Vol. 1.
New York: W} W} NOrton and Company,—l955. --

Kendler, H. H. and Kendler, T. S. Mediation and conceptual
behavior. In K. W. Spence and J. T. Spence (eds.),
The psychology of learning and motivation. Vol. 2,
NeinOrk: Academic Press, 1968.

Kruglanski, A. W. The endogenous-exogenous partition in
attribution theory. Psychological Review, 1975, 82,
387-406. ——

MacArthur, L. A. The how and what of why: Some determi-
nants and consequences of causal attribution. Journal
of Personality and Social Psychology, 1972, 33, 171-193.

Marcus, H. Self-schemata and processing information about
the self. Journal of Personality and Social Psychology,
1977, 33, 63-78.

77

Messick, S. J. and Kogan, N. Personality consistencies in
judgment: Dimensions of role constructs. Multivari-
ate Behavioral Research, 1966, 3, 165-175.

 

 

Michotte, A. Perception of causality. New York: Basic
Books, 1963.

 

Miller, G. A. The magic number seven, plus or minus two:
Some limits on our capacity for processing information.
Psycholpgical Review, 1956, 33, 81-97.

 

Mischel, W. Toward a cognitive social learning reconcep-
tualization of personality. Psycholpgical Review,
1973, 33, 252-283. .

 

Mischel, W., Jeffrey, K. M. and.Patterson, C. J. The laymen's
use of trait and behavioral information to predict
behavior. Journal of Research in Personality, 1974, 8,
231-242. —

 

Newcomb, T. M. An experiment designed to test the validity
of a rating technique. Journal of Educational Psycho-
logy, 1931, 33, 279-289.

Orne, M. T. On the social psychology of the psychological
experiment: With particular reference to demand char-
acteristics. American Psychologist, 1962, 31, 776-783.

 

Osgood, C. E. The nature and measurement of meaning.
Psychological Bulletin, 1952, 12, 197-237.

 

Osgood, C. E., Suci, G. J. and Tannenbaum, P. H. The
measurement of meanipg. Urbana: The Univer51ty of
Illinois‘Press, 1957.

 

Peabody, D. Trait inferences: Evaluative and descriptive
aspects. Journal of Personality and Social Psychology,
Monograph,—Whole N.‘644,71967, 1, NO. 4, Part 2 of 2
parts.

Rodin, M. J. Informativeness of trait descriptions. Journal
of Personality and Social Psychology, 1972, 33, 341-344.

Rosenberg, 8., Nelson, C. and Vivekananthan, P. S. A multi-
dimensional approach to the structure of personality
impressions. Journal of Personality and Social Psy-
chology, 1968, 3, 283-294.

 

Rosenberg, S. and Olshan, K. Evaluative and descriptive
aspects in personality perception. Journal of
Personality and Social Psychology, 1970,‘l3, 619-626.

78

Rosenberg, S. and Sedlak, A. Structural representatons of
implicit personality theory. In L. Berkowitz (ed.)
Advances in Experimental Social PsychOlOgy, 3, New
YOrKYP Academic Press, 1972.

Rosenthal, R. Experimenter effects in behavioral research.
New York: ‘Appleton-CenturyicrOIts, 1966.

Rotter, J. B. Generalized expectancies for internal versus
external control of reinforcement. Psychological
Monpgraphs, 1966, 33, 1-28.

 

Schneider, D. J. Implicit personality theory: A review.
Psychological Bulletin, 1973, 22, 294-309.

 

Snyder, M. and Jones E. E. Attitude attribution when
behavior is constrained. Journal of Experimental
Social Psychology, 1974, $99 585-600.

 

 

Tagiuri, R. Person perception. In G. Lindzey and E.
Aronson (eds.) The handbook of social psycholo ,
Vol. 3. Reading, MassaChusetts: Addison-WesIey
Publishing Company, 1969.

Thorndike, E. L. A constant error in psychological ratings.
Journal of Applied Psychology, 1920, 1, 25-29.

Todd, F. J. and Rappoport, L. A cognitive structure
approach to person perception: A comparison of
two models. Journal of Abnormal and Social Psy-
chology, 1964, 33, 469-478.

Warr, P. B. and Knapper, C. The perception of peppie and
events. New York: John W1ley and'SOns, 1968.

 

Wicker, A. W. Attitudes versus actions: The relationship
of verbal and overt behavioral responses to attitude
objects. Journal of Social Issues, 1969, gs, 41-78.

 

Woodworth, R. S. Experimental psychology, New York: Holt
Publishing Company, 1938.

 

Wyer, R. 3., Jr. The effects of information redundancy on
evaluations of social stimuli. Psychonomic Science,
1968, 33, 245-246.

 

Wyer, R. 8., Jr. Cognitive Organization and Change. Hills-
dale, New Jersey: Lawrence Erlbaum Associates, 1974.

 

 

"‘lllllllllllllllllf