.fi'TRiBU‘I‘EON 0F HRATEGIES 2N THE
E’BESGNEE’S DILEMMA GAME:
A SYMBOLIC ENTERACTION APPEGAEH

muesli: far {216 Degree 02: 232% D.
M3213“ 82.35223 QN‘EE‘UVY
Paul Joseph O’Grady
I???)

This is to certify that the

thesis entitled
ATTRIBUTION 0F STRI'XTEGLES IN THE PRISONER'S
DILEuguA GARE: A SYMBOLIC iNTEiACTlON APPKOACH

presented by
Paul Joseph 'Grady

has been accepted towards fulfillment
of the requirements for

Ph- D degree inwgy

. / A’ f" I." ' I /
. ’97/6739’.j ‘ 'v ‘ 90,01 2%

‘/ Major professor I

 

DMe October 13, i910

 

0-169

 

 

k C

ABSTRACT

ATTRIBUTION OF STRATEGIES IN THE PRISONER'S
DILEMMA GAME: A SYMBOLIC INTERACTION APPROACH

BY

Paul Joseph O'Grady

This dissertation investigated the relation between cognitive
factors and subjects' choices in a standard two choice Prisoner's
Dilemma Game (I50 trials). Mead's concept of the significant symbol
was proposed as an alternative to an imitation hypothesis. An
extensive review of the literature around the Prisoner's Dilemma
revealed that the major explanation that has been proposed to account
for behavior In this game is the Rapoport and Chammah imitation
thesis. It was reasoned that if this were so then behavior observed
in Prisoner's Dilemma Games where the other was simulated was
inexplicable. For these classes of experiments it was proposed that
an explanation based on social cognitions was more appropriate. This
was supported by the fact that a number of experimenters working with
this game had offered post hoc interpretations of the subject's
behavior that Included statements about the impression of the other
that the subjects must have formed. Head's theory of social Interaction
was translated into a decision theoretic model to test this cognitive
position.

Three levels of impressions of the other were assumed sufficient
to describe the characteristics of the simulated other, as a result of

his strategy choices. A noncontingent strategy reflected an other who

Paul J. O'Grady
would be considered a cooperative martyr. Such an impression was
expected to elicit an exploitive response on the part of the other.

A contingent strategy reflected an other who would be considered
cooperative but not a martyr. Such an impression was expected to
produce more cooperative responses by the subject. A third state,
neither cooperative nor a martyr, was postulated as an error state that
would elicit a noncooperative choice. Furthermore the effect of these
basic strategies was expected to vary directly as the clarity or
frequency with which the strategy was followed by the simulated other,
increased. it was also postulated that a note, reflecting the under-
lying strategy would increase for the subject the clarity of the other's
strategy.

To test these predictions 120 subjects were assigned randomly
to one of eight experimental conditions. Subjects faced either a
contingent or noncontingent strategy at high or low levels of clarity and
with or without an exchange of notes. An analysis of variance of the
final i20 trials indicated that the subjects responded more cooperatively
to a contingent than to a noncontingent strategy. Furthermore the
strategy had more effect on the subjects' responses as.lt became clearer.
Passage of a note which was expected to clarify the strategy of the
other increased cOOperative choices across all treatment conditions
resulting in an unexpected finding for the noncontingent conditions.
in addition subjects were asked to fill out a rating scale of the other
after the thirtieth trial. A multiple discriminant function analysis
of these data revealed that subjects formed impressions of the other
consistent with the strategy he adopted. While this analysis tended

to support Mead's theory the predictions generated by the model failed

Paul J. O'Grady
to do so. A modification of the response portion of the model was
suggested to obtain a better fit to the data.

Two ancillary experiments were run using the same PDG game.
The first study found that the sex of the player was important in
determining his or her reaction to the other person's strategy. Females
cooperated more with a noncontingent other whereas males cooperated
more with a contingent other. Furthermore males c00perated more with
an other whom they believed to be a human than they did with an other

that they thought was a machine.

 

ATTRIBUTION OF STRATEGIES IN THE PRISONER'S
DILEMMA GAME: A SYMBOLIC INTERACTION APPROACH

BY

Paul Joseph O'Grady

A THESIS

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

DOCTOR OF PHILOSOPHY

Department of Psychology

To my wife, Frances, and to

my family.

ii

ACKNOWLEDGEMENTS

I would like to express my appreciation to the members of my
committee, Drs. James L. Phillips, Terrence Allen, Bertram P. Karon,
and Lawrence Messé, for the help and assistance they gave in the
preparation of the thesis. I would especially like to thank Dr. Phillips
who, in directing the thesis, provided me with the support, encouragement
and freedom to work on what i regarded as an interesting problem. i
would also like to express my appreciation to Dr. C. Kenoyer, Mr. G.
Mendelsohn, Mr. J. Maas and Mr. A. Hartman for their aid and assistance.
The research was made possible from funds provided by the National
Science Foundation (GS 2358) and by the Air Force Office of Scientific

Research (F ##620-69-C-Ollh).

TABLE OF CONTENTS

LIST OF TABLES

LIST OF FIGURES

Chapter
One- Introduction
Theories of Social Perception
Prisoner's Dilemma Game Research Findings
Experiment I: Problem and Hypotheses
Experiment 2 and 3: Problem and Hypothesis
Two- Method

Experiment I: Method
Experiment 2: Method
Experiment 3: Method
Three-Results and Discussion: Experiment 1
Results ,
Discussion
Four- Results and Discussion: Experiments 2 and 3
Experiment 2: Results 2
Experiment 3: Results
Discussion
Five- Summary and Conclusions

BIBLIOGRAPHY

iv

Page

vi

vii

3l
82
91
9!:
9A

109

HO

1M

1M

M7

161

161

164

:67

I71

180

Page

APPENDIX A - Semantic Attribution Scale l88
APPENDIX B - Final Questionnaire I9i
APPENDIX C - Letter l96

Table

LIST OF TABLES

Analysis of Variance of X Responses in Blocks of
ID for Experiment l . . . . . . . .

Group Means on those Words that have the Highest
Weights on Discriminant Function I.

Group Means on those Words that have the Highest
Weights on Discriminant Function 2.

Discriminant Standardized Weights of Two Functions
Differentiating the Eight Experimental Groups .

Parameter Estimates for the Model

Frequency of Deception Responses to Experimental
Condition .

Frequency of Reclassified Reactions to the Deception
Question.

Effect on C00peration Choices of the Various Experimental
Conditions and Subjects' Reaction to Deception.

Analysis of Variance of X Responses for Experiment 2.

Analysis of Variance of X Responses for Experiment 3.

vi

Page

Il8

i2]

123

I25

I37

143

lhh

lA6
163
165

LIST OF FIGURES

Figure Page
I PDG payoff matrix . . . . . . . . . . . . . . . . . . . 27
2 P08 Matrix with payoffs to Player 2 in upper right of
each cell . . . . . . . . . . . . . . 4 . . . . . . 29
3 Group centroids in discriminant space . . . . . . . . . i26
A Schema for changes in qn. . . . . . . . . . . . . . . . 129
S Contingent .7 no note- observed and expected cooperation l33
6 Contingent .7 note- observed and expected cooperation . l34
7 Contingent .9 no note- observed and expected cooperation 135
8 Contingent .9 note- observed and expected cooperation . l36

9 Noncontingent .7 no note- observed and expected
cooperation . . . . . . . . . . . . . . . . . . . . . . I38

l0 Noncontingent .7 note- observed and expected cooperation 139

ll Noncontingent .9 no note- observed and expected
cooperation . . . . . . . . . . . . . . . . . . . . . . lAO

l2 Noncontingent .9 note- observed and expected cooperation ihl

vii

CHAPTER ONE

Introduction

 

Theories of Social Perception

The major premise of this dissertation is that the course of
interpersonal interaction between humans can best be understood with
reference to the inferences that the interactants make about their
exchanges. Implicit in this statement is the idea that differing
cognitions are possible for the same interaction. However, given the
cognitions of any interactant, his subsequent social behavior is
assumed to become understandable. Given the cognitions of both subjects
the interpersonal interaction becomes understandable. The cognitions
are thus assumed to determine the behavior observed in the interaction.
However, there is no unique relationship between the cognition and
behavior. To illustrate, it is possible to list a set of cognitions
A, 8 and C and a set of behaviors X and Y such that when A occurs X
follows, when C occurs Y follows but when 8 occurs either X or Y may
follow depending upon circumstances. From this example it can be seen
that differing cognitions (A and B or B and C) will give rise to the
same behavior (either X or Y) or that differing cognitions (A and C)
will give rise to differing behaviors (X and Y). Thus it can be
expected that differing cognitions will give rise to the same behaviors
and also that differing cognitions may give rise to differing behaviors.
It is then possible to conceive of a social situation which gives rise
to differing cognitions among the participants, which lead to the same
behavior. It is also possible to conceive of differing social behaviors
which give rise to the same cognitions and the same behavior then ensues

on the part of the cognizers.

2
There exists evidence from several sources that people make
inferences about the behavior of others and that one person's inferences
may differ from another's inferences. It is also possible that one
person's inferences about the behavior of the other differs from an
objective description of the strategy behind the behavior. Both of
these cases are described as error in the discussion which follows.
Shure, Meeker and Hansford (l965) investigated the effectiveness

of a pacifist strategy on exploitation in an experiment where each of
two subjects was expected to communicate a five unit message through
a six unit channel. In order to succeed the entire five units had to
be transmitted. Thus on any given trial only one of the subjects could
succeed on the task. Once a subject had succeeded in transmitting he
was able to control the outcome of the remainder of the game. In the
experiment, one of the players was simulated and assumed a pacifist
strategy by always letting the subject go first and never retaliating
against the subject's exploitation. When the simulated player communi-
cated his intention a number of subjects ceased to dominate the pacifist.
Those who continued to dominate in the communication condition saw the
pacifist as trying to embarrass and deceive them. The authors concluded
that for a limited number of subjects the behavior of the simulated
other led to conversion to pacifist behavior, but for others their
resolve to dominate was strengthened by attributing trickery to the
motives of the pacifist.

' Kelley, Thibault, Radloff and Mundy (I962) found that subjects In
an experimental game situation, simpler than a mixed motive game,
exhibited more cooperation than subjects who had no knowledge of the

other person when (a) they knew they were playing with another person

3
(b) were instructed to maximize their own gain. In postexperimentai
questionnaires 2i of the 26 dyads who reached a high level of coopera-
tion as opposed to only two of the eight remaining dyads, reported
making the assumption that their partner's behavior would be stable
if he were receiving positive scores and variable if they were
receiving negative ones. The authors concluded that taking account of
the partner promoted the achievement of a mutually rewarding relation-
ship. It would seem that when very simple game situations are present
very little error in attributions can intrude to spoil the process
of solution. '

For a more complicated exchange Rosenberg and Cohen (I966)
considered the interpretation of the speaker's utterance by a listener
at any given time to be determined by a probabilistic process. This
process, called the referential process of the listener, is modified by
the listener's immediate environment, his linguistic development,
previous interactions with the speaker, perception of the speaker, and
so on. There Is then a good chance for error to intrude in the inter-
personal process according to the model. Nevertheless tests of the
listener model gave a good fit to the data collected. One hundred
and ten male undergraduates were all given the 256 sets, each set having
three words In it. Of these three words two were a synonym pair and
one was the speaker's response to one or both members of that synonym-
pair. The probability that the response elicited one member of the
referent pair rather than another was predicted to be a linear log
function of the ratio, of the response's associative strengths to each
of the referent pair. This associative strength had been determined
from an Independent study In which a large number of subjects were

asked to give the first associate that came to mind when one of the

A
referent pair was the stimulus. There was no systematic curvilinearity
between the predicted choices and the observed choices. Furthermore
the constants estimated from the data were found to be independent of
the particular associative strengths used to estimate the actual
choices. The model can be considered cognitive in that a stimulus is
compared to hypothetical referents and that referent is selected which
has a higher associative scale value. Thus communication at the verbal
level is considered by Rosenberg and Cohen to involve cognitive
inferential processes.

In a more pragmatic setting, post hoc analysis by Toch (1969)
of reported violent incidents involving members of a police department
revealed a cognitive process at work. Unlike the situations investigated
above, the cognitions and inferences arising out of the situation led
to deteriorating outcomes for both participants. Toch investigated the
arrest reports of some 3AA violent incidents that occurred over a three
year period. The Investigation Included an interview of both participants
by one of their peer group. Two thirds of these reported assaults
followed one of two sequences. In the first sequence the officer
initiated it by issuing an order, making a demand, or extending a request
which he and most other members of the community define as a legitimate
function of a police officer. From Interviews it was found that the
other person did not perceive the officer's behavior as legitimate but
considered him to be acting In a highly abusive and threatening manner.
Thus the same behavior was perceived in one manner if the person were
emitting it and another manner If another person was on the receiving
end of the behavior. The cognitions were such that both participants

considered themselves as blameless.

5

Anecdotal evidence exists as well to support this viewpoint.

On April 20, l969 a group of black students emerged from the administra-
tion building they had occupied at Cornell. As they emerged a picture
was taken of the blacks and distributed to the wire services. The
picture showed them armed with bandoiiers of ammunition and rifles
poised. Some two months later H. D. Quigg and LeonPitt (1969) of

the United Press International held an inquest into the photograph. The
president of Cornell reported an onslaught of mail and telegrams in
reaction to the photo. All of these communications expressed horror,
terror and disbelief. However Perkins noted that two negotiators who
had talked with the blacks the day before, had found them extremely
frightened of coming out into an ambush. That their fears were not
groundless is supported by the fact that the police had never previously
offered them protection, and had allowed the white fraternities Into

the building to fight the Negroes. Both the black students and white
administrators had heard of threats emanating from the community that
armed white men were reported ready to descend and terminate the black
presence at Cornell. The white backlash was thus at best incomplete and
at worst erroneous in the attributions made about the armed Negroes.

Not only are Inferences possible in human communication and
interaction but people make use of this fact when they encode their:
messages for communication. Thus Rosenberg and Cohen (1966) assumed
that the referential process of the listener In decoding a speaker's
response ls similar to the comparison process the speaker undergoes
while encoding the referent. (It must be said that the authors limit
their theory to verbal stimuli only for matters of experimental

convenience.) This similarity was exploited by Krauss, Vivekanenthan

6

and Weinheimer (1968). In an experiment using females as subjects they
instructed the subjects to find verbal labels for 2h Munsel1.colorchips.
Half of the subjects were told that the names would later be used by
the subjects themselves to identify the colors (nonsocial encoding
condition). The remainder were told that the names would be decoded
by some other person (social encoding condition). After two weeks all
subjects were asked to match each of 72 names with the appropriate color.
Twenty four of the names were the ones the subject had produced under
the nonsocial encoding conditions; twenty four names were produced by
.another subject under the nonsocial encoding condition; twenty four
names were produced by another subject under the social encoding conditions.
The communication effectiveness of socially and nonsocially encoded
messages was assessed by measuring the frequency with which names given
In the encoding session enabled subjects In the decoding session to
identify colors which had elicited them. Subjects were found to be most
accurate In Identifying colors from their own encodings. They were next
most accurate In identifying colors named by another person in the social
encoding condition. The nonsocially encoded words were least accurately
identified. For present purposes it can be concluded that subjects can
select stimuli which are Increasingly communicable when they are aware
that another person Is involved In the communication act.

In a wide range of situations, evidence has been accumulated
that the inferential processes of man play an Important role In how "
he behaves and communicates and thus what will be the outcome of hIsf
social interactions. Several theories have addressed themselves to
unravelling the seam of cognitive inference from the fabric of social

Interaction. Such a theory must indeed be complex since It involves

7
cognitions about self and the other's behavior. These cognitions
between the two people may or may not be the same because behavior is
not uniquely related to cognition. Thus the inferences for two people,
A and B, may or may not agree about person B's behavior. However the
historical trend of theories dealing with this sort of situation has
been toward greater simplification. The following explication of four
theoretical positions will illustrate this historical trend.

The first theory to be considered is that of George Herbert Head.
Head was a sociologist who was heavily influenced by Dewey's philosophy
and George Cooley's theorizing. His theoretical position could be
characterized as that of a social behaviorist. This does not mean that
he was interested only in that behavior which we regard as social. It
means instead that all human actions have a social reference. To state
it another way, distinctively human behavior is accountable only on a
social basis. In particular that phenomenon known as ”the mind” can
be understood only in terms of social behavior. Mead's theory (l93h)
regards the social behavior of man as distinctly human and as qualitatively
distinct from infrahuman social behavior. Mead feels that the very
diverse patterns of human cooperation argue against instinctual
determination of social behavior.

He postulates that the individual uses a process that (a) allows
him to ascertain the intention of the acts of others (b) enables him to
make his own response to that perceived intention. To understand this we
must define some of Head's terms. The 255 for Head is the basic unit of
analysis. It starts with an experience of disequilibrium and ends when
equilibrium is restored. Included In It are the experiences and behaviors

of the Individual and the experiences and behaviors of other Individuals

8
who move that act to completion. An act may be of short or long
duration. Moreover acts may be subsets of other acts.

The gesture is that portion of the act which represents the
entire act; it is the initial,.overt phase of the act which epitomizes
it. The imaginative or behavioral completion of an act is called
meaning. Mead says when a gesture Is presented to an animal he responds
to that gesture as such. A human, however, can postpone his response
in favor of finding out the meaning of the gesture by completing it.
imaginatively. The Individual can then be expected to respond to the
perceived meaning of the gesture. A gesture is a gignificant symbol If
it has a shared common meaning among two or more persons involved in the.
act. '

Head's theory remains general in two respects. First of all while
he defines the gesture he does not limit it to any particular form. It
could as well be any one of the following or a combination of them: a
posture, a facial expression, a verbal Intonation, a word or an action.
Second, in Mead's theory of self the individual Is able to select that
gesture, by means of role-taking, that best epitomizes his own Intention
or meaning. The more adept the person Is at role-taking the more likely
the gesture will be a significant symbol. This ability to share, or
not to share, the attitudes and Intentions of another by means of the
gesture In a social situation constitutes, for Mead, "mind.”. For him
the mind is not just what goes on In the cortex but what can develop
and appear In human social Interaction through shared meanings.

Heider (1958) has postulated that Interpersonal relations proceed
In a manner similar to Head's theory of symbolic interaction. He Is

especially concerned with Interpersonal perception, the naive analysis

9
of action and the attribution by one person to another of such states
as “can, trying, desire, pleasure" and intentions and other motivational
states. In analyzing the general phenomena of social perception he
notes that humans look for invariances in their environment. When
dealing with inanimate objects, for example, psychologists have found
that a distal stimulus such as a table can give rise to a variety of
proximal stimuli in the sense receptors and yet,the human being is
capable of having an image, or focal stimulus, that is highly correlated
with the original stimulus. This phenomenon is known in the psychology
of perception as perceptual constancy.

In an anaiagous fashion Heider finds that social perception
remains veridical despite a range of mediating social situations. The
distal stimulus exists in the Intention, motivation or ability of the
other. In general Heider refers the distal stimulus to be a disposition
of the perceived other. For example, the adjective ”friendly” can be
ascribed to an other over a wide range of social behaviors and/or
mediating conditions. Heider assumes that the human engages In such
cognitive activity in order to maintain consistency in a variegated
world of stimulus patterns.

An important part of Heider's social perception theory is the
notion that the proximal social stimulus is embedded in a pattern of
stimulation. A stimulus may be Ignored or made more salient according
to Its fit to the meaning that the total stimulus pattern presents.
Heider uses the example of a child who presents a problem at school
but not at home. The teacher and the parent In discussing the problem
may ignore the differential effect that the two situations have on the

child. The problem may be attributted to the child by one or the other

IO
adult and not to the child-situation interaction as may be more befitting.
A second point that Heider makes about the concept of embeddedness is
that in social situations it may refer to the events that transpire
over a period of time. The idea is that an individual may not be able
to attribute a disposition to another until he has seen him in a variety
of situations. An action that has no social meaning when viewed out of
its social context may become understandable when what has preceded or
followed it is made salient.

Besides the ideas of social constancy and embeddedness Heider
discusses the naive analysis of action, the attribution of I'can" and
”trying" and the attribution of ”pleasure." Since two recent theoretical
developments have placed special emphasis on these processes they will be
briefly discussed here.

For Heider the common man's problem is to attribute dispositions
to another person as a consequence of observing the other in action.

The problem presents itself In several forms. Was the other just lucky
or was his success due to his ability? Was the reason for the person's
failure due to environmental forces or was it because he wasn't trying?
Perhaps, even, his failure was due to the fact that he cannot. Perhaps
he can but wasn't trying. We see here that the attribution of “can“
and ”try” constitute a formidable problem for the common man. Once
these certain dispositions can be assigned to a social other the common
man feels he understands why a certain action came about. Once he
understands he can take necessary action to facilitate or impede the
action by the other in the future.

in analyzing the epistemological style of the common man Heider

postulates that an action can come about (or not) as a combined function

ll

of the environmental forces present, and the power or ability of the
person with respect to the action, combined with how much he tries to
carry it out. The values of “can” and ”trying” are related in a
multiplicative way while the value of the first variable can take on
positive or negative values. Very often the naive attribution is in
error. The child who fails in school may be considered to be not trying
by the teacher. The parent may regard the failure to advance as due to
environmental forces since the teacher is considered hostile to the child.
In actuality the failure may be attributed to the ability factor since
the child had not learned a crucial concept in an earlier grade. it can
be understood that if a person carries out a task either he ”can” do it
or he was I'lucky.” Then the possibility is there that he has the latent
ability. If an action is not taken we are able to say nothing about
a person's ability since he may not have been trying to perform that
action. While this represents a rational approach to the analysis of
attribution the naive approach does not limit itself to such rationality
and thus leaves itself open to error.

A second area of concern for Heider's attribution theory revolves
around the assignment of desire or pleasure to the environment or to
the person interacting with the environment. Thus a person A upon
reading a book may say that a book Is not enjoyable. A person 8 upon
observing person A may say that the latter does not enjoy reading books
since he knows for a fact that several others have feund this same book
enjoyable. Person A attributes the enjoyable experience to the environ-
ment whereas person B believes the experience is due to the person
reading the book. If on the other hand, person A enjoys every book he

reads, 8 would accuse him of lacking taste or of being undiscrlminatlng.

12
In other words 8 will perceive A to be the source of the enjoyment
rather than the books.

To account for such attributions Heider states that as long as
the enjoyment is closely connected with the presence and absence of the
object then enjoyment will be attributed to that object. This is the
principle underlying Mill's methods of experimental enquiry. According
to Heider this method is the basis of attributing action to the subject
and/or to the environment. Heider points out that perception is
attributed to a person when he makes a differential response to the
presence and absence of the object or event in the environment. This
is true for both naive and scientific psychology.

The person, rather than the object, is regarded as the source of
enjoyment by any neutral observer under one of two conditions. First
if the person is observed to not always enjoy the object at different
periods of time then it is impossible to assign enjoyment to the object.
In psychometric terms one can say that this condition exists if a person
has low internal consistency. Then the enjoyment will be attributed
to him or his transient personality states. Likewise If a neutral
person has knowledge of a group of others' reactions to the object or
event he can test a particular person's reactions against this norm.

If it differs markedly, then the neutral can infer that the person Is
the source of the enjoyment or nonenjoyment. In psychometric terms
there Is little interpersonal agreement.

The difference for Heider between the scientist and the naive
observer In the attribution of enjoyment is that very often the latter
type of person relies upon a minimum data pattern to assign enjoyment.

This pattern is defined as one Instance of joint absence of the object

I

l3

and the effect, and one instance of their joint presence. Generally
the common person does not postpone attribution until a series of joint
condition-effect changes occur. Consequently one finds that people
expect they will enjoy again what they have enjoyed on one occasion or
else expect others to enjoy what they themselves enjoy. Heider refers
to the latter error as egocentric attribution.

it was stated earlier that theoretical developments in this
cognitive social approach have tended towards greater simplification.
Two of these recent developments have used Heider's psychology of inter-
personal relations as the starting points for their own formulations.
Despite this common heritage the two latest theories outline widely
differing approaches to the problem of attribution by human beings.
While this illustrates the heuristic value of Heider's theory It also
points out that he did not spell out a univocal and final model. The
most plausible explanation for this is that Heider placed great value on
the comprehensiveness of his theory and consequently analyzed a wide
variety of every day human situations where he felt the attribution
process was in effect. The interrelationship of these situations is
left unspecified except by virtue of the similar processes at work.
Jones and Davis (I965) investigated how a person or observer Infers a
variety of dispositions of another from his actions. Their emphasis
is upon delineating the necessary and sufficient conditions for assigning
the correct disposition. They assume that a perceiver seeks to find
sufficient reason why another person acted as he did and why that
action took on a particular form. This necessitates that the perceiver
establish links between stable individual dispositions and the observed

action of the perceived person. It also entails finding out which of

IA
the actions that finally do occur were foreseen by the person emitting
the action. Kelley (1967) in his theoretical analysis pursues a more
formal investigation than Heider's original investigation into the
area. At the same time Kelley is interested in applying several of the
concrete examples that Heider used, such as ”enjoyment and trying” to the
analytic scheme he proposed. The problem again deals with an observed
effect such as an accomplishment or an enjoyment which occurs when an
individual is present and interacts with a social object such as a book,
a task or another individual. A second person, or even the Individual
himself, is assumed to try and understand the reason for any particular
effect. In the simplest case this involves partitioning responsibility
for the effect between the environmental social object and the individual.
In particular Kelley has tried to determine how the naive person assigns
an effect such as enjoyment to an object in the environment or to the
person manifesting the enjoyment.

Jones and Davis define several terms in their presentation of
attribution processes. An gg£_is a molar response which reflects some
degree of personal choice on the part of the actor and which has one
or more effects on the environment or the actor himself. Effects are
distinctive consequences of actions. An act may have any number of
effects some of which may be in ggmmgg_to other acts that could have
taken its place. Inferences are made on the basis of noncommon effects.
Correspondence refers to the extent that the act and the underlying
characteristic or attribute are similarily described by the Inference.

The process of attribution Is assumed by Jones and Davis to start
with a person, A, seeking to find why another, 8, acted as he did. This

means that A has to establish a link between the Individual B's

IS

disposition and the observed action. Prior to this he must make sure
what effects, of the ones following from an action, were intended by
the actor B. In order to attribute this intention to another the
perceiver must assume that the actor was aware of the observed effect
his action would have. in addition Jones and Davis assume that the
actor has the ability to bring about the effect or that it was not
accidental. Once the intention of the act has been determined inferences
can be made about underlying dispositions. The meaning of an action or
its intentional significance can be understood by listing the effects
it produces against the effects all other possible acts could have
produced. That is, meaning becomes clearer as the number of unique
effects associated with an act decreases. Correspondence is then
postulated to be related inversely to the number of noncommon effects
and inversely to the assumed social desirability of these effects.
The latter social desirability factor relates the action to role behavior.
Therefore little information is gained about a person's dispositions
when the actions he performs are dictated by his role. To recapitulate,
the fewer the number of noncommon effects associated with an action
the more likeiy_the correct Intention will be inferred from the action.
If the action is socially desirable, i.e. if everyone wishes to perform
it, then no information is gained about an actor's unique characteristics.

Hedpnlc relevance means that the choices of the actor have some
positive or negative consequences for the perceiver. Jones and Davis
(1965) propose that as relevance Increases there will be an Increase
in the likelihood that Inferences will correspond because It reduces
the number of unrelated or noncommon effects. Furthermore they assume

that If the consequences of an act are predominantly positive the

l6
perceiver will be more favorably disposed toward the actor. If
consequences are negative the converse holds.

The inclusion of the term "hedonic relevance” introduces
attribution processes into the framework of social interaction. From
time to time Heider (I958) refers to social actions that can be taken
as a result of attributions about an actor formed by the perceiver.
However there is no psychological process outlined whereby the attributions
and the action are connected. Kelley's (I967) reformulation of Heider's
theory is also not tied to social action as much as it deals with social
perception. (Jones and Davis (1965) omit the criterion of temporal
consistency alluded to by Heider and explained by Kelley.) Kelley
focuses on the problem of assigning causality to the self or to the
object in the environment. This is closest to Heider's environmental
and personal forces. Jones and Davis on the other hand analyze the
problem from the point of view of discriminating a person's characteris-
tics from the intentions of his act which in turn are inferred from the
consequences of his actions. This is closest to Heider's description
of the assignment of ability or motivation to another. To complete this
theoretical review Kelley's theory of attribution will now be explained.

For Kelley, attribution refers to the process of inferring or
perceiving the dispositional properties of entities in the environment.
These properties can constitute any stable feature of any object.
Kelley's theory will be discussed using social situations as the objects.
Some of the features one might prefer to look for in another person In
social Interaction is his Intention, what he desires and feels or what
his basic ability is. The Kelley analysis, following Heider, assumes

that peOple tend to achieve an active mastery of the events in their

17
environment and, to do so, use a naive version of Mill's method of
difference as their basic analytic tool. This method says that an ""
effect is attributed to that social object which is present when the
effect is present and which is absent when the effect is absent.
Kelley then uses a data cube to explain the detailed workings of this
law. Along one dimension of the cube entities or social objects are
placed. Along a second orthogonal dimension a time-modality factor
is placed.. Along yet a third factor observers or persons including
one observer called the self are placed. As defined by Kelley the
problem for the self is to resolve whether a particular effect resides
in the social object or in the self. The self solves the problem by
noting whether the effect occurs when a particular object is present
but the effect does not occur when some other objects are present.
The effect is deemed by the self to be caused by the entity if other
observers respond to it in the same way. Finally the effect is attributed
to the distal object if the effect continues to be present when the
object reappears over time or In different modalities. To sum up, the
attributions to the social object require that the self respond
differentially to the social object but that this response be consistent
over time and modality. Furthermore this response must be consistent
with the other person's responses to it. As an example, Kelley IT“
examines the enjoyabillty of a movie. The problem for an observer Is
to ascertain if a movie is Inherentiy enjoyable. The sine qua non for
such an attribution to be made Is that the observer finds not all

movies enjoyable. Also the observer must find the particular movie

enjoyable If he sees It on television or If he sees it a second time.

i8
Finally other people must be seen to appreciate the movie in the same
way. If these conditions are not met the attribution is not made to
the distal object but to the self or chance.

Jones and Davis' theory is recast by Kelley into his data
cube. An inference of intention to another actor is made when a
particular observed effect produced by another observer, called actor,
follows certain conditions. First of all the demand characteristics
of the situation (i.e. that time or that modality) cannot account for
the effect. That is, other observers (including self) would produce
differing effects given the same situation. This is Kelley's inter-
pretation of Jones and Davis' social desirability criterion. A second
condition for attributing an intention to the actor is the observation
made by the self that the actor could have produced other effects in
this situation. That is, under different situations, the actor has
been observed producing several effects. The inference can be made
that the actor has the ability (as Jones and Davis conclude) to produce
several effects. Finally In order to infer an intention on the part of
the actor, the effect must be constant across differing entities for
any given social situation.

From this analysis Kelley makes the point that the self may not
actually need to examine variations In behavior over a number of other
observers. From his prior knowledge of social pressures, shared values,
and situational demands he may be able to make consensus estimates.
These estimates may be correct or Incorrect.

Kelley's distillation of Heider's attribution theory has been
discussed, first around the conditions necessary for any attribution

to be made and second, as to how the Jones and Davis' version fit Into

l9

Kelley's data cube. Kelley also introduces systematic errors that
can be made in attributions. According to Heider, errors can be
traced to instances in which first, the relevant situation is ignored;
second, egocentric assumptions are made; third, the relevant effects
have affective significance for the observer; and fourth the surrounding
situation is misleading. '

Jones and Harris (1966) found that subjects ignored the relevant
situation (Heider's first error category) by attaching too much significance
to the behavior and its effect and too little to its situational context.
Subjects observed a person express an unpopular and unexpected opinion.
The opinion emitted by the person affected subjects' judgments of his
true opinion even though the emitted opinion was elicited by strong, '
legitimate external pressure. Kelley also included in Heider's first
error category another type of situation where the causal complexity
of the situation may be such that the relevant causal factors are not
ignored but are simply not perceptible. Kelley presents the example of
a person who invests in stock during an inflationary economic period.
The true causality Is obscure enough that he perceives himself to have
the characteristics of a wise Investor.

Errors due to egocentric assumptions constitute the second bias
process. These are Important when the evidence for the attribution
Is not complete. The simple presence-absence assumption of Mill Is
simply not enough to ascertain causality. The example of violent
interactions used earlier, Illustrates well the egocentric bias. In
such situations according to Tech violent people held premises about
others that promoted the violent interchange. Very often this premise
Is Invoked under minimal Information conditions as the next example

will show.

20

The inquest into the Cornell racial incident reveals the second
type of error process at work. It would appear that because of
ignorance and/or egocentricity, the white backlash to this picture was
caused by an erroneous attribution to the black students. As pointed
out in the earlier discussion of this incident whites who reacted to
the photograph were unaware of rumors that circulated on campus that
the Negroes' lives were in danger. The reaction also ignored fact
that the police had allowed fraternity members into the building to
beat up the Negroes. Only those administrators who talked with the
Negroes were aware of the impact these two facts had on the Negro
demonstrators. The reaction to the picture also involves the third
type of error in attribution processes.

Fear was Involved on both sides. This fear dominated the whites'
perception of the picture and at the same time the blacks fear of a
minority of whites dismissed any perception of protection that the white
power structure was prepared to give them. According to Heider's third
source of error the magnitude of affective consequences Involved in
an attribution may bias the inference process. For Instance, criticism
directed towards an actor by a group of observers may be attributed
by another observer as the fault of the actor. The actor, as observer,
may attribute the common treatment to a conspiracy against him. Ego
protection of this latter sort was found by Johnson, Felgenbaum and
Weiby (196A). Their subjects attempted to teach a nonexistent student
some arithmetic. Before the teaching began, the student was represented
by a set of very poor work sheets. The student was then given a unit
of Instruction by the subject-instructors and observed to Improve by

some Instructors, but for others he continued to do badly. The former

21
instructors tended to take credit for themselves whereas the latter
tended to attribute the poor performance to the student.

Subjects have been deliberately misled by the experimenter in
order to achieve a fifth kind of attributional error. Kelley discusses
this type of error as separate from Heider's four categories. To use
this situation the subject's attention is diverted away by the experi-.
menter from an important causal factor. Thus Schacter and Singer (l962)
created a state of physiological arousal in subjects through the '
administration of a drug but misrepresented the causal link between the
drug and the physiological effect. The subject then attributed his
physiological arousal to the situation and took his cue as to the nature
of the situation from another person who was responding to the same
setting. But the social model has this effect upon the subject's
behavior only when the subject can find no reasonable explanation for
his arousal state. Another instance of this type of error Is presented
in a study by Valins (1966). The stimuli were a set of slides of
semi-nude females. The subjects at the same time as they were viewing
the slides could ostensibly hear their own hearts beating as they viewed
the semi-nude women. In actuality, variations in heart rate were
coordinated by the experimenter. Ostensible heart rate was Increased
(or decreased) for half the slides while for the other half of the
pictures the heart rate remained the same. Subjects rated as more
attractive the pictures with which changes In heart rate had been
associated and more frequently chose them as remuneration for their
experimental participation.

While the foursome of Kelley, Heider, Jones and Davis have

detailed a theoretical realm that promises to shed light on a number

22
of experiments and social situations, there is some reason to believe
that their theory is not all that well formulated. Kelley has used
experiments to illustrate one error process but these same experiments
could just as well have Illustrated other error processes. Thus
the Valins findings could just as well be a result of Heider's third
error process as well as Kelley's fifth error category. That is, Valln's
experiment contained a highly arousing or affective component in that
males were viewing pictures of females who were seminude. Perhaps If
this condition were not present there would be no chance for subjects
to make errors due to the deceptions of the experimenter. Also errors
made by ignoring the relevant situation (Heider's first category) are
most likely to occur if the magnitude of affective consequences Is
increased. Yet Kelley does not discuss this. Also Kelley does not
relate the error process to the attribution process. For example,
for some subjects group consensus Is effective in inducing misattribu-
tion as the Asch studies have shown. Thus the group consensus can be
a source of bias. At the same time Israel (I963) has found that some
subjects evoke a reference group against the group consensus in order
to maintain veridical attributions in the Asch conformity situation.
Now this would suggest a fourth dimension in Kelley's data cube. Yet
Kelley allows ”shared values, social pressures and Institutional
pressures” to take the place of the group consensus, when no group
consensus Is actually present. In conclusion Kelley's formulation of
attribution processes while being valuable is Incomplete. The practical
and theoretical Importance of such processes for human beings justifies
the Introduction of other theories which will hopefully clarify the

process by theoretical dialogue Informed by experimental evidence.

23
It can be seen that Mead's account of minded behavior has much

in common with the attribution theorists. It deals with attribution
processes and with sources of error in those processes. It recognizes
a time dimension (as does Kelley's) in the form of the social act which
must occur over time. It recognizes as does Jones and Davis, the import
of attributing intentions to the other person on the basis of observed-
actions. It places more emphasis on the social nature of the act in that
cooperation among humans is seen as the central feature of human
existence both in their thought processes (thrOugh shared meanings) and
by achieving their mutual goals. This cooperation.facilitates the
human being. Jones and Davis introduce the social nature of the social
act by way of the term ”hedonic relevance.” To them the fact that the
actor can facilitate or impede the actions of another (the self) is as
likely to cause error as it is to help the social act.

Social facilitation is introduced in yet another way by Kelley.
He remarks about the human being brings to a situation values or
expectations shared by others in his reference group. He uses this norm
to define the situation. For Mead when the situation Is the behavior
of another and that behavior is used in a self-conscious manner, the
actor can count on the other person sharing the Intent or meaning of
the gesture and of reacting to that intent rather than to the gesture
itself.

Head's formulation implicitly details sources of error in the
social Interaction. First of all one of the parties can be holding a
”conversation.of gestures” as Head terms it. In this situation one
person can respond simply to the gesture of another without regard to

the other's Intent. This source of error Is not mentioned by Kelley

2h
since Heider's theory and its derivatives do not consider such types
of interaction. Furthermore it doesn't seem to be of major importance
in Head's theory.

A second source of error would seem to be in the way the
participants define the act both in terms of its start and finish and
also of who is participating. For example the psychological experimenter
recently has discovered that he has been making this error rather
consistently. Experiments by Orne (1962) and Rosenthal (1966) showed
that the experimenter was included into the act in a reliable way by
the subject whereas the experimenter had traditionally seen himself as
a neutral object. This source of error parallels Heider's first error
as set forth by Kelley. That is, some participants Ignore the relevant
situation. It may be recalled that the experiment, used to illustrate
this error, involved a third person who commanded an unpopular opinion
on the part of the speaker. The speaker's true opinion was seen by
the subject to be related to his emitted opinion. Theellcitatlon by
the third person was ignored by the audience.

The third and major source of error covered by the Meadian frame-
work is the lack of a significant symbol. This can occur In two ways.
First, the meaning of the gesture may not be the same for the two (or
more) parties. That is, because of different upbringings or subcultures,
the participants of the social act Interpret the gesture In different
ways. The second source of error can be Inherent In the gesture itself.
Because of the complexity of the act or the limitations of communications
relative to the act the gesture cannot convey the meaning fully. That
Is two or more meanings are possible for a given gesture. The single
occurrences of the gesture leave the observer open to Infer one or the

other meaning and perhaps both.

25

Within this third error class, the first type of error is
equivalent to Heider's errors due to egocentric assumptions. The
second type of error in this class is related to Heider's first type
of error. Kelley states that ”in complex causal environments, the
relevant causal factors in the situation are not ignored but are simply
not perceptible to the person.” (Kelley, 1967, p. 219).

While Head's theory seems more cohesive than does Heider's at
the same time it remains unspecified in some areas. It does not partial
out of all behavior the portions of behavior it wishes to consider as
leading to an inference process. Jones and Davis study behavior in
terms of its effects on the environment. Head, on the other hand,
includes within the class of gestures such behaviors as grimaces,
postures, gazes etc. The central importance of such stimuli for social
interaction has been rather well documented (Duncan, 1969). However
their relationship to inferential processes has not been as well
defined as it has in the case of the effects one human being has upon
his environment and/or upon another human being. Given these postulated
weaknesses Mead's approach Is not easily reduced to a standard laboratory
situation. However, combined with Heider's formulation and Jones and
Davis' adaptation a list of the characteristics of the Ideal experimental
test situation can be drawn up.

In search of a laboratory situation. To test Head's formulation
of significant symbols a situation Is required In which the following
criteria are present:

(I) The social aspects of the situation must require the mutual
Interaction of at least two people.
(2) The situation must be well defined In terms of the number of people

who are participating and In terms of the onset and completion of the

26
interaction. This is so in order to eliminate the second class of
errors (that is, the number of people participating in the act) that
Mead's theory postulates.
(3) The situation must be of such a nature that the subjects are
likely to engage in reflective thought processes. Essentially this
means eliminating Head's first class of errors, what he calls ”a
conversation of gestures.”
(A) The situation must be of such a type that the gestures are ambiguous.
That is the gestures are not totally significant symbols using Mead's
terms.
(5) For methodological reasons the situation must be such that the
possible significance of the symbols is well defined. This means that
the number of possible gestures must be limited and the significances
attached to these gestures must be well articulated and very few in
number.
(6) The gesture, again for measurement reasons, must be made In one
mode or dimension. That is, If facial expressions are to be used, then
verbal behavior, postural gazing and mutual reinforcement cues must
be controlled.

The Prisoner's Dilemma Game (PDG) (Rapoport and Chammah, l965a)
fits all of the qualities mentioned above. To verify this, the game
will be explained and then the compatibility of the game with the
above criteria will be demonstrated.

The PDG can best be described as a game played by two people
over an extended sequence of discrete trials which may or may not
have meaningful consequences to the players. Each person has two

choices (usually called C for cooperation and D for defection). On

27
any given trial four outcomes are possible: CC, CD, DC, DD (where
the first letter signifies player 1's choice and the second signifies
player 2's choice). Each of the outcomes is associated with a different
reward to both players as determined by the matrix in Figure l. The
payoff to player I is in the lower left hand of each cell and that
to player 2 is in the upper right hand of each cell. Thus a joint
DC choice would result in a payoff of T units to player i and a payoff
of S units to player 2. The values of the cell entries can be varied
in amount but the following inequalities must hold to maintain the

structure of the PDG.

 

 

 

S<P<R<T (I)
2R>S+T (2)
Player 2
C D
‘\~ R‘\\\\ T
c \\\\ l .\\ .
“\\ \‘\\
Pla er 1 R \\\\ S
Y ‘ .
‘-\\ s"\\\\ P
o \‘x \‘\\
\ \
T P S

 

 

 

Figure l. PDG Payoff Matrix

R is called the reward for cooperating; P is the punishment for
defectlng while T.Is the temptation to defect; 5 Is the “suckers”
payoff for cooperating while the other player defects (usually S and
P are negative). Assuming that each player wants to win as many
points (or cents) as possible and given the two Inequalities above,

the mixed motive nature of the game becomes clear. If both players

28

want to get.the highest possible payoff, T they will both play D and
both receive P, a mildly punitive outcome. If one persists In choosing
C he leaves himself open to exploitation by the second player who

will go after outcome T by choosing C. If the first person retaliates
by choosing D after he has been exploited the dyad runs the risk of"
staying in the DD cell. Such a result can only lead to both players
receiving nothing for their efforts (either in money or points).

Under the restrictions of these two Inequalities and the
assumption of symmetry, R, S, T, and P are free to take on a wide
range of values. Rapoport and Chammah (l965a) looked for an Index
to describe the relationship between the incentive value of the payoff
matrix and the course of game behavior. Their first conjecture resulted
in the following hypothesis. If other payoffs are kept constant, C
increases as R and S increase and decreases as T and P increase. If
the entries of the payoff matrix are thought of as utilities then C
(the frequency of cooperation) depends not on the individual values
but on the ratio of their differences. For three variables one ratio
suffices but for four variables two ratios are needed to describe
the relationship. The authors found the one ratio that best describes
the incentive value of the matrix In terms of frequency of cooperation
in the various games to be given by Equation 3.

CI-R-P

T-S I (3)
The particular PDG payoff matrix to be used in the present study

Is shown in Figure 2.

29

Player 2

\ l¢ \ 2:

IC \ 'ZC \
Player l ::g -2¢ -I¢

 

 

 

2¢ -i¢ \\‘\\

Figure 2. PDG matrix with payoffs to Player 2
in upper right of each cell.

 

 

This matrix has a CI-.5 which is a little larger than the CI used by
Terhune (1968) and much larger than the CI used by Pruitt (l967).

From this explanation and considering some other findings about
the PDG we can move to an examination of the six criteria outlined
above for a suitable experimental situation in which to test Mead's
theory.

The first criterion demanded that the situation require the
interaction of two or more people. This is true of the PDG since,
before any reward can be received by either, both persons playing the
PDG have to respond. Furthermore In order to attain a profitable
outcome both have to learn to cooperate by choosing C.~

The second criterion related to the definition of the situation
poth insofar as the number of people Involved is well known and
insofar as the beginning and ending ls defined for people. The PDG
is presented as a game for two people and the onset and finish of

each trial Is well established just by the features of the game

itself.

30

The third criterion was concerned with selecting a situation
in which the thought processes of subjects were likely to be operative.
For the PDG, the experimenter can increase the likelihood of such
thought processes occurring by making increased financial reward
contingent upon apprehending the strategy of the other player. That
players pay attention to the other person's strategy and behavior
in bargaining games has been documented by Murdoch (l968). He found
that the variance accounted for by social factors in a bargaining
type game explained some 50% of the total variance. Personality
factors explained only 30% of the total variance. Murdoch's analysis
came from a study by Druckman (1967). Murdoch concluded that situational
as opposed to personality variables should be studied in bargaining
situations. It can be expected that variations in the other person's
behavior in a PDG will have a large influence on subject's performance
over and above personality differences In the subject. In addition
Kelley et al (1962) in a similar game type situation to the PDG
found from post experiment questionnaires that subjects often engaged
in strategy assignment to the other player during the game.

The fourth criterion concerns the ambiguity of the gesture
either on the part of the stimuli used or on the part of the encoding-
decoding process. McClintock and McNeel (l966) have analyzed PDG
choices In terms of the ambiguity they pose for the experimenter in
inferring the motivations of his subjects. A C choice by any player
can mean that he wants to maximize joint game. A D choice can mean
that he wants to get the greatest possible gain for himself or that
he wants to minimize his maximum possible loss. There Is no reason

to assume that subjects do not face the same dilemma when they

3i
confront the other person's choices. Since both the experimenter
and the subject achieve competence over their environment by the
method of differences, one can expect that dilemmas confronting the
experimenter also face the subject. This line of reasoning leads
to a view of the PDG as a mixed information game.

. Regarding the fifth criterion the PDG is reasonably well
defined as far as choices and outcomes are concerned. The significance
of PDG choices will be delimited in a model at the end of this
chapter.

In addition to meeting the criteria discussed above, an
additional reason for using the PDG is that the interpersonal effects
can be controlled as the sixth criterion calls for. The PDG is an
excellent way of establishing communication in one mode only. The
experimenter can control the situation so that the alter cannot be
seen or heard. In fact the only source of information about the
alter is through his choices in the game. This methodological fact
satisfies the sixth criterion for a suitable experimental situation
in which to test Head's Ideas about social thought. If the PDG can
be accepted as such an experimental situation then It behooves us to
look at the results from the magnitude of experiments performed around
this game situation. '

Prisoner's Dilemma Game Research Finding§_

This summary of findings will be divided Into five sections.
The first section will deal with the personalities of both players.
The second section will outline experiments In which the strategy
of the other player was controlled. The third section will present

experiments where the communication pattern was varied. The fourth

32
section will deal with experiments in which the reward structure was
manipulated. The fifth section will serve to review those experiments
relating theoretical positions to the PDG. Preceding each section
will be an abstract of the findings in that area. PDG findings have
been reviewed in other articles (Terhune, 1968, Vinacke, l969, Gallo
and McClintock, 1965). The major conclusions of these reviews will
be mentioned insofar as these apply to the PDG but articles mentioned
by them which do not appear to be of great importance will not be
reviewed here again. Within each section reviews of articles will be
grouped so that there is at least an intuitive similarity among the
manipulated variables.

Personality. The results from personality studies in the PDG
give a very confusing picture. In examining studies in which sex of
the subject was varied Gallo and McClintock (1965) conclude that sex
is not related to PDG choices. Rapoport and Chammah (1965b) on the
other hand, fOUnd that both the sex of the player and his Opponents
have large effects on the level of c00peration in a 300 trial PDG but
that this difference does not emerge on the first two trials. Tedeschi,
Lesnick and Gahagan (1968) found sex differences emerged by trials
3, 5 and 10 In 100 Iterations of the PDG but these differences
disappeared after the full play of the game.

Measured personality variables have also been found to not
reliably affect the level of cooperation from experiment to experiment.
This has been pointed out in a review by Terhune (1968) and Is
illustrated again in recent studies by Sermat (1967b) and Lutzker
(1960). On the other hand real life personality differences such

as the degree of mental health, the race of the opponent, the degree

33
of friendship of the two players and an economic orientation on the
part of the players seem to affect the level of cooperation consistently.
The economic factor promises to be especially important since Doibear
and Love (1966) found that it overrode the effects of friendship.
Crowne (1966) found that subjects with an entrepreneurial.background
were more competitive than bureaucratic background subjects. Subjects
taking business courses are more competitive than students in non-
business courses. Finally, frequency of economic crimes among juveniles
is correlated with defective choices.

Gallo and McClintock (1965) have summarized findings from the
personality research on the Prisoner's Dilemma prior to 1965. They
claim that a large number of studies show no relation between sex and
the amount of cooperation. This last conclusion was based on a
relatively large number of studies. From a fewer number of experiments
they found that scores on the F scale tended to correlate negatively
with cooperation while subjects who were scored as high lnternationalists
made more cooperative choices.

The best study on sex differences has been done by Rapoport and
Chammah (l965b). Using several game matrices they ran 70 pairs of
males, 70 mixed sex pairs and 70 female pairs for 300 trials. They
concluded women pairs tended to cooperate mUch less than men when
both sexes played In like sex pairs. In mixed sex pairs women became
more cooperative than when they played with other women. Men became
more competitive when they played with women than when they played
with men. An index, M, was developed which takes Into account the
success of a I'martyr” run by a player. A successful run was defined

as the conversion of an exploiter to cooperation. An unsuccessful

34

run meant that the martyr changed to defection. These two frequencies
were combined into M. When a man was the defector in the martyr run,
the run failed to be successful about 2.5 times more than it was
successful regardless of whether the martyr was a man or a woman.
When a woman was the defector, nonsuccesses outnumbered successes 3.7
to I if the martyr was a man, and h.8 to one if the martyr is a woman.
Tentatively, women have more of a tendency to exploit. However as
Rapoport and Chammah point out, if the martyr run had continued longer
women might have succumbed to the Inducement and switched to cooperation.

In this same article, RapOport and Chammah present an explanation
of why male groups come to cooperate more than do female groups. First
of all they define pI as the correlation of a subject's response on
trial n with the other subject's response on trial n-I. For the male
pairs pl was .51. For the female pairs It was .3A while for the mixed
sex groups it was .35 and .39. They consider this measure as evidence
that men tend to be more imitative. This leads to a consideration
of two automata with a tendency to perfectly imitate the other I.e.
pI-l.OO. Under conditions that the automata choose randomly on the
first trial, have a bias to maximize their payoff and minimal ”noise”
is introduced into the system, the tendency will be.for the system
to lock in on the mutually cooperative cell. Rapoport and Chammah
reason that men's higher cooperative response frequencies stem from
their greater tendency to give tit-forhtat (as measured by p') In
situations of this sort.

Tedeschi, Lesnick and Gahagan (1968) found no differences In
cooperative choices over 100 Iterations of a PDG between males and

females when faced with an alter who cooperated as much as he defected

35

in a highly competitive game. Females did cooperate more than males
for the first ten trials of the game. Under low levels of competitive-
ness the authors feel that the sex differences in cooperativeness may -
persist throughout the game.

Vinacke (1969) in his review of PDG findings found four studies
where males were clearly more cooperative than females and three studies
where there was no difference in playing that could be reliably
assigned to sex differences. Vinacke concludes that the laboratory
game situation makes males less exploitive than they would otherwise
be while females become more exploitive. By the words ”laboratory
game situation“ Vinacke means that females infer from the PDG matrix
that the experimenter wants them to compete. Thus females become
more competitive in order to cOOperate with the experimenter. Presumably
males are inherently more competitive and once they perceive the
experimenter's intent react against It.

Komorita (1965) found that females tend to be more cooperative
than males when facing an unconditionally cooperative other while the
reverse was true for a conditionally cooperative other. Horai and
Tedeschi (1969) In a study to be reviewed later found no effect on
subjects' PDG plays that could reliably be attributed to sex. Kanouse
and Wiest (1967) could not find any difference in cooperation that
could be attributed to the player's sex or the sex of the other person.
Bixenstlne and O'ReIlly (1966) found males to be more cooperative
than females.

Terhune (1968) used three one trial games and two 30 trial
games to assess the effect of the subject's motivational structure

on the amount of cooperation he displayed. The one trial games varied

36
along Rapoport's c00perative index. In all three one trial games
subjects who were high need achievers were more cooperative; subjects
who measured high in need affiliation were more defensive where defensive
means they chose 0 and expected the other to choose D. Subjects who
had a high need for power were most exploitive. In a 30 trial game
where communication was not allowed the difference due to motivation
disappeared over trials. This was true also of a thirty trial game
in which communication was allowed. Subjects in the noncommunication
games perceived the other person as a competitor and Opponent. This
was used as a reason by the subjects for their own competitiveness.
yin addition for all motive types it was found that the less favorable
the first trial outcome, the more conflict there was in subsequent
trials.

Terhune also reviewed the findings of the research on personality
and cooperation-conflict in games. Most personality scales do not
relate to cooperation. Those which are related to cooperation are not
replicated by other studies.

Deutsch (l960b) found that the F scale did predict PDG behavior
but this was not found by Wrightsman In a non-zero-sum game. The
Internationalism-Isolationlsm scale was found to relate to behavior
in several non-zero-sum games but Pillsuk et al (1965) found it did
not explain behavior In an expanded PDG. Christie's Machiavellianism
scale was found to predict behavior in some games but not In a non-
zero-sum game. Some of the needs (autonomy, abasement, aggression,
deference) as measured by the Gough adjective check list were related

to PDG behavior but other needs (nurturance, dominance) were not. The

37
following scales did not predict behavior in a non-zero-sum game or
an expanded PDG: Guilford Zimmerman test, eight basic values
questionnaire, self acceptance, monetary risk preference, risk taking
scale, verbal hostility, rigidity, social responsibility, personal
optimism, anti-police attitudes, political cynicism, Crowne Marlowe's
social desirability and Edward's social desirability (Terhune, I968).

Lutzker (1960) administered his internationalism-Isolationism
scale to a group of 600 students. From the pOpulation of returnees
he took the highest 11% of scores and the lowest 19%. (Isolationists
were unwilling to participate in the experiment.) In addition a
control group of untested subjects was included. He found that
lnternationalists cooperated more than the isolationists as did the
control group. Also competition increased over the 30 trials for the
isolationist and control groups but not for the internationalists.

Sermat (1967b) found that neither the MMPI dominance scale
nor the Lutzker Internationalism-Isolationism scale correlated
significantly with cooperative choice frequencies under a 100%
competitive treatment condition in the PDG.

In two, two trial PDG games involving various combinations of
real and imaginary payments, Wrightsman (1966) defined as a trusting
person one who chose C, expected the other to choose C and gave as
his reason for this a concept of trust, fairness or cooperation.
Trustworthy behavior was defined as present when the subject chose C
after the other had chosen C. On the Philosophies of Human Nature
scale trusting subjects had generally more positive attitudes towards
others, saw people as more trustworthy, more altruistic and more

Independent than did distrustlng subjects and other subjects. Several

38

other scales including the F scale were not related to trusting behavior.
None of the scales predicted trustworthiness. In the second experiment
trusting subjects had more favorable views of human nature but the
mean values of the three groups were arranged differently from that
of experiment one.

Bobbitt (1967) in his doctoral study found that subjects who
were classified as internally controlled, that is those who try to
alter present social conditions for the sake of personal gain, as
measured by the I-E scale were significantly more c00perative than those
who were classified as externally controlled. This latter phrase means-
those persons willing to accept the social condition. This difference
occurred when the other was simulated and was predominantly uncooperative
for an initial block of trials. The two groups were not differentiated
when the simulated other switched his strategy in the second block of
trials. In a second experiment when the simulated other was predominantly
cooperative in the initial block of trials, internals cooperated more
than externals. In this second experiment when the other was uncooperative
in the first block of trials internals again cooperated more than externals.

Knapp and Podell (1968) decried the lack of relationships found
between personality measures and PDG game behavior. They postulated
that this occurred because only paper and pencil measures were used.
As a basis for Individual differences they used samples of prison
inmates, mental patients and students. For 2A trials all three groups
faced a simulated other who responded at the 502 cooperative.level.
All three groups were then split Into two halves. One half of the
subjects faced an 802 cooperative other while the other half met a 202

cooperative other. All three groups under the latter condition responded

39
at approximately the same level of cooperation although inmates tended
to be more c00perative. Under the 80% C condition students were more
cooperative than patients who were, in turn, more cooperative than
inmates. In addition this is one of the few stUdies to find a
differential response on the part of subjects to any unconditional
strategy on the part of the other player. Rapoport (1968), in
commenting on this article, mentions that subjects may have been
responding to the perceived change in the other person's strategy and
not to the unconditional strategy itself. In the same comment he
mentions an unpublished manuscript by Chammah in which about one half
the subjects cOOperate and one half exploit a simulated other who
responds cooperatively 100% of the time.

Harford and Solomon (1969) used a better defined patient
population in their comparison of normals and schizophrenics. The
schizophrenics were split into paranoids and nonparanoids and then
matched on a number of pregame psychological tests. Twenty four
patients were in each group. However normals were college students
and did not match the patient group. All subjects then played a PDG
for 30 trials against either a reformed saint or lapsed sinner strategy.
In the latter condition the confederate chose cooperatively for the
first three trials and then switched to conditional cooperation for
the remainder of the game. In the reformed sinner condition the
confederate chose noncooperatively for the first three trials, then
switched to unconditional cooperation for the next three trials, and
finally changed to conditional cooperation. In describing the reaction
to these strategies the authors conclude that, when the sinner ls

punitive and tough the paranoid Is docile and cOOperatIve; with the

ho
sinner's abrupt shift to a soft benevolence the paranoids become
exploitive. With the saint strategy the reverse is true. Initially
they are exploitive, but then switch to cooperation when the confederate
becomes conditionally cooperative. The authors conclude that the
paranoids' interpersonal stance is not indiscriminant and can be
modified by variations in the strategy of the other party. The normals
were more sensitive to strategy changes than either of the patients.
Also the reformed sinner strategy was more effective in establishing
a trust relationship with the students than with the patients.
Finally the students chose competitively more than the patients.

Heller (1967) in his doctoral thesis found that Caucasian and
Mexican-American subjects played more competitively when the other
subject was a Negro than when the other was a member of their own race.
This study was limited to juveniles who had been incarcerated. At
the same time measures of racial prejudice in whites and Mexican-
Americans did not predict their level of cOOperativeness. Frequency
of economic crimes was correlated with the level of cooperativeness
as was a measure of socialization and the Marlowe Crowne Social
Desirabllity Scale. Both of the last two correlations disappeared
however for some of the experimental conditions.

Crowne (1966) set out to Investigate the correlates of two
person nonezero-sum game behavior in the economic motivation of the
subject. Subjects were classified as entrepreneurs or bureaucrats
according to the jobs their fathers held. Entrepreneurial parents
are those engaged In risk taking occupations (farmers, small businessmen,
doctors etc.). The sole criterion of bureaucracy is employment In a

relatively large organization. He also measured level of aspiration.

41
Although he had markedly unequal cell frequencies Crowne used an
unweighted means of analysis of variance. Entrepreneurial backgrounds
produced more exploitive bargaining over the 20 trial PDG although this
was marginally significant. Interaction between family orientation
and level of aspiration was significant also. Bureaucrats and entre-
preneurs with tendencies to escape from competition were more cooperative
than the remaining entrepreneurs.

Dolbear and Love (1966) felt that the risk orientation that
subjects take toward games would predict how the subjects would respond
in the PDG.~ Risk was measured in a lottery type game where the subjects'
propensity to take risks could be assessed from their choices. In the
PDG itself risk was measured by the number of times a subjectchose C
when faced with a stooge who chose D 13 straight times. Their data
indicate that risk orientation is not a good predictor of the subjects'
risks in the PDG. The subjects were from business school and this may
reflect in the negative findings. When the stooge switched to a 100%
cooperative strategy the subjects became suspicious of the experiment.
This is one of the few studies to report on this suspicion.

The effect on cOOperation of the prior degree of friendship
between the players has also been studied (Oskamp and Perlman, 1966).
The subjects from the two colleges were divided into four groups In
which the pairs of males had differing degrees of mutual friendship.
This was a standard PDG game played for thirty trials. Points won
were converted to money after the game was over. They found that
cooperation was lower at the college where business was the prime
subject matter. Friendship at this college had no effect on the way

the game was played. At the smaller nonbuslness school cooperation

#2
was higher and here the level of attained cooperation depended on the
degree of mutual liking. This finding seems to be similar to Crowne's
(1966) and to Swingle and Gillis' (1968).

Swingle and Gillis (1968) found that high school subjects whose
opponent was a friend were initially more cooperative but gradually
moved to a competitive strategy. They were markedly affected by their
friends' strategy changes. Subjects who played against a disliked
other, settled early in the game into a highly competitive strategy and
remained unaffected by abrupt strategy changes on the part of the other.

As Sheff (1967) noted, social psychological experimenters shy
away from measuring subjects' perceptions in mixed motive games.

Lumsden (1966) ran an experiment in which this was done. No statistical
tests were used to assess results. One of Lumsden's aims was to judge
the degree of similarity between the traditional laboratory experimental
games and real life international conflicts. Consequently he used a

PDG matrix and a ”chicken” matrix for two of the games and two inter-
national problems to simulate the traditional games. The latter two
games evoked a very high percentage of C responses from the participants.
This was explained by way of a peace norm, in the Norwegian subjects,
evoked by the internation context. Of the two traditional games the
‘subjects were consistently less cooperative in the PDG game. In
addition, after the 100 trial games were completed the subjects were
administered a semantic differential. There were marked differences
between the profiles of self and other In the two traditional experi-
mental games. In particular subjects evaluated themselves more

favorably than they evaluated the other. Although no test of significance

A3
is reported the percentage of cooperation was found to be negatively
correlated with the level of self evaluation.

Deutsch (l960b) had subjects play a two trial two position
game. On the first trial they played from the first position. That
is each 5 made his choice before the other had to make his. (In
fact the other person was fictional). Here the S had to decide whether
to trust the other or not. In the second position the S chose second
after he knew the choice of the fictional other. Here the subject
could be trustworthy or not since the fictional player always chose
cooperatively. Subjects who were trusting tended to be trustworthy.
Those who were SUSpicious tended to be untrustworthy. A point biserial
correlation was obtained on the subjects' game behavior and F scale
score. The obtained value was significant. That is, trusting and
trustworthy subjects had low scores on the F scale while suspicious
and untrustworthy subjects scored high.

Marlowe, Gergen and Doob (l966) manipulated the personality
of the other player. The subjects played a PDG for 30 trials where
the simulated other player was programmed to play an 80% cooperative
strategy. Prior to playing the game, the personality of the simulated
other player was manipulated on an egotism-humility dimension. A six
Item pregame questionnaire indicated this manipulation was effective.
Crossed with this dimension was a dimension wherein half the subjects
were led to believe they would meet their partner after the game.

As predicted the Interaction was significant, Indicating that the
subjects exploited the humble other when no post game Interaction was
anticipated but exploited the egotistical other person when post game

interaction was expected by the subjects. Furthermore the same six

AA
item questionnaire was administered after the game to the subjects.
Perception of the humble partner was found to be virtually unaltered
by the game experience. The subjects' perception of the egotistical
other was changed. They came to see him as significantly less
egotistical, less independent and more likeable. No assessment of
the experimental deception was mentioned by the authors. In a second
experiment, subjects exploited a simulated partner significantly more
when they anticipated no future interaction than when they anticipated
they would meet him after the game. No prior information was given
about the personality of the other person.

Sampson and Kardush (1965) used a sample of preadolescents who
were involved in a summer camping trip in a 50 trial PDG. Points won
during the game could be convered to candy. They found that the overall
level of c00peration was low and that it decreased over the game.
Using a marginal significance level (p<.10) these authors conclude that
Negroes initially were more cooperative than whites but that their
cOOperation declined faster than for the white children. In addition
for male white pairs, the older children cooperate more than the
younger ones. The relation is reversed for the females. For Negroes,
the younger children were more competitive. Sex did not Interact with
age for Negroes.

Simulated strategy findings. This section deals with studies
wherein the game behavior of the other is controlled in some way.
Success or friendship prior to the game itself was manipulated and
the effect on a standard strategy studied. The simulated subject made
promises with varying credibility or utilized a power advantage in

differing ways. Experimenters frequently varied the simulated person's

AS

strategy by making it contingent or noncontingent on the subject's
choices. Sometimes the strategy was changed in frequency or
contingency in the middle of the game to see if the primary strategy
exerted more of an effect than did the later strategy or vice versa.
Finally experimenters are beginning to investigate the effect of the
simulated person's strategy on the subject's ratings, impressions and
preferences for the supposed other. '

A wide variety of experiments showed that extra game experience
affects the way the PDG is played. Thus strategy change was found
to affect cooperation only if the subject considers he is playing
against a friend as opposed to a person whom he doesn't know. Subjects
who witnessed a competitive other cooperated more than a control group
against a cooperative other. Pregame dyadic interaction was found
to increase cooperation even if the game occurred one week following
the original interaction. Subjects were more c00perative in a short
game when another promised to cooperate. In a longer game the promise
to cooperate affected only the trial after it was sent. In the latter
experiments the message promised cooperation only on that trial and
did not refer to an overall strategy. When a simulated other held power
over the subject, benevolent use of that power increased cooperation
over and above when both players held the power. Mere possession of
the power did not affect cooperation when two subjects possessed it
or when only one subject did.

Komorita (1965) found that females tended to be more cooperative
when facing an unconditionally cooperative other while males were more
cOOperatIve when faced with a conditionally cOOperative other. What

the simulated other did early In the game affected how the subject

46
responded to him later in the game. Thus Sermat (1967a) found that
an unconditional strategy, either cooperative or competitive, produced
more cooperative choices in a later contingent condition than subjects
who early in the game faced a contingently cooperative other. Sermat
(1967b) in another experiment found early choice strategy by a simulated
other affected subjects' later responses in a PDG in a cooperative
direction only when the other was known to be free to change his
strategy anytime. Swingle (1968) in an experiment similar to Sermat's
found only a significant increase in variability of response after
the strategy change occurred.

Several experiments find that subjects do not respond differentially
to the differing levels of noncontingent cooperation. it is only in
indirect fashion that subjects can be inferred to respond differentially
to different unconditional strategies. Thus Gahagan and Tedeschi (1968)
found that subjects under some conditions were more cooperative when the
simulated other chose C 75% of the time than when the other chose C 50%
of the time. Faucheux and Moscovicl (1968) found a similar effect when
the name ”nature“ was associated with the Opponents 62% cooperative
strategy and the name ”chance” was associated with the 502 cooperative,
strategy. While the game was not a PDG the authors found that subjects
chose cooperatively more Often when they played against ”nature” than
when they played against “chance.” Komorita (1965) found that females
respond cooperatively more to an unconditionally cooperative other
than do males. Furthermore as the simulated other became more cooperative
females increased their level of cooperation. Neither Solomon (1960)

nor Bixenstlne, Potash and Wilson (1963) found any difference in the

#7

level of competitive choices when groups of subjects faced an
unconditionally competitive other or an unconditionally cooperative
other.

Impressions of the simulated other have also been studied as
a function of the strategy he adopts. If he changes this strategy
in the middle of the game Swingle and Coady (1967) found that impressions
so formed were determined by a primary effect at least in a game run
for a relatively large number of trials. Wilson and Insko (1968)
in a shOrter game found that impressions of the other were formed by
a recency effect. Tedeschi, Aranoff, Gahagan and Heisler (1968) In
an experiment using several simulated strategies felt that any explana-
tion of the subject's PDG behavior must involve knowledge about the
other person's cognitions of the opponent's intentions. Harford and
Solomon (1960) also interpret PDG behavior in terms of cognitive
inferences although no data 15 presented for this. Solomon found that
subject's inferences about the other person's strategy are contingent
upon the type of strategy the simulated other uses. Komorita and
Mechllng (1967) also explained their findings in terms of impressions
the subjects formed about the other person.

Evans (l96b) created three groups in his study. For group one
a promise to cooperate was made to the subject and E backed this up
with a threat to take away points if the stooge didn't keep his promise.
In group two the confederate made the same promise but no threat was
mentioned. In group three no promise was mentioned. The subjects were
run on a six trial game where the other player always gave the same
response on any trial as the subject. Group one was slgnlficantlymore

cooperative than group three. Both groups one and two rated the

#8
confederate as more trustworthy than did group three. For trial one,
12 out of the 15 subjects in group one chose C whereas only seven out
of 30 in the other two groups did so.

Komorita, Sheposh and Braver (1968) introduced a third
alternative choice to the normally two choice PDG. The difference
between the third choice and the regular two choices of the PDG was
that use of this choice assured negative outcomes for both players.

If used unilaterally the outcome was always worse for the other player
no matter what choice the other player made. This third choice was
used by a stooge in one experiment as a unilateral threat. The stooge
used the power in a passive manner under one condition, in a benevolent
way for a second condition and in a malevolent way in a third condition.
Under a second experiment where both the subject and the stooge had

the power (bilateral threat), the stooge was programmed to utilize his
in a benevolent or a passive way. A matching strategy was added as a
fourth condition in experiment two. Under unilateral power, the
benevolent use of power increased cooperation. Under bilateral power
the benevolent strategy increased cooperation but significantly less
than the benevolent condition In the unilateral threat experiment. In
experiment one malevolent use of power resulted in increased mutual
defection. With regard to the passive power condition for both
experiments the authors conclude that such power must be exercised

if the power is to be effective. The theoretical discussion centered
around the perception and communication Of power without any empirical
evidence that such perceptions occurred.

Gahagan and Tedeschi (1968) point out that there are two general
ways of studying the strategy of the other. The first Is to use a

post hoc method, as did Rapoport and Chammah (l965b), to look at the

49
choices of one player and how they affect the choices of the other
on various measures. The other is to simulate the choices of the other
player and see what result this produces in the subject's simulated
choices. These authors chose the latter course and as an experimental
manipulation had one half of their subjects play against a 502 cooperative
strategy, while the other half played against a strategy wherein the '
opponent played cooperatively 75% of the time. Each of these strategies
was crossed with three levels of a factor called credibility of promise.
Throughout the game subjects received one Of five possible promises.
It was always the same promise though and informed the subject that
the confederate was going to choose cooperatively on the next trial.
The stooge kept his promise 30% or 60% or 90% of the time. Thirty five
females and 37 males were assigned at random to the six cells. Subjects
played the PDG for 110 trials under instructions to maximize gain.
They participated as part of course credit and received no pay. The
authors analyzed the proportion of cooperation over the entire 110 trials
(CPI); the proportion of cooperation following messages (CP2); the
frequency with which promises were reciprocated on the ten message trials
(FRI); the behavioral credibility of their messages (CR1); Rapoport
and Chammah's trustworthiness (TW), repentance (R), forgiveness (F)
and trust (T). Strategy by Itself, had no effect on any of the dependent
variables. The discussion centered around the fact that while subjects
can discriminate between the two strategies they can still not make
predictions of the other's choice and, hence cannot act upon their
discriminations. The credibility factor produced differences on the
CP2, FRI and T variables. The subjects sent more messages of Intention

to cooperate when the promises were 902 credible than at either the 60%

50
or 30% levels, which, in turn were not different from each other.
Subjects not only reciprocated highly credible messages but they
also tended to cooperate on the next trial. The messages had no
effect on the overall level of cooperation. This is interpreted as
meaning the subjects discriminated between the postmessage trial and the
remaining trials. Also trust (T) diminished as credibility diminished.
The strategy by credibility interaction was significant on CPI, CP2
and F (forgiveness). The 75% cooperative strategy produced more
overall cooperation at the 60% and 30% credibility levels than did
the 502 cooperative other. NO difference existed among the strategies
at the 90% credibility level.

Swingle and Gillis (1968) in a PDG, investigated whether liking
had any effect on cooperative plays. If the opponent changed his
strategy from cooperative to competitive or vice versa this affected
the subjects' responses in the expected direction only if the programmed
other was considered a likeable person. The reaction to strategy change
on the part of the opponent was Interpreted as a tendency on the part
of friends to match the behavior of the friend.

Swingle (1968) found that male subjects who witnessed a non-
cooperative other before playing the game themselves were more likely
to respond cooperatively to a benevolent strategy than were a control
group. The data Indicates that such cooperation was increased the
longer the subjects witnessed the noncOOperatIve other In practice
sessions. Witnessing an unconditionally cooperative other seemed to
have little effect when the other became uncooperative In the game
itself. As In another study the author felt that subjects responded

differently to similar strategic postures of their opponents. From

51
correlation data the author concluded that highly cooperative subjects
retaliate against the Opponents defection immediately and severely.
Highly competitive subjects tend to switch to a more OOOperative
response when the opposing player defects.

Harrison and McClintock (I965) manipulated pregame dyad success.
If subjects were run immediately following the experimental manipulation,
dyads who had experienced success were significantly more cooperative
than subjects who experienced failure or a control group who had no
previous contact besides the PDG. If subjects participated in the
game one week after the manipulated success failure condition, coopera-
tion for both the successful and unsuccessful dyads was higher than for
the control group but not different from each other.

Sermat recently published two studies on initial cooperation-
competition. In the first study (Sermat, l967a) the subject was faced
with a 1002 cooperative other or a 1002 competitive other for 30 trials.
Following this the subject's opponent was switched to conditional
cooperation for 200 trials. A control group faced a conditionally
cooperative other throughout the game. The first two groups were
found to become increasingly cooperative over the last 200 trials
unlike the control group. Sermat relates two other unpublished studies
that found a decrease In the cooperation over trials. Regardless
of the type of pretreatment, subjects who received a cooperative
choice from the simulated other player on the first trial In the PDG,
were consistently more cooperative than subjects who received a
competitive first choice. The subjects played the game for course
credit. Little Information is given as to whether the subjects were

aware they were playing against a confederate. The pretreatment

52
phase was presented as one experiment. The conditional strategy was
presented as a separate experiment.

In a second experiment Sermat (1967b) had a simulated player
choose D for 50 trials and choose C for 20 trials. Three conditions
were grafted on to this basic strategy. ‘Group one was informed that
the other player was free to change his strategy at any time and that
he knew the outcome of every trial. Group two was informed that the
other player had committed himself to playing a strategy and was
informed of the outcome of every trial. Group three was told the other
player had committed himself to playing a strategy and was uninformed
about the outcomes. The three groups did not differ in the rate of
competitiveness over the first 30 trials although the rate declined
over the blocks. In the last 20 trials, group one became significantly
more cooperative than either of the committed groups. Neither the
MMPI dominance scale nor the International-Isolationism scale
correlated significantly with cooperative behavior under the initial
treatment condition. Again no mention is made of testing the experi-
mental deception.

Pylyshyn, Agnew and Illingworth (1966) found that pairs of
subjects played more cooperatively than did individuals against a
confederate who played a high percentage of tit-for-tat. Furthermore
the pairs increased at a faster rate than did the Individuals. By
trial 150 individuals were picking C 16 out of 25 times. Subjects
played for money.

Minas, Scodel, Marlowe and Rawson (1960) ran three PDG games.
In the first game subjects played in natural pairs for 30 trials.

in the second game subjects faced another player who made the same

53

choice as they did. In yet a third game subjects played a confederate
who chose 0 100% of the time. In all three games the subject chose
competitively about 60% to 65% of the time.

Harford and Solomon report using a ”lapsed saint” strategy as
well as a "reformed sinner” strategy on both the mentally 111 (1969)
and normals (1967, 1969). In the 1967 study the “lapsed saint” is
defined as a strategy where for the first three trials the simulated
other cooperates and then responds in a tit-for-tat fashion until
30 trials have been completed. The ”reformed sinner” strategy is
identical to the ”lapsed saint” strategy except that three 0 choices
precede the "lapsed saint” strategy. Harford and Solomon report
that a "reformed sinner” strategy is more effective in eliciting
COOperation than is a “lapsed saint“ strategy. The interpretation
is made in terms of impressions that subjects must glean from the
two strategies although no empirical evidence is presented for this.

Komorita (1965) Placed males and females against a conditionally
cooperative and an unconditionally cooperative other player for 80
trials. Females tended to be more cooperative (but always .3 or less)
to an unconditionally cooperative other. For the conditionally
cooperative other males were more cooperative. Females increased their
cooperativeness as the simulated other player increased hers. Males
cooperated at the same low level throughout the range of values that
the unconditionally cooperative other performed. The interpretation
that Komorita places on his findings Is that females are more concerned
with Interpersonal relations than males whereas males are more

concerned with maximizing gain.

5A

Tedeschi, Aranoff, Gahagan and Heisler (1968) conducted an
experiment to test the interpretation of PDG results according to
partial reinforcement theory. Seventy undergraduates (30 females
and #0 males) were randomly assigned to one of seven conditions.

The C strategy choices of group one subjects were “reinforced” 100%
of the time and subjects' 0 choices were reinforced 0% of the time.
The other groups had different schedules. In group two the subjects
were reinforced 75% of the time for C's and 0% for D's; group three
had 100% for C's and 25% for D's; group four had 75% for both C's and
0's; group five received 50% for C's and 0% for D's; group six was
reinforced on 100% for 0'5 and 50% for D's. The data for cooperative
strategy selections over 100 iterations of the PDG generally
supported a partial reinforcement interpretation. However the fact
that no differences existed between conditions on trials to extinction
was interpreted as indicating that recourse to inferences about the
other's intentions may be necessary to handle PDG data.

Sermat (l967a, l967b) in studies previously reviewed here,
found that subjects responded more cooperatively to a contingently
cooperative other when they have faced an unconditionally cOOperatIve
other or an unconditionally competitive other early in the game than
subjects who faced a conditionally cooperative other early In the game.
In addition subjects were more likely to exploit an other whom they
knew to be committed to a cooperative strategy than an other who may
change his cooperation at any moment. This occurs when the other has
consistently chosen competitively over an Initial block of trials and

then consistently chooses cooperatively for a block of trials.

55

In a Study to be reviewed later in this section, Swingle and
Coady (1967) did not obtain the latter finding of Sermat despite a
similar procedure. No information was given by the experimenter
about the degree of commitment of the other in the experiment. However
Swingle and Coady observed that the variability of response increased
when the strategy changed.

Bixenstlne, Potash and Wilson (1963) placed 2“ males and 2A females
in a PDG which ran for 90 trials. Unbeknownst to them they were playing
against a confederate who, for one half the subjects played an 83%
cooperative strategy, and for the other half of the subjects played
an 83% competitive strategy. This condition was In effect for 30 trials
after which the confederate adopted a strategy that matched the other
player's responses 83% of the time. The purpose was to determine
the effects of initial level of cOOperatIon on the subjects~own level
of cooperation. The only significant effect they found In predicting
percent cooperation was an arbitrary measure of ethicaiity both for
the initial 30 trials and the final 60. The authors concluded that
the Initial strategy difference did not make any difference In the
way the subjects chose. However for the analysis of the last 60 trials
the reported F value for the Initial strategy by sex Interaction was
2.A39. For one and AD df the tabled F Is 2.8h (p<.10). The authors
seem to have accepted the null hypothesis on only marginal evidence.
They report that none of their subjects evidenced any suspicion that
they were not playing the person they were paired off with. Further-
more they found In pilot work that unvarlable benevolence led subjects

to believe they were not playing a real person.

56

Bixenstlne and Wilson (1963) using the same game format but
with a different strategy sequence ran #0 male and #0 female subjects.
Forty eight of the subjects faced a simulated strategy of 95% C for
#0 trials, followed by 50% C for 20 trials, then 80 trials of 5% C,
then 20 trials of 50% C and, finally #0 trials of 95% C. The second
group of 32 subjects received the following sequence of strategies for
the same trial lengths as did the first group: 5% C, then 50% C,
95% C, 50% C and 5% C. The design then calls for differing initial
cooperation levels for A0 trials followed by a reduction to a 50%
cooperation by both groups for 20 trials. For subjects who had faced
a 95% cooperative other during the first A0 trials the probability
of responding with a C response to a C response by the other on the
previous trial was generally lower than the probability of responding
with a 0 response to a C response by the other on a previous trial.
For subjects who faced a 5% C strategy in the first 40 trials just the
reverse was true. Moreover the probability of a cOOperatIve response
was higher for the 5% group than for the 95% group. The authors
conclude here, unlike their previous experiment that tacit communication
is possible in the PDG. They further indicate that subjects were not
suspicious that they had not been playing another person.

Faucheux and Moscovicl (1968) executed an experiment which
is relevant to the common finding that subjects respond with a
competitive strategy to the unconditional choices of a simulated player
regardless of the level of cooperation at which the opponent plays.
According to the authors, nature can be conceived, on the one hand,

as a condition representing some sort of order governed by fixed rules

‘.

57

which enable one to understand it and then to make predictions about
it. Chance is thought by them to mean that the opponent is in an
erratic state which evades comprehension and prediction. The opponent
is perceived to be working actively against the subject perhaps raising
hopes in order to deceive. On the other hand, nature because of its
regularity is perceived by the subject to be on his side. Half of
the subjects (French medical students) were informed that they were
to play a game against chance. The other half were told they were
playing a game against nature. The latter condition resulted in an
opponent strategy of 62% cooperation. The former experimental
condition gave the opponent a 50% cooperative strategy. The game
was played for 50 trials. As expected, subjects played more
exploitively against nature than chance apparently because of a
preconception that nature is systematic and can be solved, while
chance is capricious and even malevolent. The 55 did not so clearly
perceive the winning exploitive strategy when playing against chance.
The experimenters also experimentally manipulated feelings of self
esteem. Those subjects with high self esteem played more exploitively
than did those with low self esteem. When the experimenters assessed
feelings of natural self esteem no relation was found between subjects
with high self and low self esteem and exploitive play.

Solomon (1960) had subjects play a six trial game for
imaginary money. Conditions of power were varied by presenting
four matrices. One of the groups was In an equal power matrix
condition which In reality satisfied the conditions for a PDG
matrix. Each subject received one of three types of games strategies

from the simulated other. The other was unconditionally cooperative

58

for group one; for group two, the other made a conditionally cooperative
choice; for group three, the other made an unconditionally uncooperative
choice. Groups one and two were significantly more cooperative for all
four games. For the PDG subjects a conditionally cooperative other was
evaluated more favorably than an unconditionally cooperative other.
Furthermore the conditional cOOperatIon in the PDG was more effective
in inducing subjects to play cooperatively. There was no statistical
difference between the unconditional strategies. Also data indicated
that the subjects competitive choices were based upon an exploitive
intention more for the conditionally cooperative other than for the
other two strategies. Finally subjects made attributions about the
motives of the player they were facing. The responses were classified
as (a) cooperative (other seeks mutual gain), (b) individualistic (other
seeks maximum gain), (c) out of field (other does not want to win or
else he doesn't understand). In the PDG, the conditionally cooperative
other never received motive (c). Seventy five percent of the time he
was seen as being cooperative. The remainder saw him as individualistic.
For the unconditionally cOOperatIve other, 50% saw him as OOOperative,
7% saw him as individualistic and 36% put him out of field.

Komorita and Mechling (1967) studied the effect of betrayal on
a group of 6A subjects, 32 women and 32 men. Each subject played with
a member of the same sex and was instructed to maximize joint gain.
The authors were Interested In studying distrust after a betrayal and
how long before a reconciliation occurs. They consequently had a pre-
programmed strategy to play against the subject. For one experimental
condition the preprogrammed player defected on two consecutive trials
after four cooperative responses or after ten such responses. In

addition, matrix values were manipulated so that there was high and

59

low temptation to defect and high and low losses for a double cross.
The major unexpected finding was that reconciliation was slowest under
the high temptation, high 1055 and four trials to defection condition.
In addition the programmed player in the high temptation condition was
evaluated more negatively. One of the interpretations of the three
way interaction centered around an ambiguous impression the subjects
must have held of the other player after just four trials. Fear of
a doublecross would then be high for this group. For subjects who
faced an opponent who doublecrossed after ten trials, it was postulated
that such subjects held a firmer impression of the other's basic
cooperativeness despite his doublecross.

Swingle and Coady (1967) found that subjects were insensitive
to abrupt strategy changes when the programmed player began in a 100%
competitive or 100% cooperative state for 50 trials and then switched
to one of the following four levels of cooperation - 0%, 25%, 50%, 100%.
Both of the initial strategies elicited high competitiveness and this
remained at the same level for groups who changed as for the group which
served as a control. Variability of response was found to Increase
after the strategy change. Subjects In postexperiment ratings Indicated
they would rather play again with the initially cooperative player.
No preference effect was reliably related to the last half strategy
change. Evaluation of the other was most strongly determined by the
partner's initial strategy. Partner's final strategy was reported to
affect evaluation but the authors appear to be in error here. That
Is Factor A, partner's initial strategy, Is treated as If It had 3 df
when it had only I df In the experimental design. Factor B, partner's

final strategy, is treated as though It was associated with 1 df when

60
in fact it had 3. This has the effect of reducing the mean square
due to the initial programmed strategy and inflating the mean square
due to the final programmed strategy. If the df were changed it
would result in a change in the conclusion. Specifically the partner's
final strategy would not be seen as affecting the subject's evaluation
of him. Only the initial strategy would affect evaluation.

Wilson and Insko (1968) found that subjects' impressions of
another person were formed by his recent behavior rather than by
behavior he had performed previously. The PDG was played by #0 females
and A0 males for 20 trials. Under one condition in the first block
of ten trials a stooge played cooperatively 80% of the time. For the
final ten trials the stooge played competitively 80% of the time.

Under other experimental conditions the stooge adopted the same strategies
in reverse order. Subjects impressions of the stooge were explained

by a recency and not a primacy effect. Only two subjects suspected

the experimental deception.

Communications In the Prisoner's Dilemma Game. The studies In
this section refer to studies where various types of notes or
verbalizations are exchanged. Notes can assume several forms. They
can refer to the sender's strategy (with varying degrees of reliability
or credibility) either on the trial following, or a series of trials,
following the transmission of the note. The note can also convey
an expectation of what the other should do and convey (or not) what
will happen if he doesn't do this.

The communication can also clarify outcomes where the joint
outcome Is not available to one or both subjects. It can affirm

a cooperative norm or it can establish a cooperative intent over a

61

period of time. The note can be issued by the experimenter as a form
of establishing an orientation in the game.

Several studies have found that transmission of a note increased
cooperativeness in the form of cooperative choices or else in the form
of the other's perceived trust or intent. One study hOwever found
that threatening notes of varying credibility influenced cooperation
in an extended game but only on that trial to which the notes referred,
namely the one following their transmission. In another study the
credibility of the note had an effect on the subject's cooperation
in the entire game but the type of effect it had depended on the
level of cooperation of the other subject.

Pregame free social interaction facilitated cooperation
especially when it involved a statement of a cooperative norm or an
exchange of a series of standard notes. Experimenter instructions
have been found to facilitate or Impede cooperation at least in
shorter length games. Kanouse and Wiest (1967) reported that subjects
expected the other to carry out the same strategy to which they were
oriented by experimental instructions. Thus experimental Instructions
may in some sense be considered a form of indirect communication.

In general, communication seems to markedly influence the
subject's choices In a PDG. The reason for this given by most authors
is that it establishes the intent or state of the opponent. The
effect of the note seems to Increase the more It is nonthreatening
or nonaggressive In tone at least as far as can be told from the
Horai and Tedeschi, and Gahagan and Tedeschi studies.

Loomis (1959) used the PDG to study the effects Of communications
on the frequency of cOOperatIve strategy selection by subjects.

The notes expressed varying degrees of intention, expectation,

62

retaliation and absolution. The subject did not have to honor his note
but for two levels of note sending the note informed the subject that
the confederate would retaliate if the subject did not do as the note
bade him. Ten groups of subjects were run for five trials, of which
the first trial outcome was analyzed. Half of these groups were receivers
and half were senders of notes. Perceived mutual trust was considered
present if the subject expected a cooperative choice by the other
person on the first trial and also perceived the other person as
expecting him to choose cooperatively. They found that regardless of
the type of note sent it was effective in increasing perceived mutual
trust. As the level of the note increased (i.e. from expectation
only, to intention only, to expectation plus intention, to expectation
plus intention plus retaliation, to expectation plus intention plus
retaliation plus forgiveness) the level of perceived mutual trust
increased. Finally Loomis found that as perceived mutual trust
increased, subjects increased their level of cooperation.

Swanson (1967) allowed subjects to send notes to one another
prior to playing any trial in a 30 trial PDG. The game differed
from the usual PDG In that the outcome Information to both subjects
was varied. For.one group only one subject had complete Information
about the joint outcomes, the other person had knowledge only of his
own outcomes. .For a second group both had knowledge of joint outcomes
for any particular play of the PDG. Crossed with this condition
was a condition where the subjects were allowed to send or receive
notes of the form, ”I have chosen _. You have chosen _."
In one group the sender of the note had complete outcome Information.

For another group the sender was the person who had Incomplete outcome

63
knowledge. For yet a third group both parties could send notes, but
they alternated communications. As is common he found that coopera-
tion decreased over the first 30 trials. However there was evidence
of less of a decline in cooperation in the alternating communication
group.

Radlow and Weidner (1966) investigated the effects of unenforced
commitments in a game that ran for 98 trials. ~The experiment allowed
a pregame phase in which five standard notes could be exchanged until
a standard strategy had been agreed upon or until 15 exchanges had
taken place. Even though the strategies were unenforceable the
results indicate that negotiations with unenforceable commitments
results in a very high percentage of C being chosen, (about 90%).

When no communication was permitted, subjects started choosing C at
about 55%, went down to between A0% and 50% by the thirtieth trial

and returned to between 60% and 65% by the end of the game. Radlow
and Wiedner hypothesize that the familiar U curve of the noncommunica-
tion groups is a part of the dynamic process by which players explore
one another's intent. Preplay communication could be expected to
eliminate this phase of exploration.

Horai and Tedeschi (1969) allowed a simulated partner to threaten
the subject in various ways during a 150 trial PDG. The procedure
allowed the simulated partner to threaten the subject with a loss
of five, ten or twenty points in addition to the usual trial outcome
if the subject did not make a particular choice on the next trial as
the partner demanded. Further the experiment called for the simulated
partner to back up his demand on 10%, 50% or 90% of the trials on which

the subject failed to comply with the demand. In order for this

6A
procedure to be carried out ten noncompliance trials had to be elicited
from the subject. This was done by increasing the frequency of the
demands until the noncompliance criterion had been reached. The major
dependent variable was taken to be the number of threats issued across
the three credibility levels, the three magnitude of punishment levels
and the two sexes. The major findings with respect to credibility was
that all three credibility levels differed among themselves with
the ten percent credible other eliciting more noncompliance and the
90% credible other eliciting more compliance, while the 50% credible
other was intermediate. The punishment factor showed that although
the five and ten point conditions did not significantly differ from
each other they both differed from the 20 point condition. None of
the interaction factors nor sex factor was found to be significant.
Moreover, when the percentage of cooperation was used as the dependent
variable it was found to differ as a function of credibility. The 90%
credibility condition yielded significantly lower cOOperatIon than
the other two levels. This was explained in the following terms. By
complying more often the 90% credibility group experienced more threats
since ten noncompliances were needed. Since the simulated partner
cooperated 50% of the time and since the partner defected on all the
threat trials, the proportion of cooperative strategy selections on
nonthreat occasions by the partner was greater for the high credibility
players. These players had a greater Opportunity to discover that
threet trials were clearly different from nonthreat trials and that
they could exploit the more cooperative play of the other on the non-
threat trials. No other condition, Including sex, had any statistical

relation to the amount of cooperation.

65

Kanouse and Wiest (1967) gave a one trial, paper and pencil
version of the PDG to a group of subjects. They had two levels for
each of the following: partner condition, instructional set, sex of
the other player and sex of self. The subject was told that his
partner was a member of his class or an imaginary partner and, that the
partner was a member of the same or opposite sex. Furthermore the
subject was given a cooperative or competitivewset. The data were
analyzed by Sutcliffe's multiple classification procedure. Of the
subjects given the cooperative set, 67% chose C as compared to only 37%
who were given the individualistic set. The amount of cooperation was
not affected by the imaginary or real status of the other player.

Nor did sex of the subject or sex of the partner affect the percentage

of C. The expected choice of the partner was related to the instructional
set given the subject in that those given cOOperatIve instructions felt
that the partner would behave cooperatively while those given
individualistic instructions felt that he wouldn't.

Deutsch (l960a) investigated the concepts of “trust“ and
”suspicion” using the PDG. The essential features of ”trust“ according
to Deutsch are (a) that the individual Is confronted with an ambiguous
path, a path that can either lead to an event perceived to be beneficial
or an event perceived to be harmful (b) he perceives that the beneficial
or harmful event Is contingent upon the behavior of another person
(c) he perceives the strength of the harmful events to outweigh the
strength of the beneficial events. The Individual Is more likely to
choose the ambiguous path If he perceives the beneficial events as

stronger as compared to the harmful events and he has greater confidence

66
that beneficial events will occur rather than harmful. This is said
to be related to how much he trusts the other person.

Suspicion is postulated by Deutsch to occur (a) when the
individual is confronted with the possibility that a potentially
harmful event will occur (b) he perceives that the occurrence of the
harmful event is contingent upon the behavior of another person
(c) he perceives the possibility of engaging in behavior that will
prevent or reduce the harmful consequences of the other person's
behavior, if it occurs. Deutsch considers a choice gg£_to take an
ambiguous path a type of suspicious choice. An individual is considered
as more likely to make a suSpicious choice if he perceives the strength
of the harmful effects to be greater and he is confident that the
suspected behavior of the other will occur, and if he is more able
to prevent or reduce the harmful consequences of the other's behavior.

Deutsch postulated that a cooperative motivational orientation
would lead a subject to expect that the other had a reliable benevolent
intention toward him and would facilitate trust. The subject with
a competitive orientation would be most likely to expect a reliably
malevolent intention from the other just as his own intention would
be malevolent. When subjects are individually motivated only an
exchange intention was postulated to occur and that under special
circumstances (I.e. they both make their choice at the same time).
Also the opportunity to communicate with the other would lead Deutsch
to expect that individualistic Individuals would encourage mutual
cooperation.

In both the one trial and the ten trial PDG Deutsch found that
a cooperative orientation leads the Individual to make a cooperative

choice. The competitive orientation leads to 0 choices. However with

67

the individualistic orientation, the choice to cooperate or not is
a function of the experimental treatments. The opportunity to communi-
cate by note prior to choosing among the individualistically oriented
subjects inoreased their cOOperatIon. The opportunity to communicate
In the other two motivationally oriented groups did not enhance or
deter cooperation.

‘ Gahagan and Tedeschi (1968) pointed out that.most of the research
on P00 and communication had been done with relatively few trials.
Also none of the researchers had investigated the I'cre'dibility” of the
written messages. In their experiment, the subjects were 37 males
and 35 females. They played the game for no money and for a total
of 110 trials. A communicated message was considered by the authors
to be credible for them if the “other person” does what he says he will
do on the particular trial in question. They established three
credibility levels. For group one, 30% of the time the other player
did what he said he would. For group two, 60% of the time the message
was true. For group three the message was true 90% of the time. The
subjects in each of these three groups were randomly assigned to an
opponent who was 50% or 75% cOOperatIve. It Was found that the subjects
who received highly credible (90%) messages cooperated more often
than the other two groups on the trial following the message.
(The message was, ”I am going to push the cooperative switch on this
trial”). This group was also more likely to reciprocate promises.
There was a significant strategy by credibility Interaction on the
overall proportion of cooperation. The 75% strategy produced more
cooperative choices than the 50% strategy when messages were 60%

credible. The reverse was true when the messages were 30% credible.

68
No difference was found when the messages were 90% credible. They
conclude from this that the strategy of the other did not affect the
subject's behavior if the other's promises were consistently backed
up by his behavior. At the lower credibility levels the recipient
of messages was more likely to take the overall strategy into
consideration.

Oskamp and Perlman (1965) found that a small amount of social
interaction prior to beginning the game increased the amount of
cooperation in_a PDG that ran for 30 trials. Also they found that
the public commitment involved in stating a norm, namely that
cooperation is desirable, can increase the level of cooperation.
Subjects who publicly disagreed with a competitive norm were the
most cooperative. This experiment was unusual though in that it found
a relatively high level of cooperation among subjects. It did find
that males were more cooperative than females and that very few
subjects expressed that deception by the experimenter was present
even though the simulated other chose C 100% of the time.

Reward matrix manipulations in the Prisoner's Dilemma Game.
Gallo and McClintock (1965) In their review of game studies predicted
that there would be a difference in cooperation If real money was
used to pay off subjects In PDG experiments. In studies reviewed
below there was no statistical difference In cooperation when money,
shock or exam credits were used as the reward outcomes. Thus while
the Gallo and McClintock hypothesis has not been disproved no support
for It has been evidenced In the very few studies reviewed here.

More attention has been paid to finding an Index to describe

the incentive value of any PDG matrix. While the exact index has

69
not been specified the work of Rapoport and Chammah (l965b) and
Steele and Tedeschi (1967) have deveIOped powerful predictive indices.
The latter authors point out that more subtle personality effects
may emerge only when the cooperation index is in an intermediate
range.

Recently there have been attempts to modify the outcome matrix
so that it will reflect a very real fact of social interaction.

There is a feeling that once an interaction occurs the outcome matrix
is modified. This type of experimental manipulation while it may
increase the realism of the game situation, introduces unnecessary
complications that hinder proper experimental design.

Evans (196A) reports an experiment in which half of the
subjects were told that the points they won in a six trial game
represented money that they could take home with them. The other half
were told that the points they accumulated would be converted to exam
creditsand added to their final exam scores. The subjects interacted
with bone fide subjects and there were 18 dyads in each condition.

No statistical difference was found between the two groups in the
frequency with which they chose C.'

In previous experiments Radlow had used a one trial payoff
selected randomly from a large number of trials. Evans and Grumbaugh
(l966b) tested that procedure against one whose payoff depended upon
the outcome of every trial. The authors found that the payment
schedule did not affect the percentage of cooperation.

Bixenstlne and O'Reilly (1966), prior to the PDG, had subjects
roughly scale shock In terms of amount of money they would forego to

avoid the shock.' Forty males and AD females were then run In a PDG

70

with a member of the same sex as their partner. Half of the pairs
received shock for the first part of the game and then money for the
second part as the payoff for the outcome matrix. The other half of
the subjects received shock and money in reverse order. No difference
in cooperation was observed when shock was the outcome and when money
was the outcome, if other factors were held constant. Like other
studies males were observed to be more cOOperatIve than females.
This difference was most apparent In the game in which shock came first.
Unlike most other studies the level of cooperation was high.

Radlow (1965) performed two experiments on a 98 trial PDG.
Payoff was on the average between three and four dollars and was
based on the outcomes of one unknown trial picked randomly from the
98. No statistical tests were performed on the experimental results.
For both experiments Radlow found that subject's cooperative choices
initially went down In frequency then went back up. The experimental
difference between experiments one and two was that the latter
contained more explicit (but essentially neutral) Instructions about
the nature of the game and the reward matrix. In experiment two he
found that subjects cooperated about 25% more than In experiment one.

Using a standard PDG matrix and paid, recruited subjects.
Gumpert, Deutsch and Epstein (1969) found that subjects who played
the game for 20 trials for real dollars (RD) were less cOOperatIve
than subjects who played for Imaginary dollars (10). Subjects
In the RD and 10 conditions were paid the same amount for participating
but the subjects In the RD condition were given $10 In real money and
asked to attach great importance to winning or losing the money.

The authors do not specify whether the R0 subjects were, allowed to

71

keep their game winnings. Subjects in the ID condition were given
only an imaginary $10 and instructed to play as though they were
playing with real money. Subjects in the ID condition chose
cooperatively about 50% of the time whereas subjects in the RD
condition chose cooperatively about 31% of the time during the 20
trials. The authors conclude that Gallo and McClintock's contention
that subjects would increase cooperation if paid real money is incorrect.

Hesse, Dawson and Lane (1970) advance an equity explanation for
the effects on subject's PDG choices of differential reward level.
They speculate that subjects who receive high rewards for the very
little work involved in short PD games would be in an aversive state
since they received more than they felt they deserved. Subjects to
escape from the aversive state would give themselves less money by
stabilizing their choices at mutual noncooperation. To test this
hypothesis Messe et a1 had half the subjects face a high reward condition
and half face a low reward condition. In addition for each of these
two subgroups half of the subjects were required to fill out an
extensive pregame questionnaire while the other half were placed
directly into the ten trial game. It was predicted that subjects
in the high reward condition who performed a pretask should be
significantly more cooperative than those who did not fill out the
questionnaire. it was predicted that under low reward conditions
subjects would cooperate less If they worked on the pregame question-
naire than If they did not. Both of these predictions were borne
out supporting the position that a "norm of equity” mediates the

effect of reward level on behavior In the PDG. That Is when their

(N

72

senSe of equity is not violated either for them or against them
subjects are motivated to do well. When so motivated mutual
cooperation is more easily achieved. _

McKeown, Gahagan and Tedeschi (1967) had a stooge play a
game for 100 trials in which he held relatively high power over the '
other person's outcomes. In one condition he made sure the other
got detrimental outcomes 10% of the time, in another condition 50%
of the time, and in a third condition 90% of the time. After the
first 100 trials were completed the subject was given the Strong
power position. They found that varying the strategies ofia strong.
”dummy" player against a weak player who knows he will shift to the
stronger position at the midpoint of the experiment does not affect
the subjects' cooperation choices when he does become more powerful.
However the groups do respond differently on a ”plaintiveness” strategy.
This was measured as the frequency with which the subjects cOOperate
following a mutual defection. The 50% other produced more ”plaintiveness“
than the 10% powerful other. The 90% powerful other produced an
intermediate amount of ”plaintiveness.” No mention was made of subjects
suspecting the experimental deception.

Under the restrictions of the two inequalities defining a P00
and assumption of symmetry, R, S, T, P are free to take on a wide
range of values. Rapoport and Chammah (l965a) looked for an index to
describe the relationship between the Incentive value of the payoff
matrix and the course of game behavior. Their first conjecture
resulted In the following hypothesis. If other payoffs are kept
constant, C increases as R and S Increase and decreases as T and P

Increase. If the entries of the payoff matrix are thought of as

73

utilities then C (the frequency of cooperation) depends not on the
individual values but on the ratio of their differences. In an exten-
sive discussion they point out that for three variables one ratio
suffices but for four variables two ratios are needed to describe

the relationship. This means that all other ratios can be derived
from these basic ratios. The authors found the one ratio that best
describes the incentive value of the matrix in terms of frequency of
OOOperation in the various games to be given by C1, where CI is equal
to R-P/T-S. Axelrod (1967) from purely strategic considerations
developed a second CI which was equal to (T-R) (T-S)/(T-P)2. Rapoport
(l967a) suggested that his own formula is more compatible with a
psychological interpretation whereas Axelrod's has a nice geometric
interpretation.

Steele and Tedeschi (1967) generated a total of 208 indexes
and, in a correlational study, found that the index, log (T-S)/(R-P),
had the highest correlation (r-.6A) with the proportion of 0 choices
across #2 game matrices. This correlation, while high, was considered
incomplete since only one dyad was used with each matrix. In
addition not all of the game matrices were of the PDG type.

Jones, Steele, Gahagan and Tedeschi (1968) set out to make a
statistical test for linearity of their Index using only PDG matrices.
They designed a two factor experiment with one of the factors having
three levels of CI (.1, .5, .9). The second factor Increased the
absolute magnitudes of the payoffs while holding the CI constant.

The proportion of cooperative responses was found to Increase linearly
as a function of the log of CI. Furthermore the matrix with negative

values produced more cooperation especially over the high levels of

74

CI (.5 5 .9). Cooperation was observed to decrease over time.
Subjects here were playing for points and not for money.

Recently there has been a determined effort to make the PDG
a more authentic representation of the real life situation by adding
values to the outcome matrix. Guyer (1968) argued that the utilities
in the payoff matrix in social situations are not stable, that they
tend to change as a function of the past history of the participants.
The experimental game that he used featured game payoffs which were
responsive to the strategy choices that the players made. The
procedure called for eight different sequences of Prisoner's Dilemma
Games each containing 25 distinct games. Each sequence was characterized
by successively increasing or decreasing the R or the P parameter or
both. The parameter changes were made contingent upon the subject's
joint response by a series of eight rules. It was found that when
reward is either moderate and fixed or large and increasing a decreasing
severity of punishment produced as much cooperative behavior as did an
increasing severity of punishment. He Interpreted this to mean that
a decreasing severity of punishment in a dynamic environment produces
as much cooperation by engendering good will as does an increasing
severity of punishment which promotes cooperation only by its deterrent
effect.

Rapoport and Cole (1968) created a Multiple Prisoner's Dilemma
Game (MPO) for the same reasons that Guyer modified his outcome matrix.
The MPO In addition to the joint outcome Information presented by the
PDG matrix gives the players Information which sends them to one of
several other PDG matrices. Which matrix they go to is a function of

their joint outcome. Thus one or both players may take a temporary

75

loss in order to maneuver the dyad to a more favorable game. To
test the game three female groups faced a player who played an '
unconditionally cooperative strategy. Three game matrices were
employed. For groups one and three, the confederate increased his
cooperation as he went from game one to game three. The players in
group three were told they were playing against a preset strategy.
This group attained more points than those in group one. All groups
were also found to play the same percentage of C across the three
games and a significant decrease in C occurred over trials. Subjects
were asked to estimate how the stooge would play in the future. Mean
estimates of each group corresponded fairly closely to the actual
probabilities in each subgame, with high confederate probabilities
underestimated and low probabilities overestimated. However the
standard deviation of these estimates was large. The subjects'
estimates were used to test a dynamic programming model (Rapoport,
l967b). The test involved ranking optimal policies of playing the
MP0 on the basis of the subjects' perceived strategy of the other
player. At the end of the game the experimenter asked the subjects
to rank eight possible policies to yield a chosen ranking. The
correlation between the optimal policy rankings had chosen rankings
for the two models was not statistically different from zero thus
refuting the model. To account for this Rapoport and Cole explained
there were indications that the subject made conditional probability
estimates of the other person's behavior. The model calls for
unconditional estimates to be made.

Theoretical formulations about the Prisoner's Dilemma. Not

many studies have been oriented toward direct theoretical tests of

76

behavior in the PDG. The few tests of learning theory predictions
have not fared too well. Rapoport and Chammah's imitation theory
of PDG behavior is discussed later in this chapter.

Vinacke (1969) claims that it is crucial to determine how the
players perceive the game. He also noted that the players' perception
of the choices may not be the same as the experimenter's. Recently
two models have been advanced to account for PDG behavior in terms
of the attributions that the players make about one another's behavior
or the estimates of their future choices based upon their current
choices. The former model is developed by Scheff (1967) from a con-
sideration of a number of cusrent theories including Schelling's
theory of tacit coordination, Mead's theory of meaning and gesture,
and Schutz's theory of intersubjectivity. Anatol Rapoport is the
author of the latter model. It represents the player's choices now
as being at least partially determined by what will happen in the
future. Thus any player is assumed to estimate the policy of the
other player from that player's current choices. However this last
model has not stood up too well to the limited empirical tests to
which it has been put.

An article (Tedeschi, Aranoff, Gahagan and Heisler, 1968)
has already been reviewed demonstrating that the partial reinforcement
effect of learning theory does not account for behavior in the PDG.
These authors found that all groups, even though they had been on
differing schedules of reinforcement required the same number of
trials to a specified criterion of extinction.

An attempt was made by Halpin and Pillsuk (1967) to relate
the more complex behavior of the P00 to the findings of probability

matching studies. The usual finding In two choice experiments

77

is that subjects will come to select the more frequently rewarded
alternative with a frequency asymptotically equal to the_frequency
with which that alternative occurs. Sixty subjects were run under
three conditions. In group one, subjects were presented with a
series of 200 choices, 138 of which were C's while 62 were D's.
Subjects knew they were playing against a preset strategy, played for
money and were asked to predict the lights. In the second group,
a game with a ”real” other, the subjects were told they were playing
with another person of the same sex.' Payoffs were for money and
were determined by a PDG matrix. In addition subjects were
awarded a 1c gain or loss for correctly or incorrectly predicting
other's chOices. The sequence of choices was the same as that
specified in the first condition. Group three had the same conditions
as group two but they were told they were playing a computer. For
groups one and three the mean number of predictions was quite close
to the observed number of C choices. For group two, subjects
overestimated the number of C occurrences. (There were 72 Cs in the
last 100 trials; group two estimated 81; groups one and two estimated
71.) While this difference was not significant for the final 100
trials, it was significant for the total 200 trials. Thus there was
no indication that subjects match even their estimations to the
opponent's choices in a human social situation. The authors do not
indicate if the subjects in group two realized they were playing
against a preset strategy.

Preliminary evidence would indicate that simple learning theories
are not capable of handling PDG choices. Rapoport and Chammah (l965a,

l965b) have advanced an Imitation model to explain game playing.

78
Game theory also has not as yet predicted PDG choices. For a one
trial game, game theory predicts that rational players should choose
0. However empirically this has not proven to be the case (Terhune,
i968).

Rapoport (1967b) developed a semi-rational theory of game behavior
for iterated plays of the PDG. In this model the players are assumed
to play a sequence of different component games. The players are
considered not only sensitive to the outcomes of a particular choice
on a particular trial but also considerate of future gains or losses.
This consideration comes about because the players realize that their
choices make tacit communication or collusion possible. Thus Rapoport
believes that players realize that their present decision may partly
determine future decisions of the partner. Furthermore Rapoport
assumes that players are capable of estimating their partner's
propensities. Thus X'Is player A's subjective probability that
player B will choose cooperatively following a pair of joint C choices
by both players. The propensity Y'is A's subjective probability that
B will choose cooperatively following a joint CD choice. These subjective
probabilities plus two more, reflect what A thinks that B will do at
time t after a joint choice at time t-l. Thus it Is assumed that A
believes that B has a policy and his estimates of this policy come
from the tac1t communication that is said to occur in the PDG.

Taking into account these subjective probabilities, the expected
immediate reward and the future consequences of the decision, Rapoport
developed expected outcomes for each of all possible distinct

policies in the PDG. As yet however he has failed to show that this

ha!

and

3111
IKE

811

{III
100
life

lid 4

79

imaginative model applies to what subjects actually do. (Rapoport
and Cole, 1968).

T. J. Scheff (1967) has developed a theory of social consensus
and coordination. He reviewed Schelling's theory of tacit coordination,
Head's theory of meaning and gesture and Schutz's theory of inter-
subjectivity. His purpose was to point out that humans are capable
of making attributions of value and intent. In particular he views
mixed motive games as interactions in which it is necessary for a
participant to coordinate his behavior with the other's when the inten-
tion of the latter are unknOwn. However no adequate theoretical
argument has been advanced to account for the PDG. Schelling, as
reported by Scheff, felt that what was present in human social
situations was the mutual assessment and often successful coordination
of subjective information by the players or the meeting of minds.

Scheff repOrts that for Head communication was only a means to an end.
That Is communication increased consensus between the participants.p'
Schutz called such joint thought ”Intersubjectivity“ when there was a
shared consciousness between two persons Involved in a collaborative act.
Scheff reports that Schelling had a similar explanation for behavior
in mixed-motive games. ”Tacit coordination” was the term Schelling
applied, and explained it In terms of "the meeting of the minds,”
”mutual perceptions” and suggestions.

To provide a more specific and operational statement to these
theories Scheff defines three terms; consensus, communication and
coordination._ Consensus, the key term, Is defined not merely as
_agreement about a statement X, but also awareness of the agreement

and also awareness of the awareness. Thus consensus is said to exist

80
between two persons, with respect to the issue X, if there is an
infinite series of correct, mutually reciprocating attributions
of each to the other's attributions. The decisions of the game
players, that is their pattern of moves Scheff calls coordination.
Mutual coordination Is cooperation in terms of PDG behavior.
When both players defect Scheff calls it failure of coordination.
If either player exploits the other Scheff terms it asymmetric
coordination.

The basic postulate is that coordination is a function of
consensus and communication. Scheff makes several specific predictions.
1) With no communication between players, coordination is a function
of consensus:

(a) The greater the consensus, the more mutual coordination.

(b) The less the consensus, the more failure of coordination.

(c) The more asymmetric the consensus, the more asymmetric
the coordination, with the person making the most accurate attribution
profiting by the asymmetric coordination.

2) When communication ls allowed, consensus is a function of
communication:

(a) The more the communication, the greater the consensus.

(b) The more unidirectional the communication, the more
asymmetric the consensus, with the person receiving the messages
making more accurate attributions than the sender.

3) CoordinatiOn depends upon consensus, which, In turn may depend
upon communication.

(a) if consensus is constant, then the amount of communication

may be Increased Indefinitely without leading to more coordination.

81

In proposition 3) communication is seen as a causal process in
coordination. However communication is causal only in that it leads
to more consensus.

The author finds that the Deutsch experimental findings support
his theory but only by assuming that attributions were involved.
Consensus as defined here was explicitly measured by Loomis, and
Scheff finds that this experimenter's findings with respect to the
initial trial of the PDG are consistent with the major tenents of the
predictions outlined above. In the Deutsch experiment Scheff notes
that under the cooperative and competitive set of instructions
explicit information about the other player's intent is given. For
the cooperative instructions It is stated that the other player Intends
to play cooperatively, while in the competitive instructions the other
player's intent to play competitively was made salient. Moreover for
both types of instructions the idea of reciprocity of perspectives is
explicitly stated. The ”Individualistic” instructions of a third group
did not delineate the intent of the other player. Scheff claims that
the findings of this study can be explained in terms of consensus
rather than motivation. In Scheff's terms, Deutsch found that with
an increase in consensus there was more mutual coordination (89%
chose mutual coordination) while with less consensus there Is failure
of coordination (88% chose mutual defection). Finally the more
asymmetric the consensus the more asymmetric the coordination.

Loomis' study defined the “state of perceived mutual trust”
as a situation in which the Individual understands that the other
person is aware, and believes the Individual Is aware, that there
is no reasonable alternative but cooperation. Scheff feels that

this definition reflects what he has called consensus. To ascertain

82'

the state of perceived mutual trust, Loomis asked each player two
questions before the first move. ”What do you think the other person
will choose?" “What do you think he thinks you will choose?” LOomis
in terms of Scheff's proposition found that the more the communication,
the greater the consensus. Asymmetry of communication produced asymmetry
of consensus. Finally Loomis found that the greater the consensus
the more mutual coordination. That is, subjects in consensus cooperated
79% of the time. Subjects in the dissensus condition failed to
mutually coordinate their moves. This occurred 78% of the time.
Experiment 1: Problem and Hypotheses

An imitative explanation for PDG behavior has been advanced by
Rapoport and Chammah (l965a, 1965b). This explanation has been
proposed by Bandura and Walters (1963) for other dyadic social
situations and a general reciprocity norm has been advanced by
Gouldner (i960). Rapoport and Chammah correlated the total frequency
of C responses and the presence of C responses by the other person
on zero, one and two trial lags. From the high correlations obtained
for several hundred subjects in several types of game situations
these authOrs state: ”The Interaction effect In repeated plays of
Prisoner's Dilemma Is strong and positive. In the single sessions
a pronounced tendency is observed of each player to lmltate the
other. It is fairly clear, therefore that a search of Individual
correlates of this frequency will not be rewarding." (Rapoport and
Chammah, l965a, p. 67).

However under some game conditions, Terhune (1968) found
differences In cooperative responding when players were selected for

personallty'dlfferences on need for power, need for achievement and

83
need for affiliation. In single trial games, where feedback did
not occur, opportunity for communication was minimized and the
cooperation index was relatively high, the need achievement and
need affiliation subjects were more ”cooperative” than the need for
power subjects. in an extended trial game with a low cooperation
index, personality differences in cOOperatIon were obtained when
explicit communication was allowed. Personality differences due
to internal-external control were also obtained by Bobbit (1967).
Knapp and Podell (1968) obtained differences in cooperation due
to gross personality differences as did Harford and Solomon (l969).
Rapoport and Chammah (1965a) show that females have a lower Imitation
index.than do males. Furthermore males have a higher imitation
index playing males than they do playing females. It is concluded
then that personality differences do affect the frequency of
cooperative responding. '

In addition to these personality differences various authors
have shown that a subject's reactions to confederate strategies
cannot be explained by imitation. Loomis observed that 13% of the
time subjects exploited a cooperative strategy on the first trial.
Harford and Solomon (1969) showed that subjects reacted differently
to ”lapsed saint“ and ”reformed sinner” programmed strategies over
a short period of trials. Swingle and Gillis (1968) Showed that
imitation would occur only If the programmed other was considered
,a likeable person. Swingle and Coady (1967) observed that subjects
were lnsensitlve to abrupt strategy changes when the programmed player
began In a 100% competitive or 100% cOOperatIve state_for 50 trials

then switched to several different levels of cooperation. Also

84
Gahagan and Tedeschi (1968) found that under conditions of low
credibility subjects reacted differently to two different
unconditionally cooperative strategies. (50% and 75%). However
this is also in the direction of exploiting the confederate.

Furthermore, the correlations which Rapoport and Chammah
computed do not necessarily constitute an ”explanation” of PDG
behavior. Rather, they can be said to describe the behavior of
the subjects. Therefore the concept of imitation can be considered
only a description and not an explanation of PDG choices.

The postulates of Mead's theory would explain the communication
effect found in PDG iterated plays. (Wilson and Insko, 1968; Komorita
and Mechling, 1967; Swingle and Coady, 1967; Evans, 196“) The
gesture is that portion of an act which epitomizes it. The gesture
has two functions. If the player is the recipient of gestural
behavior the problem becomes to match the strategy of the other
person with that behavior. If the participant is about to act In the
social situation the problem becomes one of selecting that behavior
(or sequence of behaviors) that will epitomize (or not, If deceit is
the goal) the strategy. When a gesture can stand for two strategies
its clarity can be said to decrease as the probability of Its
assignment to one of the strategies approaches .5. For example, a
0 choice in the PDG without further Information, can only tell the
observer that a subject plans to maximize his own gain (since T is
the highest payoff) or that he wants to maximize relative gain (since
T-S Is the biggest differential payoff).

It Is hypothesized that the Information gained by a knowledge

of the other person's choices In the P00 serves as gestures about

85

that person's state or strategy. Subjects are hypothesized to act
on this information in order to advance their own best interests

in a PDG with monetary payoffs. This hypothesis, at least in part,
is in opposition to the hypothesis that Rapoport and Chammah put
forward. Specifically it is maintained that subjects will imitate
the other only so long as it remains in their best interests to do
so.

The strategies of a hypothetical person in the PDG can be
placed along two dimensions--a cooperative dimension-and a martyr
dimension. Thus from a player's choice one can infer that he is
cOOperatIve (C) or not (C). At the same time one can infer that he
is in a state of martyrdom (M) or not (H). Since these strategies
are independent, four combined strategies are possible: CM, CM}

CM, CH: The psychological interpretation is straightforward for
three of these strategies. CM would describe a player who is
unconditionally cooperative, whether the other subject is cooperating
or defecting. CH'refers to a subject who will cooperate only as long
as the other player does. In other words such a player would be
unwilling to receive an S payoff very often. CH describes a strategy
for a player who Is not willing to receive less than a maximum reward.
The type Of player who uses this strategy hopes to maximize own gain,
or relative gain between subjects, without regard to social inter- ,
dependence. CM does not seem psychologically feasible since such

a subject would be unwilling to settle for even moderate social
payoffs and, instead, looks always for the other subject to win.

A number of the following hypotheses have been substantiated

for short runs of PDG plays. In view of Rapoport and Chammah's

86
(l965b) findings that there is an abrupt change in game behavior
following trial 30 they need to be confirmed for a much larger
number of trials. While Gahagan and Tedeschi ran their game for
110 trials, on an unconditionally cooperative strategy, they did
not pay their subjects nor did they include conditionally cooperative
strategies for comparison. It is still unresolved if the paying of
subjects serves to motivate them. In view of the Gallo and McClintock
review (1965) one can expect subjects to behave differently in monetary
payoff conditions.

Hypothesis 1. As the clarity of the gesture increases with
respect to a given strategy, that strategy of the player will have
a stronger effect on the frequency of cooperative responses of the
other player for a given displayed PDG matrix.

Hypothesis 2. When the type of strategy is ignored no
difference among the groups in the frequency of cooperation will be
due to clarity alone.

Hypothesis 3. Frequency of cooperative responding will be
lower in subjects who face a partner employing a CM strategy than
for subjects who face a partner employing a Cfi'strategy.

Hypothesis A. The frequency of cooperative responding will
increase as the game progresses for subjects facing a partner employing
a CM'strategy and will decrease for subjects who face a partner
employing a CM strategy.

These hypotheses represent gross predictions about the behavior
of a subject operating in a mindful way as opposed to the behavior
of a subject operating In an Imitative way. But the theory ls
sufficiently precise that more rigorous predictions can be made about

the trial to trial behavior of a subject.

37

52921: The model specifies that the behavior of the subjects
can be divided into two psychological processes, the activation
process and the decision process. The activation process describes
the way in which four events or stimulus conditions are related to
three strategies that are postulated to describe the intentions of
an opposing player in the PDG. By intention is meant the meaning of
another player's choices. The decision process describes the action
that can be taken once the intentions of the other are known. In
specifyingthe model for the PDG, four stimulus conditions are sufficient
to activate three strategies or meanings which are then related to two
choices in the decision portion of the psychologiCal process.

The activation portion of the model is given by Equation A.

It is assumed that a subject uses the events from two trials as the
basis for activating one of three possible strategy states. On any
trial, n-l, the subject, S, can choose C or D in a PDG game. On any
trial, n, the other player E can choose C or 0. Then four possible
joint stimulus events can be considered. A cooperative response by
the other can follow a cooperative response by the subject

(C /c5,n-l)3 a cOOperatIve response by the other can follow a

E,n
defectlng response by the subject (°5,n/°s n-l)‘ a defectlng response

by the other can follow a cooperative response by the subject

(0 /C I); finally a defecting response by the other can follow

E." s,"-

a defecting response by the subject (DE n/DS n-l)°
B 9

CM CR '61?
cEqn/CS,n-| l-s e O
C /O _ s l-s 0
AI E'n S," I (H)
DE.n/Cs’n-1 O l-s s
O /D 0 s l-s

E,n S,n-l

88
These stimulus conditions are given as starting the four rows on
the right hand side of Equation A. The three strategy states head
the columns of the matrix in Equation A. These strategies are
represented by the labels CH, CH} and CH1 Thus if on trial n-l the
subject chooses C and the other player follows with a C choice than
the function designated by row one is in effect. That is, the other
player is attributed with probability s to be usinga CM'strategy and
with probability l-s to be using a CM strategy. The model predicts
the other player is never cast into the CF condition.

If on trial n-I, the subject chooses a D and this is followed
by a C on the part of the other player then the second row of the
transition matrix is brought into play. Under this rule the subject
assigns the other player to the CM strategy with the probability 5
and with probability l-s to the CH'strategy. It is considered that
given these stimulus conditions the other player Is never attributed
to being in the Cfi'strategy by the subject. The A matrix can be
examined for the remaining two rows by the same methods in order to
acquire an intuitive understanding Of the activation process.

The construct validity of the transition Into the three
postulated states Is straightforward. If the programmed player
follows a C response on the subject's part with a C.re5ponse then
the subject can Infer that the other player Is in an unremittingly
cooperative state and that he is open for exploitation. This state
is labelled CM for a cooperative martyr. The subject can also Infer
under other circumstances that the subject Is in a cOOperatIve state
but will not tolerate a doublecross on the part of the subject. This

state Is labelled as Cfi'for a cooperative but not a martyr strategy.

89

'The CH label would characterize a strategy wherein the
programmed other player is seen as being both uncooperative and not
a martyr insofar as his playing style is concerned. A player who
followed a subject's C choice with a C choice, according to the
model would never be perceived as uncooperative and nonexploitable.

An analysis of the other stimulus conditions as presented in
the rows of matrix A lead to similar formulations. ’The second row
describes a stimulus condition that has the subject assigning the
other player to the same two states as in row one, but with reversed
probability. The necessary stimulus for this row is the occurrence of
a 0 response by the subject on trial n-l. If the other player responds
with a C on the next trial then the subject is construed as placing
the programmed player with probability 5 into the CM strategy. With
probability l-s the programmed player is assigned by the subject to
the Cfi'strategy. The latter statement is warranted since the subject
may see the other player as trying to lead them out of a dangerous
mutual loss situation.

'The third row of the A matrix specifies the function rule when
the subject chooses C on trial n-1 and the programmed player follows
with a D on the very next trial. The model calls for the subject
to attribute a noncOOperatIve and unmartyrlike strategy (Ci) with
probabhllty s to the programmed player. This combination of choices
Is seen by the subject as doublecrossing behavior on the part of the
programmed other player. At the same time depending upon past events,
suchbehavlor can be Interpreted by the subject as an attempt on the
part of the other player to "get even.” So with probability l-s, under

these conditions the programmed player is placed in the CH'state.

90

The fourth row of the matrix specifies the mutual defection
condition of the two players. If the other player follows a subject's
D response with a 0 response of his own then the model predicts that
with probability 5 he is being assigned a cooperative but nonmartyrlike
strategy (CH) by the subject. Under the same stimulus conditions the
programmed player is cast into a noncooperative and nonmartyrlike
strategy with probability I-s.

Again depending upon the recent history of the game, the
subject can assign the other player to what has been called the
minimax strategy. Also depending upon the game history, he can
assign the hypothetical other to a strategy which will not tolerate
exploitation but will practice mutual cooperation (Ci).

While the A matrix specifies the hypothetical cognitions
of the subjects it does not address Itself to the way these
cognitive states affect the subjects' own responses. A decision
matrix (D) will be constructed which will predict the subject's choices

as a function of his cognitions of the other player's strategy.

I: 0
CM 0 I

0- CM qn Min (5)
ER - o l

The Interpretation of this matrix Is straightforward. Consider the
first row. It can be presumed from previous experiments that If
the other player was construed as a pacifist he would be exploited.
Thus subjects could be expected to choose 0 with a probability 1.
At the same time such subjects would be unlikely to adopt a

martyr role themselves. Thus If they found the otherplayer uncooperative

91
as the third row of matrix D specifies, they would remain uncooperative
themselves by choosing O with probability 1.

The second row of the 0 matrix denotes the intermediate strategy
of cOOperatIve but not a martyr. Under these conditions the subject
who is out to maximize his gains can be expected to adjust his own
level of X responses to that of his partner in an attempt to yield
maximum gain.‘ Such an adjustment is expected to be gradual as suggested
by the subscript n affixed to the q in the 0 matrix. For the moment,
the rate of change is assumed to be linear and will be specified in
greater detail in the chapters to come.

Experiments 2 and 3: Problem and Hypothesis

2 Very few PDG experiments investigate the effect of varying
the nature of the opponent upon subjects' choices. One study by
Marlowe, Gergen and Doob (1966) found that the degree of humility
of the opponent on a humility-egotism dimension affected cooperation.
Subjects exploited the humble other player when no postgame interaction
was anticipated but exploited the egotistical other person when postgame
interaction was expected. Kanouse and Wiest (1967) could not find
any significant difference in a one trial game that could be attributed
to the sex of the other player or the sex of the subject. Rapoport
and Chammah (l965b) found males to be more cooperative than females
when they faced members of their own sex. This difference disappeared
when the males and females faced a subject of the Opposite sex. The
game was played for a large number of trials and the other was not
simulated. Tedeschi, Lesnick and Gahagan (1968) found no difference
due to sex of the subjects In cooperation over a 100 trial PDG. In

this experiment the other was simulated and chose C 50% of the time.

92
In addition the game was highly competitive as measured by the
incentive value of the matrix. Several studies (Komorita, 1965,
Tedeschi, Lesnick and Gahagan, 1968, Bixenstlne and O'Reilly, 1966,
Rapoport and Chammah, I965b, Bixenstlne and Wilson, 1963) reviewed
previously, found males to be more cooperative facing a tit-for-tat
strategy while females were more cooperative than males when the
other was unconditionally OOOperative. Little or no information is
available about the effect of the sex of the simulated other on the
choice behavior of males and females in a contingent and noncontingent
PDG situation. Contrary to the findings of Tedeschi, Lesnick and:
Gahagan it is predicted here that males and females perform differently
from one another with males being more cooperative against the
contingent strategy while females are more cooperative against a
noncontingent other. No predictions are made about the effect of the
sex of the opponent.

Messick (1967) found subjects employed very effective counter
strategies in a zero-sum game where subjects were told they were playing
a computer. It has been reported that subjects come to suspect the
other In a PDG where the choices of the other are arranged according
to a simulated strategy. (Bixenstlne, Potash and Wilson, 1963,
Solomon, 1960) No effective assessment of this suspicion on the
subject's choices has yet been achieved. One way of assessing the
effect of the experimental deception Is to compare the actual
performance of subjects who are told they face a simulated other
against the performance of subjects who are led to believe they are

facing another subject.

93

In terms of Mead's theory and Scheff's theory of social
consensus, communication of meaning or intent behind human actions
is assumed possible. Therefore as long as one assumes he is
interacting with another human the meaning of an action must convey
the subject's-own intention to cOOperate in order that the other
does not retaliate. It is therefore expected that, for subjects
playing against a supposed human other, less effective counter
strategies will be used than if subjects are told they are playing
against a simulated other.

To test the two exploratory hypotheses presented above, two
experiments were carried out. Subjects were assigned to the
experimental conditions as outlined in Chapter Two. For both
experiments some of the experimental groups used in Experiment I

were used In the data analysis. The apparatus is described in

Chapter Two.

 

CHAPTER TWO
Method

Experiment 1: Method

 

Subjects. Subjects were recruited for the experiment by
means of an advertisement placed In the Michigan State University
newspaper over a two week period during Spring term, 1969. Under-
graduate subjects who were interested in participating in a social
game experiment, not taking more than one hour, were requested to
call a given phone number. In return for their participation the
advertisement promised that subjects could earn up to $2.50. When
subjects Called this number a secretary was available who recorded
the student's name and telephone number. The caller was also
classified by sex. The times when he (or she) was available for
the experiment were recorded. He was then instructed by the
secretary that an appointment would be arranged by phone within
the following two or three weeks. This delay was eXplained to be
necessary since the experiment Involved a small group of people whose
available times had to coincide. In response to any questions about
the experiment she was told to state that the experiment Involved
no hypnosis Or electric shock, that It was found to be interesting
by the majority of people and beyond that, that, she (the secretary)
did not know any details.

The sheets for male volunteers were then grouped by threes
according to available times Indicated on the sheet. Subjects were
contacted by the experimenter and an appointment was arranged. it

was stressed on this second call that It was extremely Important that

9A

95
a subject be there since a small group was necessary in order to run
the experiment. The same procedure was used for the second experiment
except that groups were formed in which there was one female and two
males, or two females and one male, or three females, as the experi-
mental design called for. In the third experiment any number of males
was deemed acceptable but an effort was made to have at least two
males run in each time period. On any given day there were ten
available one hour periods In which the experiment was scheduled.
Subjects were instructed to come to the experimental room in the
Psychology Research building and were given appropriate instructions
as to the location of the building and room. Within experiments,
groups of subjects were randomly assigned to each of the experimental
conditions.

Apparatus. The experimental room consisted of a main chamber
and six side chambers, three on each side of the main chamber. All
six chambers were Interconnected by means of a conduit. Four of the
side chambers were used In the experiment; one contained electronic
equipment while the remaining three were each provided with one
table, one chair, a pair of stereophonic earphones and a small display
and response cOnsole designed to convey the subjects' responses to
the Laboratory Control Apparatus (LCA) and to convey Information
from the LCA to the subjects by means of appropriately labelled
lamps. The console was connected to the LCA by means of telephone-
type fifty pin connectors fitted to both ends of a cable which ran'
from the console to the conduit In the wall and through the conduit
to the LCA. The console was placed on the table In such a way that

the subject response button labelled X, was on the left and the

96

button labelled Y was on the right. In terms of the traditional PDG
labels the X button represented the cooperative (C) choice while the

Y button represented the competitive (D) choice. immediately above
these two buttons and centered between them was an amber ready light,
labelled “ready.” The LCA automatically initiated each trial, after
the subject's response had been made and recorded, and the appropriate
feedback given to the subject. Feedback consisted of two types. The
first type was delivered by a row of four lights situated in the middle
of the box just above the ready light. Two of these lights were green
and were equally spaced to the left of the vertical central line on

a horizontal line in the middle of the box. The other two lights

were red and were placed in the same way on the same horizontal
midline but located to the right of the vertical centre line. This
row of four lights was used to convey to the subject the outcomes for
both subjects of any particular trial. If the subject had pressed
button X and the other player had chosen X the leftmost green light

in the middle row was turned on. Labels below this light indicated
that both players had won lc on that particular trial. If the

subject had pressed button Y and the other player did so as well the
rightmost red light was turned on. This informed the subject that
both he and the other player had lost lc. In the event that the

other player responded with an X on a trial and the subject responded
with a Y the remaining green light was turned on and informed the
subject that he had won 2c while the other player had lost 2c. The
remaining red light in this middle row was turned on when the subject
responded with an X response while the other player pressed his Y

button. The subject then knew that he had lost 2c and the other

97

player had gained 2c for that trial. Above the row of outcome lights
were two more lights, one green and one red. These lights informed
the subject of the other person's choice. The green light which was
positioned to the left of the vertical centre line was labelled with
an.X and indicated, when on, that the other player had chosen his X
button. The red light was on the right side of the vertical centre
line and, when on, indicated that the other player had responded Y.

A trial began with the onset of the amber light which was turned off
in each room by the LCA after each subject's response. When the
subjects who were present for that session had all responded, the LCA
fed back information indicating the other player's choice and the
outcome for each subject and his other player. The two lights giving
feedback to the subject remained on for an arbitrary three second
interval. Their onset immediately followed the last subject's
response in that group. The amber ”ready” light was turned on by

the LCA one and one-half seconds following the offset of the feedback
lights. The offset of the amber light was determined by each subject's
response latency.

The LCA is a four by three by five foot portable electronic
machine. its inner workings contain a set of standard commercialized
printed circuits which are connected to a 32 x 50 MAC Panel. The
printed circuits are designed to perform a particular logical
operation or a combination of logical operations. They have the
capability of communicating with the MAC Panel through connecting wires.
The MAC Panel itself can join these logical circuits by simply inter-
connecting the matrix of holes on the MAC Panel with a set of

flexible wires. Output from the logical circulty was delivered from

98

the LCA to both the subject's console and a Priden paper tape punch.
Input in the form of logical bits was received by the MAC Panel
from an eight channel paper tape reader attached to the LCA and
from the subjects' consoles. A description of the purely technical
' capabilities of the LCA as well as additional functions it can
perform will be available in a forthcoming technical report. (Kenoyer,
C. and Mendeisohn, 6., technical report in preparation).

The LCA located in the experimental chamber served a number
of functions for the present experiment. Primarily It was programmed
to simulate for each subject the other player who played according to
one of four possible preprogrammed strategies. The LCA received the
subjects' responses and controlled a Friden eightchannel Paper tape
punch. Thus each subject's response and that of the programmed other
player could be automatically recorded for each trial. As already.
mentioned the LCA fed back to the subject, by means of the console,
information about joint outcome, the choices of the other player
an<i initiated the next trial. The LCA was also programmed to perfbrm
the above functions for experimental units of from one to three players.

The design of the study called for the choices of the other
player to assume a certain conditional frequency. This conditional
frequency had several values for the several conditions of the
experiment. The LCA was programmed to allow the preprogrammed choices
of the other player to be punched in channel one on a paper tape for
the length of the game. Channel eight was also punched on all tapes
prior to the first trial and just after trial l50. The experimental
design also called for the game to be interrupted after trial 23 in

one half the conditions and after trial 30 In all conditions.

99

Channel eight punches were inserted after the apprOpriate trials for g
each of the conditions. The paper tape reader mounted on the LCA
recognized such channel eight punches as a cue for stopping the game
and signalled the experimenter that such was the case. A signal from
the experimenter to the LCA was necessary before the LCA would accept
a new response from any of the subjects. The punches in channel one
were recognized in different ways by the LCA depending upon a two
position switch controlled by the experimenter. If the experimental
design called for a group to be in the noncontingent condition then .
the switch was placed in position one. The LCA then recognized a
punch in channel one of the paper tape as an X response on the part
of the simulated other player. If no punch was present the LCA
was programmed to play a Y response on the part of the simulated
player. Thus, in the noncontingent condition all players in a given
group were given the same X or Y response on the part of the other
player. if the experimenter placed the switch in position two, the
LCA was programmed to respond in a contingent manner to the subject.
Under the contingent condition a punch in channel one of the paper
tape meant that the programmed other player did what the subject
had done on the previous trial. If the LCA read no punch it chose
the opposite response to the response the subject had chosen on the
previous trial.

For each of the two strategies two types of tapes were punched.
The first type reflected a high clarity level of a particular
strategy while the second type of tape reflected a low clarity condition.
On a given tape, clarity was decreased by decreasing the frequency of

punches in channel one. In the high clarity condition 902 of the

100

trials contained punches whereas under low clarity Conditions only
702 of the trials contained punches. These tapes were punched by the
experimenter prior to the start of the entire experiment. For each
level of clarity under each strategic condition, two tapes were
punched. These were subject to the clarity and strategy conditions
but were generated by a random number table. These eight tapes contained
a punch in channel eight before the first trial and just after trial 30
as well as after trial lSO. A duplicate set of tapes, which added a
channel eight punch after trial 23, was also constructed to allow for
the introduction of a message just after this trial.

In addition to the LCA and its input-output devices the apparatus
chamber contained a Vollensak tape recorder which was outfitted with
a multiple output jack. This jack was connected to each of the subjects'
sets of earphones by means of the conduit connecting the experimental
chamber with each of the subjects' chambers. Also, the experimenter
was equipped with a pair of earphones with which to monitor the output
from the tape recorder. The tape recorder was used to convey pre-
recorded instructions toail subjects participating in an experimental
session.

Procedure. After arriving at the experimental room the subject
was conducted to one of the side chambers. There he was Instructed
in the use of the earphones but told not to put them on until told to
do so by the experimenter. He was told that four people were assigned
to that particular time period but that as few as two could ensUre
. that the hour would not be cancelled. All sessions were started
(or cancelled) within ten minutes of the appointed time. If two

lappeared the subjects were Informed of this fact and the session begun.

i0l
If three appeared, a pretence of a fourth subject was made and the
session begun. 'That is, just after the third subject was admitted
the experimenter went through the motions of seating a subject in the
fourth chamber. (This was possible since the subjects had no visual
contact with a fourth chamber, and by virtue of the fact that there
was often a separate experiment being run in yet a fifth chamber.)
If only one subject appeared the session was cancelled and the subject
told he would receive by mail a full $2.50 for appearing.

When a group had been formed the subjects were instructed to
put on their earphones as the experiment was about to begin. The
experimenter then retired to the apparatus chamber and started the
tape recorder with the instructions to the subjects as required'by the
experimental design. The instructions given to the subjects were as

follows:

If the volume level on your earphones is uncomfortable
just adjust the two knobs on your earpieces until a comfortable
level is reached. Now let us begin. You are about to play a
game with another player of the same sex in one of the other
rooms. He is now listening to the same instructions you are.
Since there is more than one team playing at a time neither
one of you will know who he is playing with. Both of you can
make as much as $2.50 by playing the game.

Before you is a small box that is connected electronically
to a machine in another room. The other player has a similar
box connected to the same machine. The purpose of the machine
is to record both your response and the other player's response,
and to record how much you both make and also to give you
information.

The first piece of Information it will give you is when to
respond. You respond when the yellow ready light goes on.
You can then make a choice by pressing one of the two buttons
which are labelled X and Y at the bottom of the box just below
the ready light. When you and the other player have both
responded the machine will supply two other pieces of information.
First, one of the two top lights will light up indicating whether
the other person made an X or a Y choice. Thus if the green
light lights up that means the other player has chosen an X
button. If the red light lights up that means he has pressed
the Y button.

l02

The final piece of information the machine gives is how
much money both of you made. In the middle of your box are
four lights in a row. Now if you press button X and the
other person presses button X the leftmost light in the
middle row will go on. Underneath this light is a sign with a
lc immediately below the light and another l¢ just a little
lower down. The first l¢ tells you how much money he got and
the second lc tells you how much money you earned. This means
that since both of you pressed X you both get lc. If you both
press button Y the rightmost light will go on. Below this light
is a -lc and below that is another -I¢. This means when you
both choose Y you both lose lc. ,

Now the second light from the left in the middle row lights
up when you choose Y and the other person chooses X. Below
this light is a -2c and below this is a +2¢. This means that
the other player loses 2c and you make 2c. The third light
from the left lights up when the other person has pushed button
Y and you have pushed button X. Below this light is a +2¢ and
below that is a -2¢. This means that he makes 2: and you lose
2¢.

You will be able to make as much as $2.50 for less than one
hour of participation. However to do so you must play your
cards right since how much you make will be determined both
by your choices and the choices of the other player. Also,
how much he makes is dependent on the choices you make. Some
people make more than others, some less. To begin the game both
of you will start with 50¢. The machine will keep track of
how much you make and a check will be mailed to you for the
correct amount.

 

Each of you will be allowed to communicate by note with the other
one, early in the game, about how he feels the game should be
played. '

 

In addition after you have played a number of trials I will ask
you to fill out a form indicating what you feel the other player
is doing.

To sum up, this is a game for two people played a large number
of times. Each time the ready light goes on, just above the
two choice buttons, you are to press either the X or Y button
but not both. After the other player has made his choice the
two lights In the top row will tell you whether he pressed the
X or the Y button. The four lights in the middle tell you how
much you both have made. For example if you both press Y, the
light on the right will go on and you both lose l¢. These
lights will remain on for a short period of time and then go
off. Then after a short period of time the ready light will
come on and you make another choice. The machine will keep

l03

track of your earnings. You can make as little as nothing

or as much as $2.50 depending upon how you play the game.

Take your time. There is no rush. Make your choice when

the yellow ready light goes on.

The design called for half the grOUps to communicate written
messages to the other player. The instructions for this condition
were the same except for the sentence inserted indicating that notes
would be exchanged.

After the instructions had been completed the experimenter
initiated the experiment by activating the LCA and the Friden tape
punch. Subjects responded to the ”ready" light and the experiment
was run off until the LCA signaled the experimenter that the game
had been stopped according to the experimental plan. For one half
the groups this occurred just following trial 23. At this point the
experimenter delivered a blank sheet of paper and pencil to each
subject and instructed each of them to write a note to the other player
instructing him how the game was to be played. Each subject under this
condition was informed that he would exchange these notes with the other
player. The experimenter acted as a messenger and when out of sight
from the subjects substituted one of the following notes depending upon
which condition the group happened to be under. Subjects in the non-
contingent condition received Note l. 'Subjects in the contingent
condition received Note 2.

fig£g_l, If we both choose X it seems to me that will
give us the most money. Pressing button Y seems to be a

hopeless situation so i will press X much more than Y no
matter what you do. .
Note 2. If we both choose X it seems to me that wll
‘ give us the most money. Pressing button Y seems to be a
hopeless situation but I won't continue to press X if you
don't.

loll

Subjects were allowed ample opportunity to read and write the notes

but no subject received a note until he had written his. After the
notes had been read they were collected by the experimenter who then
returned to the experimental chamber. The game was then continued
until the LCA stopped for all groups following trial 30. At this

point the experimenter delivered the Semantic Attribution Scale (SAS)
to each subject. This scale as it was given in the experiment is
reproduced in Appendix A. The scale was designed to measure impressions
that the subject held about the other player. Both the scales and the
instructions were closely modelled after Osgood's semantic differential
(Osgood, Suci, and Tannenbaum, l957). Osgood et al have shown that

a similar set of words are capable of yielding a multidimensional
structure in a variety of contexts and when the analysis uses a
variety of methods to yield the structure. The multidimensional
structure usually factors into words reflecting subjects evaluations
of the events rated and of their relative activity and potency.

Both the words and instructions were modified to make them suitable
for the present study.

After the scales were completely filled out by all subjects
the experiment was continued and the game completed.~ After a total
of .l50 trials had been completed for all groups the LCA signalled
the experimenter that the game was complete. The experimenter then
asked the subjects to fill out a SAS form and final questionnaire.
The, SAS used was the One found In Appendix A and used early in the
game. The final questionnaire is presented in Appendix B. Its
composition was as follows: Part A of the questionnaire obtained

the information needed in order to mail a check to the subject in

l05

payment for his participation in the experiment. The four questions
of Part B obtained the information from the subjects about their
biographies that might be relevant to the data analysis. Questions
I, h, 6 and ID of part C were designed to get at the perception of
the other person's strategy. Questions 2, 3. S, 7, 8 and 9 were
designed to elicit information about the subject's own strategy.
Questions ll through l6 were intended to elicit information about
the experimental deception. After these final forms were filled out
the subjects were permitted to leave and the apparatus was made ready
for the next group of subjects. After the three experiments had been
completed, checks were made up by the university business office for
the entire $2.50. These were then mailed to all the subjects at the
address they had indicated in the final questionnaire. Along with
the check went an explanation of the Prisoner's Dilemma Game and the
reasons for both the experiments and the experimenter's deception.
This explanation is contained in Appendix C.

Degjgn and analysis. The independent variables investigated
in Experiment l were the effect of the other player's strategy,
the clarity with which the other player presented that strategy, and
a written communication verbalizing the strategy employed. Two levels
were used for each of the independent experimental variables. The
game was played l50 times by each subject. There were l5 subjects
per cell for each of the eight independent cells in the experiment.
Thus the design constituted a 2x2x2xl50xl5 factorial design with the
last factor considered random and the first four as fixed. The
fourth factor was trials and repeated measures were taken on the

same subjects over all trials. These measures cannot be considered

l06
as independent of one another. This fact introduces special problems
for the quantitative analyses which will be discussed later.

The first factor in the experiment represented the strategy of
the other player. Strategy was manipulated by changing the conditional
probabilities with which the programmed player followed a subject's
X or Y response. The probability that the programmed player played
an X on trial n, given that the subject had chosen an X on trial n-l,
was set equal to w. In notation, letting C stand for X and 0 for Y,

w.P(cn,E/cn-l,5) (6)
where the subscript E stands for the programmed other player and S

denotes a subject. The probability that the other player would play

an X on trial n following a Y by the subject was set equal to y. Thus:

-P

Y (cn.E/Dn-1 S) (7)
Then by the laws of probability:

l-w-P(0n,E/Cn_l’s) (8)
and l-Y'P(Dn,E/Dn-I,S) (9)

To simplify notation for purposes of discussion the vector Bi is
defined as:

Bi-(w, y, l-w, l-y) (10)
When w was set equal to y the programmed player was said to be playing
a noncontingent strategy. When w was set equal to l-y the preprogrammed
player was said to be playing a contingent strategy. From Equations
6 and 7 it can be noted that the programmed player played X with
probability w regardless of what the subject did when w was set equal
to y. From inspection of Equations 6 and 9 it can be understood that
the programmed player chose what the subject had chosen on the previous

trial, with probability w when w was set equal to l-y.

l07

The second factor of the design was concerned with the clarity
of the first factor. For the high clarity, noncontingent manipulation
the values of w and y were determined as follows:

314.9, .9, .l, .1) (11)

For the low clarity, noncontingent manipulation the values of a w and
y were lowered as shown by Equation l2:

324.7, .7, .3, .3) (12)

For the high clarity contingent condition the values of w and (l-y)
were given in Equation l3:

33=(.9, .l, .1, .9) (13)

For the low clarity contingent condition the vector Bk was used.

Bh'('7’ .3. -3. -7) (lh)

The third factor represents the presence or absence of a message
exchange condition after trial 23, where the programmed other player
advocated increased usage of the X button.

The fifth factor was considered random and represented l5 subjects
for each of the eight independent cells in Experiment l. Since the LCA
was capable of playing only one strategy per experimental session it
was impossible to assign subjects at random to conditions. However
groups of subjects were assigned randomly to the experimental
conditions. A total of 297 subjects arrived for the three experiments.
3eventeen of these subjects were not used because of apparatus failure
or because only one subject appeared. 0f the remaining 280 subjects
126 were used In Experiment l. (The last six subjects were booked
to complete groups and were excluded from the analysis to obtain an
equal number of subjects per cell.) The analysis of the experimental

data was divided Into several parts according to the methodology and the

l08

Vdependent variable. It was predicted by Head's theory that the
attributions made of the programmed player's behavior would be
stabilized early in the game or, more specifically by the occurrence

of trial 30. Consequently a 2x2x2xl2 fixed effects analysis of
variance was carried out on the number of times the subject pressed

the X button. The first three factors of this design constitute

the two levels of contingency, the two levels of clarity and the two
message levels. The fourth factor referred to blocks of trials. A
number for each block was arrived at by summing the number of X responses
for the ten trials, starting with trial 3], making up the consecutive
blocks. This design is essentially the Case ll analysis of Winer where
there are multiple factors and repeated measures on one of the factors
(Winer, l962). Conservative F-tests as outlined by Greenhouse and
Geisser (l959) were employed with the F ratios of the repeated measures
resulting from the usual analysis of variance procedures.

In addition to the tests of the general effect of the treatments
more specific predictions of the trial to trial changes of the subjects'
behavior were generated by a decision theoretic model shaped from the
formulations of minded behavior as set forth by G. H. Mead (l93h).

The decision theoretic approach has been used successfully in models
of signal detection behavior and was thought to be compatible with
the present more molar approach to social behavior.

The theory as applied to the Prisoner's Dilemma suggests that
the subjects are in one of three cognitive states depending upon how

they perceive the other player's strategy for any two adjacent trials.

In turn the theory specifies, by means of a decision matrix (see

Equation 5), how these cognitions influence the subject's X and Y

109
choices. Furthermore the theory calls for these cognitions to arise
relatively early in the game and to be two dimensional in character.

To assess these last predictions a multiple discriminant
function analysis was performed on the 30 trial semantic attribution
scale data. The discriminant analysis approach was first developed by
Rao (l952) and is presented by Cooley and Lohnes (l962). This method
can be applied to data for which there are several measures on groups
of people known to differ on some a priori dimension such as sex or
pathology. The first discriminant function finds the best weighted linear
combination of the measures that will maximally discriminate among the
groups. This is done by selecting those weights of the measures which
maximize the ratio of the between groups variance to the pooled within
groups variance. The variance due to this first function is then
extracted from the measures and a second function is computed on the
residual variance. If there are J groups J-l functions can be derived.
Rao has derived a chi-square test to ascertain whether the functions
will reliably discriminate among the groups in a manner that cannot
be due to chance.

Experiment 2: Method

 

This experiment varied the sex of the subject, the sex of his
or her opponent and the type of strategy that the opponent employed.
éoth males and females were recruited as subjects and they were told
that their Opponent was either male or female. In reality the opponent
was a simulated strategy of either the contingent or noncontingent
type and that strategy was carried out under high clarity conditions.
No notes were exchanged In this experiment. Thus 30 male subjects

were led to believe that they faced a male opponent while another 30

llO

believed that they were playing with a female. In fact l5 subjects from
each of the two groups played against a simulated strategy that was
contingent and another l5 in each group played against a noncontingent
strategy. Sixty female subjects faced exactly the same experimental
conditions. Subjects, as nearly as possible, were selected randomly

for the experimental treatment groups. Groups were randomly assigned,
within the limits of the design, to experimental conditions.

The same procedure was followed as in Experiment I with the
exception that the instructions were modified on the tape recorder
to take into account the sex of the opponent. For four of the groups
the pronouns “he or she” were inserted in place of the pronoun ”he”
and the words ”of the same sex” were changed to ''of the opposite sex.”
For the two all female groups the pronoun “she” replaced the pronoun
“he” throughout the main experimental instructions.

For purposes of the analysis of variance a 2x2x2xl2xl5
factorial design was utilized with repeated measures on the foUrth
factor, trials, and the last random factor representing subjects.

The first factor (A) stood for the type of strategy (contingent or
noncontingent) that the subject faced throughout the PDG. The second
factor (B) was the sex of the subject while the third factor (C)
represented the sex of the opponent. The fourth factor (0) was used
'or the l2 blocks of ten trials that were used.

Experiment 3: Method

For this study four experimental groups were created and
compared with four comparable groups from Experiment I. This experiment
investigated the effect of facing a contingent-noncontingent strategy
(Factor A), the effect of the subjects being led to believe they were

playing a human or a machine opponent (Factor B), and the effect on

ill

the subjects of receiving a note supposedly clarifying the strategy
of the other versus those who received no note at all. The analysis
of variance, a 2x2x2xl5xl2 design, was exactly the same as the
previous two experiments. The second last factor represented the IS
subjects considered as randomly assigned to each of the cells and
the last factor represented the last l2 blocks of trials of ten trials
a piece. The dependent variable was the frequency of X choices across
the trial blocks.

The same procedure was followed as the previous experiments with
two exceptions for those subjects in the machine condition. First,
if only a single subject appeared the experiment was still carried out.
Out of a possible 60 subjects seven were run when no other subject was
present. Second, subjects were assured by the experimenter before the
instructions were read that they were playing against a machine. If
they wished they were allowed to view the machine. The following
instructions were then read to those subjects who were facing a machine.

If the volume level on your earphones is uncomfortable
just adjust the two knobs on your earpieces until a comfortable
level is reached. Now let us begin. You are about to play a
game with an electronic machine in another room. As many as
three players can play against the machine at any one time.
All of you are listening to these same instructions but in
no way will you be playing with or against one another.

The machine is capable of assuming several strategies or
ways of playing the game. These strategies represent the ways
that several different types of people play the game once they
understand it. You can make as much as $2.50 by playing the
game when the machine assumes these strategies. The strategies
are simple and straightforward and involve no trickery or deceit
on the experimenter's part. For example while the machine may
or may not pay attention to the way you responded prior to any
trial it does not have foreknowledge of your current choice.
That is it must make its choice on any one trial without knowing
what yOur choice is on that trial just as you don't know what
choice it has made until you have made yours. Now I will explain
the game itself.

ll2

'Before you is a small box that is connected electronically
to a machine in another room. In addition to being the other
player for you, the machine can record your response and the
amount that you make. Also it gives you some information.

The first piece of information it will give you is when to
respond. You respond when the yellow light goes on. You can
then make a choice by pressing one of the two buttons which
are labelled X and Y at the bottom of the box just below the
ready light. When you and the other players have responded the
machine will supply two other pieces of information. First,
one of the two top lights will light up indicating whether the
machine made an X or Y choice. Thus if the green light lights
up that means the machine has chosen an X button. If the red
light lights up that means the machine has pressed the Y button.

The final piece of information the machine gives is how much
money you and the machine make. In the middle of your box are
four lights in a row. Now, if you press button X and the machine
presses button X the leftmost light in the middle row will go
on. Underneath this light is a sign with a It immediately below
the light and another It just a little lower down. The first
l¢ tells you how much money the machine earned and the second l¢
tells you how much money you earned. This means that since both
of you pressed X you both get l2. If you both press Y the
rightmost light will go on. Below this light is a -l¢ and below
that is another -l¢. This means that when you both choose Y you
both lose It.

Now the second light from the left in the middle row lights
up when you choose Y and the machine chooses X. Below this light
is a -2c and below that is a +2c. This means that the machine
loses 26 and you make 2c. The third light from the left lights
up when the machine has pushed button Y and you have pushed
button X. Below this is a +2¢ and below that is a ~2c. This
means that the machine makes 2¢ and you lose 2¢.

You will be able to make as much as $2.50 for less than one
hour of participation. However to do so you must play your cards
right since how much you make will be determined both by your
choices and the choices of the machine. Some people make more
than others, some less. To begin the game you will start with
50¢. The machine will keep track of how much you earn and a
check will be mailed to you for the correct amount. After you
have played a number of times I will pass a note to you that a
person who was playing such a strategy might write if he were
communicating by note with you.

Remember you will be playing with a machine for a large number
of times. in addition after you have played a number of trials
I will ask you to fill out a form 23_jj_you were playing with
another person.

To sum ug, each time the ready light goes on just above the
two choice uttons, you are to press either the X or the Y button

but not both. After the machine has made a choice the two lights
in the top row will tell you whether it pressed the X or Y
button. The four lights in the middle row tell you how much

you and the machine have made. For example if you both press Y

ll3

the light on the right will go on and you both lose lc. These
lights will remain on for a short period of time and then go off.
Then after a short period of time the ready light will come on

and you can make another choice. The machine will keep track

of both your earnings. You can make as little as nothing or as
much as $2.50 depending upon how you play the game. The machine's
strategies are simple and straightforward. They involve no
trickery, deceit or foreknowledge on the machine's part. Take
your time. There is no rush. Make your first choice when the
yellow light goes on.

CHAPTER THREE
Results and Discussion: Experiment I
Results

The results can best be presented in four sectiOns. The first
section deals with an analysis of variance of the data from Experiment
I. The second section will present the data regarding the perceptions
of the other subject that the subjects held early in the game. The
third section will present the decision theoretic model and an assessment
of its accuracy. The fourth section will give the subjects' reactions
to the experimental deception.

Assessment of the experimental manipulations. The statistical
analysis of the frequency of the (subjects') X responses as a function
of the three bilevel independent variables is given in Table l. The
independent manipulation involved a note-no note condition crossed
with a contingent or noncontingent strategy which was communicated with
high or low clarity across l2 blocks of ten trials apiece. it will
be recalled at this point that the experimental hypotheses predict
that: first, the clearer the other person's strategy is the more
effect that strategy has on the cooperative choices of the subjects;
second, the frequency of cooperation is not affected by the clarity
level by itself; third, cooperation is lower for those groups who face
a CM strategy than for those who face a CM strategy; fourth, over trials,
when faced with a partner using a Cfi'strategy subjects increase their
cooperation and decrease it across trials when faced with a CM

strategy.

llh

llS

In connection with the first hypothesis, the note was expected
to clarify the other player's strategy. Thus in the contingent
strategy it was predicted that the note condition would raise the
frequency of (subjects') X responses above the no note condition.

For subjects who faced a noncontingent other, those who receiVed a note
were expected to lower the frequency of their X responses below that
level of responding shown by those who received no note. However no
interaction of the note condition with any other condition reached an
acceptable level of significance. Instead the exchange of notes itself
operated to raise the frequency of X responses. By sending and receiving
a note the frequency of X responses was raised from h.h9 to 5.l9 in an
average ten trial block.

From the between subjects portion of Table I it is seen that
variations in the Opponent's contingency pattern produced reliable
differences in the subjects' X responses. Subjects who faced a
contingent opponent chose X 5.96 times on the average trial block as
opposed to only h.hh such choices by subjects who faced a noncontingent
other player. This finding has to be qualified by the significant
interaction observed between type of contingency and the clarity of
the strategy. Subjects who played with another who chose according to
the high clarity contingent condition played X 6.93 times out of ten
gn the average while the subjects faced with a less clearly contingent
strategy chose X h.99 times out of ten. A clearly noncontingent condi-
tion resulted in subjects choosing X only 3.99 times out of ten while,
faced with a less clear noncontingent condition, subjects made the X
choice “.90 times out of ten on the average block of ten trials.

Subjects who faced a clearly noncontingent other player made fewer X

ll6
responses than those facing a less clearly noncontingent other when
no notes were exchanged (F=3.7, p<.08, df-l/llZ) as was predicted by
hypothesis one. This effect was only marginally significant when the
note exchange condition was ignored. (F-2.3, .25<p<.l0, df=l/ll2).
Under both note and no note conditions subjects who faced a clearly
contingent other were more cooperative than those who faced a less
clearly contingent other (F=l0.h, p<.0l, df-l/ll2). No reliable
difference was obtained between the group who faced a less clearly
contingent other and a less clearly noncontingent other.

Hypothesis l was partially verified in that as the opponent's
strategy became clearer the subjects' responses were increasingly
affected in the expected direction. However the exchange of notes
in the noncontingent condition produced an unexpected effect. Subjects
who received a note expressing the noncontingency of the other increased
their own level of cooperativeness.

Hypothesis 2 was verified in that the subjects who faced an other
who performed his strategy in a relatively clear fashion did not behave
differently from those subjects who faced an other with a strategy
defined as less clear. 0n the other hand subjects who played with a
contingent other exhibited more X responses than did those subjects
facing a noncontingent other. This verifies the predictions of
iypothesls 3.-

Effects taking place within subjects as the game continued
are given in the second half of Table l. Over all conditions there was
a statistically significant decrease in the number of X responses.

(From the first block to the last block the average number of X responses

was as follows: 6.29, 5.58, 5.39, 5.08, 5.09, 5.l8, 9.87, “.78, 5.20,

ll7
5.l5, 5.00, 4.86.) The type of contingency differentially affected
the rate of change of X responses over blocks of trials as is shown
by the significant contingency by trials interaction. The clarity by
trials interaction indicated that subjects changed their response
rate as the other person's strategy became more or less clear. Under
low clarity conditions, disregarding the type of strategy, the rate
of change is low while under high clarity conditions the rate becomes
high. Finally the second order interaction of type of contingency
and clarity of that contingency affected the rate of change across
blocks of trials. The contingent strategy under the high clarity
condition elicited a high degree of c00peration by the fourth block
and maintained that high level with a slight decline in the middle
block of trials. The contingent but less clear strategy started at an
equal level of cooperation but this declined over trials. Both non-
contingent strategies elicited about 55% to 55% X responses on the
fourth block of trials. In the high clarity condition this had declined
to 39% by the eighth block whereas it had remained at 54% on the low
clarity noncontingent strategy. 0n the remaining trials six out of
seven blocks remained at 45% or more X responses, whereas under the
high clarity condition all seven blocks were “0% or less X responses.
The low clarity conditions in both strategies do not appear to differ
substantially across blocks of trials. This pattern of responses is
not inconsistent with Hypothesis l although for the contingent low
clarity condition it runs counter to Hypothesis h. For that condition,
it was expected that subjects would increase their X responses as the

game progressed.

ll8
Table l
Analysis of Variance of X Responses

in Blocks of ID for Experiment l.

 

 

 

Source df MS F
Between Subjects
Note (A) 1 725.336 11.304**
Contingency (B) 1 825.069 12.859*
Clarity (C) l 95.069 l.h82
A x B 1 34.225
A x C l 63.336
B x 0 1 725.336 11.304**
A x B x C 1 55-225
Error Between ll2 6h.l65
Within Subjects
Trials (T) 11 20.207 5.855*
A x T 11 1.719
B x T 11 8.562 2.481***
c x T 11 6.653 1.928****
A x B x T 11 1 781
A x C x T ll l.856
B x C x T ll l2.hOl 3.593*
A x B x c x T 11 2.057
Error Within l232 3.45]

 

 

 

 

="p<.0005
*“p<.001
p<.005

""'§‘-€p< . 05

ll9
A major difficulty with the use of analysis of variance
procedures and the necessarily dependent measures taken repeatedly on
the same subjects as the game continues is that the variance-covariance
matrix must be symmetrical (Winer, l962, p. 305). A conservative test
has been developed by Greenhouse and Geisser (1959) which avoids
assumptions about equal covariances in the pooled variance-covariance
matrix. In their technique the same F ratios are used but the df used
to consult tabled F values are modified by dividing the usual df for
both the effect (numerator) and error (denominator) terms by the df
associated with the main repeated measure (in this case df=ll). This
test is generally considered overly conservative. When applied to the
present data the trials' effect remains significant at better than
I
the .05 level of significance. The first order interaction of type of
contingency by trials becomes only marginally significant (F=2.h8, p<.l3,
df=l/ll2). The level of clarity by trials interaction also becomes
marginally significant (F=l.93, p<.25, df=l/ll2). The second order
interaction of type of strategy by clarity by trials remains significant
(F=3.59, p<.07, df=l/ll2). This suggests that when the level of clarity
is considered, Hypothesis four is essentially verified. That is subjects
facing a CM strategy across trials decrease their cooperation level at
a rate dependent upon the level of clarity. Facing the CE strategy
.ubjects increase their cooperation when the level of clarity is high
and decrease it when the level of clarity is low. Data will now be
presented to show that the subjects' impressions of the other person
are not independent of the other person's strategy.

Assessment of social perceptions. The purpose of the discriminant

 

function analysis was to attempt to measure the subjects' impressions

of the other as a function of his strategy. The strategy, it will be

l20
recalled was expected to be subject to an error process as a function
of the clarity of that strategy. Therefore to varying degrees it would
be expected that the impressions of the other while clear would not be
great. That is, while a noncontingent strategy is associated with the
CM state, at the same time it has a positive probability of being
assigned to the CM state. However it can also be expected that subjects
reliably discriminate between strategies expressing the CM state and
strategies reflecting the CM state. Post hoc analysis of the words
involved in the discrimination should reflect the content of the CM
and CM strategies. The SAS was administered to all subjects following
trial 30. The subjects' ratings of the other player were then subjected
to a multiple discriminant analysis utilizing a standard computer program
(Thomas, l968). For the eight groups this program extracted several
discriminant functions of which the first two functions were found to
discriminate among the eight groups in a statistically reliable way.
The 30 bipolar variables and the standardized weights assigned to them
by the first three discriminant functions are presented in Table A.

Both of the two functions contain words loading heavily on
Osgood's evaluative dimension. The chi-square of 7i.“ for the first
function was significant (p<.0l, df=36) indicating that the groups
discriminated reliably among the various strategies they were facing.
n3 can be seen in Figure 3 three of the four noncontingent groups
(noncontingent, no note groups for both high and low clarity levels

and the noncontingent high clarity note group) and one of the contingent

.7
r
/

groups (contingent, low clarity, note group) are high on the first
function relative to three of the contingent (contingent, no note

groups at high and low clarity levels and the contingent, high clarity,

Group Means on those Words that have the

l2l

Table 2

Highest Weights on Discriminant Function l

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

Function Noncontingent Contingent
Weights No Note Note No Note Note Group
I 2 .7 .9 .7 .9 .7 .9 .7 .9 Mean
Believing .38 .111 3.8 2.9 3.8 2.714.}. 3.7 11.3 11.1 3.7
Skeptical
Independent r30 .09 4.3 4.9 3.7 4.8 2.7 3.3 3.7 3.5 3.8
Yielding
Grateful '30 .28 4.l 4.3 4.l 3.3 3.9 3.6 4.4 3.5 3.9
Ungrateful
Trustworthy -29 .2l 3.4 3.3 3.7 2.8 3.5 3.3 4.2 3.1 3.4
Untrustworthy_ I
Formed .29 .Ol 3.4 3.l 3.7 3.5 3.5 3.7 3.6 3.6 3.5
Formless '
Strong -28 -05 4.2 5.0 4.3 4.0 3.3 3.7 3.9 4.0 4.0
Weak
Meek .27 -05 3.8 3.3 3.8 4. 5.2 5.l 4.4 4.3 4.0
Aggressive ‘

 

 

 

 

 

 

 

 

l22
note group) and one of the noncontingent groups (the noncontingent, low
clarity, note group). Groups or persons who assigned the other a high
negative score on this function felt that he was relatively ”believing,
yielding, ungrateful, untrustworthy, formed, weak, meek, calm, and
bad.” The group means for each of the first seven words along with
their weights on functions are given in Table 2. It may be noted that
two of the words ”ungrateful and untrustworthy,” load substantially
(.20 or better) on the second function. Any interpretation of the
first function of the group means on these two words is confounded by
these two words' contribution to the second function. The following
results use the overall mean as a point of reference. In all four
contingent groups and the two low clarity noncontingent groups the
other is seen by the subjects as ”skeptical.” For the two high clarity
noncontingent groups he is perceived as ”believing.” All four contingent
groups perceived him as ”independent” while only the noncontingent low
clarity note group thought him to be so. The other three noncontingent
groups believed him to be “yielding.” The groups do not separate into
any distinct pattern when they rate the simulated other as formed or
formless. However all four noncontingent groups rated the other as
”weak” compared to all four contingent groups. In like manner all four
noncontingent groups saw the other as ”meek” as opposed to all four
contingent groups who saw the other as ”aggressive.”

The second discriminant function indicated that six of the
groups (all four noncontingent groups plus both high clarity contingent
groups) perceived the other person as similar relative to perceptions
of the other by subjects in the two low clarity contingent conditions.
The chi-square for this function reached 50.8 (p<.05, df=34). The

group means for the first eight words on this function are presented

l23
Table 3
Group Means on those Words that have the

Highest Weights on Discriminant Function 2

 

 

 

 

 

 

 

 

 

 

 

 

 

Function Noncontingent Contintingent
weights No Note Note No Note. Note Group)
I ' 2 .7 .9 .7 .9 .7 .9 ~.7 -<.9. Mean -
..._.__L..._
Faithful -.Ol -.37 3.l 3.] 3.3 2.7 4.2 3.2 4.5 2.9 3.4
Unreliable
Offensive -.I4 -.35 4.2 4.3 4.3 4.3 3.6 3.2 4.7 3.6 4.0
Defensive
Masculine .06 .33 3.7 3.7 4.3 3.2 2.9 3.l 3.3 3.4 3.4
Feminine
Good -.25 -.28 3.6 3.2 3.3 2.6 3.4 3.l 3.8 2.7 3.2
Bad
Grateful -.30 .28 4.l 4.3 4.l 3.3 3.9 3.6 4.4 3.5 3.9
Ungrateful 4.9
Stingy .I0 .26 4.7 4.7 4.i 4.9 3.8 3.9 3.l 4.4 4.2
Generous
Altruistic -.IS -.25 4.i 4.l 3.7 3.6 5.l 4.4 5.l 4.6 4.3
Self-serving;
Disreputable -.05 .23 4.8 4.7 4.7 5.2 4.3 4.9 4.3 4.8 4.7
Reputable

 

 

 

 

 

 

 

 

 

 

 

 

l24

in Table 3. Again two sets of bipolar adjectives “good-bad, grateful-
ungrateful” load heavily on the first function and thus the group means
for these words are considered uninterpretable. Both low clarity
contingent condition subjects viewed the other as ”unreliable” relative
to the four noncontingent and the two high clarity contingent groups.
Subjects in both no note contingent groups as well as the group who faced
a clearly contingent other who exchanged a note judged the other as
less defensive and more offensive than the other groups. All four of
the contingent groups as well as the clearly noncontingent who exchanged
notes saw the other as more masculine than the remaining noncontingent
groups (both no note noncontingent groups as well as the noncontingent
low clarity note group). The two contingent no note groups, the low
clarity contingent note group and the less clearly noncontingent note
group saw the other as ”stingy.” All four contingent groups but especially
the two low clarity contingent groups saw the other as being ”self-serving.”
These same two low clarity groups saw the other as being relatively
”disreputable.“

The third discriminant function yielded a chi-square value of
4l.2 (p<.ll, df=32). High loadings on this function indicated that the
other subject was seen as ”pessimistic, irrational, cruel, generous,
meek and meaningful.” Three of the noncontingent groups had high
discriminant scores on this function. The noncontingent groups that
were high on this function were both low clarity conditions and the
high clarity no note condition. 0f the contingent conditions only
the low clarity no note condition was high on this function. Thus

in three of the contingent conditions and the noncontingent, high

l25

Table 4

Discriminant Standardized Weights of Three Functions

Differentiating the Eight Experimental Groups

 

 

 

Bipolar Adjectives Function
‘ l 2 3
good--bad -.25_ .-.28_ ll
trusting--skeptical -.l0 -.05 -.l4
independent--yielding -.30 .09 -.06
optimistic--pessimistic -.04' .07 .36
sociable--unsociable .l7 .0l .08.
unfair--fair -.l0 -.07 .14
meek-~aggressive .22_ -.05 -.25_
disreputable--reputable -.05 .23_ .l3
passive--active -.I5 .2l -.l0
grateful--ungrateful -.30 .28: -.I3
pleasureable--painful [09 .ll -.l0
severe--lenient .09 .03 .08
irrational--rational -.09 .l7 -.45
believing--skeptical .38 .l4 .04-
offensive--defensive -.T4' -.35 -.02
trustworthy--untrustworthy -.22_ .2T' .04
st i ngy--generous . IO .26: ._2_9_
strong--weak -.28 -.05 -.23
masculine--feminine .06. .32_ .T6'
altruistic--self-serving -.i5 -.25 -.07
calm--excitable .26_ .T5' .03
formed--formless .29 .Ol -.l9
unsuccessful-~successful .06' -.09 .ll
wise--foolish .l4 .l3 .l7
kind--cruel .00 -.I8 .3l
changeable--stable -.l9 .05 -.20
meaningless-meaningful -.02 .l0 .2§_
cautious--rash -.09 -.l6, .l2
competitive--cooperative -.0l -.00 -.l5
faithful--unreliable -.0i -231 -.04

 

 

 

 

Note. - Underlined weights are above .20 and are used for inter-

pretation.

FUNCTION 2

 

FIG. 3.

I26

 

 

 

 

 

3
8
i,
l 2
6
5
group Strategy Clarity
l .7 no note
2 Noncontingent .9 no note
3 .7 note
4 4,9 note
5 .7 no note
6 Contingent .9 no note
7 .7 note
8 .9 note
-2 -3 -4
FUNCTION 1

Group centroids in discriminant space

127
clarity, note condition the Opponent was seen as more ”optimistic,
rational, and kind.“

Results of the decision theoretic model. Equations 4 and 5 of
the first chapter specify the process by which a person's choices are
related to the choices of the other person. If Equation 4 is post
multiplied by Equation 5 then potentially observable chains of events
can be deduced.. Two intermediate cognitive events as specified by
s and qn are postulated to intervene between any trial n and trial n+l.
The parameter s specifies the indeterminacy of the four possible stimulus
conditions with respect to any two of three psychologically possible
cognitive states. The psychological state then determines the responses
of the subject. For this realization of the model the subject was
expected to choose Y with probability I if he perceived the opponent
to be in a CM or a CM state. If the opponent was perceived to be in a
C8 state then the subject was expected to choose X with some probability
qn. The model further states that the CM perceptual state is attained
from any one of the four stimulus conditions whereas both the CM and
CM conditions can be attained from two mutually exclusive pairs of
stimulus conditions.

While the stimuli for the subject, that is the responses of the
other player, were programmed to respond according to Equations ll through
14 of Chapter two they were in themselves contingent upon the play of
the subject. in order to generate the predicted performance matrix
it is necessary to introduce the subject's own level of cooperative
responding through a parameter Ln-l' It will be recalled that when w
was set equal to l-y in Equation ID the preprogrammed player was said

to be playing according to a contingent strategy. The subject's

l28

stimulus conditions playing against this strategy are then given by:

R=[Ln_lw, (l-Ln-1) (l-w), Ln-](l-w), (l-Ln-])w]
The model specifies that the probability of obtaining a C response by
any subject on any trial n can be found by multiplying the activation
matrix. (Equation 4) and the decision matrix (Equation 5). The
resulting 4x2 matrix then gives the probability of either a cooperative
or defecting response occurring under the four activating stimulus
conditions. Furthermore the probability of each of those stimulus
conditions occurring is given by the vector R. When this is used to
multiply the 4x2 matrix the result is a lx2 matrix which gives the
probability of a c00perative response on trial n as well as the probability
of a defecting response.

In the contingent case the probability of a cooperative response
is found to be:

Ln=Pr(Cs’n)=qn(l-w-s+2$w) (lSa)
It will be found convenient to let:

D=(l-w-s+st) (15b)
Then Equation I5a becomes:

Ln=Pr(CS,n)=an (15c)

For the contingent model it remains to specify the process by
which qn changes. That process is schematized in Figure 4. Here
An+l is the probability of the programmed other emitting a cooperative
response on trial n+l and is given by:

An+l=Pr(CE,n+]/Cs’n)+Pr(CE’n+‘/Ds,n)

=wLn+(l-w)(l-Ln)

=I-Ln-w+2wLn (l6)

l29

 

 

 

 

 

An+l
c1:,n+1 ‘——'>Q1
CF
n
B
n+l
DE,n+l ; Q2
A
n+1 TE,n+l "“‘> Q2
CM
n
8WM 0
E,n+l ——-->Ql
A
n+l
_ cE,n-i-l ——'>Q1
E71"
n
Bn+l

J)E,n+l ——>Q2

 

Figure 4. Schema for changes in qn

130
Then the probability of the programmed other emitting a defecting
response is given by:
=I-A (17)

In schematic form, Figure 4 Specifies that two Operators Q] or
Q2, change the value of qn' Which operator is applied depends upon
the subjects' perception of the programmed other's psychological state
(CM or CM or CM) and whether the other's response (CE,n+l or DE,n+l) on
trial n+l confirms the subject's perceptions or not.

This growth model postulates that changing the probability of
the subjects emitting a cooperative response depends in part on what
state the programmed partner is perceived to be in and what the programmed
other does following that perception. For example if the other person
is perceived to be in the cooperative but nonmartyrlike state (CM) and
this is followed by a c00perative response by the other player then the
operator Q] is applied according to the following rule:

qn+1=Q](qn)=(l-t)qn+t (18.)
If the other person is perceived by the subject to be in the CM state and
the other player follows this with a defecting response than the Operator
Q2 is applied in the following way:

qn+l=Q2(qn)=(l-t)qn . (18b)
From Figure 4 the eXpected growth of qn is a joint function of being
in one of the three perceptual states and the probability of applying
one of the two linear operators. With this model the change in qn is
given by:

E(q )=Bn+]Ql[Ln-]w(l-s)+(l-Ln_])(I-w)s]

n+l

(19a)
+An+lQl[sLn_‘w+(l-s)(l-Ln_l)(l-w)+(l-s)(l-w)Ln-]]

' +sil-w)[Ln-]+s(l-Ln_l)w+(l-s)(l-Ln_])w]

+An+lQ2[(l-s)Ln_]w+s(l-Ln_,)(l-w)]

l3l
+Bn+lelsLn_lw+(l-s)(l-Ln-i)(l-w)+(l-s)(w)Ln-]
+s(l-w)Ln-]+s(l-Ln-l)w+(I-s)(l-Ln-])w] (l9b)

-Bn+'Q'(s-sLn_‘-sw+wLn-l) 4

+An+lQl(l-s+sLn-l+sw-wLn_I)

+An+lQ2(s'5Ln-l'5"+"Ln-l)

+Bng'Q2(l-s+sLn_'+sw-wLn_I)
Multiplying and simplifying this expression yields thefollowing:

E (qn +1 )eE (qn+t[ (s-sLn- I -sw+wtn_ 1 ) (2qn0+2w-qu0.- l) _

+ (i -q,,D-W+2wq,,0) -<1,,l) (20)

This expression is a nonhomogeneous, nonlinear difference equation and is thus
not amenable to analytic solution. Instead estimation of the parameters
t, 930 and s was accomplished by means of iterating the parameters through
their range of possible values and establishing those that yielded a
minimum chi-square value. 4

This procedure was carried out by means of a program for the CDC6500
computer and was done separately for each of the four contingent groups.
The parameter estimations were made in two stages. First using Equation
lSc, values of q and s were selected which minimized the chi-square values
fOr Observed and predicted cOOperatIon in trials 28, 29, and 30. The
values that produced minimum chi-square estimates are given in Table 5.
Using these estimates of s and q, Equation 20 was used to estimate values
of t which would minimize the chi-square values for predicted and
observed frequency of cooperative responses on trials 3i through 35.
(The values of t which minimize this value are given In Table 5. Using
these three estimates and Equations l5c and 20, estimates of the
probability Of cooperation were generated for the entire game and the

graphs of the observed and predicted probability of cooperation are

l32
presented in Figures 5, 6, 7 and 8 for the contingent conditions.
By inspection the model predicts that c00peration will decrease very
slowly throughout the game. The graphs also show that the level of
expected cooperation stays in the same general region of observed
cooperation for at least three of the four contingent groups throughout
the course of the game.

The model for the noncontingent conditions was derived in the,
same manner as the contingent model. The stimulus conditions for non-
contingent subjects becomes:

R'=[Ln-lw,(l-Ln_l)w, Ln_](l-w),(l-Ln_l)(l-w)]

The probability of a subject emitting a C response is given by post-
multiplying the R' vector by the product of being in the CM state and
the probability of emitting the C response once in that state given by
Equations II and l2. This leads to the following equation:

Ln=Pr(CS,n)=[Ln_lwsqn+(w-wLn_l)(qn-sqn)+(Ln-]-wLn_])

(qn-sqn)+(l-Ln-l)(sqn-wsqnll
-q,i(1-01+L,-1(20-111 <21)

Exactly the same two linear operators Q1 and Q2 are applied to
specify changes in 9n° Figure 3 also specifies the way in which qn
changes with the exception that An+1=w and Bn+l="w° Then:

E(qn+])=an+Wt(S+W+Ln-]'2WS‘2WLn-]'ZSLn-]+4swLn_])

-wtqn (22)

Parameters s and q3o were estimated by the computer using Equation
2i so that they minimized the chi-square value between the predicted
level of cooperation given by Equation 2] and the observed level of
cooperation for trials 28, 29, and 30. These values are recorded in

Table 5. Estimates were then made for t so that Equation 2i was

PROPORTION OF COOPERATIVE RESPONSES

l33

-observed

.expected

 

 

6 18 3° 42 5" 66 78 90 '02 110 '26 l38 '50

TRIAL BLOCKS OF SIZE 6

FIG. 5. Contingent .7 no note- observed and expected COOperation.

PROPORTION OF COOPERATIVE RESPONSES

l34

.2 -observed

.expected

 

 

6 13 3° 42 5" 66 78 90 '°2 11:. '2°

TRIAL BLOCKS OF SIZE 6

138 '5°

FIG. 6. Contingent .7 note- observed and expected cooperation.

PROPORTION OF COOPERATIVE RESPONSES

'35

-observed

.eXpected

 

 

6 54 l02 l26’ l50

18 3° 42 66 78 90 111i 138

TRIAL BLOCKS OF SIZE 6

FIG. 7. Contingent .9 no note- observed and expected cOOperation.

PROPORTION OF COOPERATIVE RESPONSES

I36

-observed

.expected

 

 

6. l8 30 “2 54 66 78 l02 1‘“ l26' ISO

90
TRIAL BLOCKS OF SIZE 6

I38

FIG. 8. Contingent .9 note- observed and expected cooperation.

137
minimized for trials 3l through 35. Values of the t estimates for the
four noncontingent conditions are also given in Table 5.

Using these estimates of t, q and s, predicted values were

30
generated by Equation 2i for the remainder of the game. Both the observed
values and their estimates are graphed in Figures 8, 9, l0, and II. From
inspection of these figures it can be seen that the model predicts too
many uncooperative responses both in terms of the rate of change of these
responses and the level which they finally reach. In all conditions

the model predicted that the subjects' responses would drop almost

immediately to a near zero level.

Table 5

Parameter Estimates for the Model

 

 

 

Conditions 5 Q30 t
Contingent-.7-no note .625 .925 .0l
Contingent-.7-note .800 .975 .Ol
Contingent-.9-no note .750 .900 .0l
Contingent-.9-note .975 .975 .0l
Ikrncontingent-.7-no note .025 .975 .50
Noncontingent-.7-note .975 .975 .0l
Noncontingent-.S-no note .325 .725 .0l
Noncontingent-.9-note .975 .975 .0l

 

 

 

 

From Table 5 it will be observed that for seven of the eight
conditions the learning parameter t was equal to .0l while in one
condition it was abnormally high. The parameter q30 was in general

quite high as one might suppose it to be. In seven out of eight

PROPORTION OF COOPERATIVE RESPONSES

l38

-observed

.expected

 

 

6 0 4 8 l02 l26 l 0
i8 3 42 5 66 7 90 ll4 I38 5

TRIAL BLOCKS OF SIZE 6

FIG. 9. Noncontingent .7 no note- observed and expected cooperation.

PROPORTION OF COOPERATIVE RESPONSES

139

-observed

.expected

4‘ A A -_.A A_‘- A _A I A A A I A ‘4‘L

I8 3° 42 5" 66 78 90 '°2 lI4 ”6 l38 '5°

TRIAL BLOCKS 0F SIZE 6

 

6

FIG. l0. .Noncontingent .7 note- observed and expected cooperation

PROPORTION OF COOPERATIVE RESPONSES

I40

-observed

.expected

 

 

6 0 4 8 l02 i26 l 0
l8 3 42 5 66 7 90 ll4 I38 5

TRIAL BLOCKS 0F SIZE 6

FIG. ll. Noncontingent .9 no note- observed and expected co0peration.

l4l

 

 

.9
.8
.7
.6 - - _
u, _ -
I.”
U, _ -
g; .5
Q.
U, _
L”
m -
LLJ
22 _
:5 .4 ‘
o:
u, _ -
(L
O
O
(.3
LL
C) .3
2
Si
I'-
(K
O
O.
2
a. .2
-observed
.l I .expected
.0 ................
6 18 3° 112 5" 66 73 90 '°2 11L. '26 I38 '50

TRIAL BLOCKS OF SIZE 6

FIG. l2. Noncontingent .9 note- observed and expected cooperation.

l42
conditions it was .90 or better. In the noncontingent, .9, no note
condition it reached only .725. Estimated values of s were quite
consistent with the theoretical predictions. In the four contingent
conditions 5 took on the two lowest values for the no note subconditions
and was highest (.975) when both the behavior of the other was clear
(.9) and the other had sent a note clarifying his strategy.

In the four noncontingent conditions 5 was again lowest for both
of the no note conditions but was well below .5 for both of them.
According to Equation 9 the latter is not to be expected. In the non-
contingent conditions a cooperative response by the other most often
has followed a defecting response by the subject. From the second row
of Equation 9 it can be seen that a low value of 5 would then send the
subject into the CM state which is the only state from which a COOperative
response can be made. Thus the low values of s in these two conditions
probably reflect the mechanical attempt of the model to raise the level
of predicted cooperation for the subjects in these two groups.

Assessment of the experimental deception. Question l6 of the
final questionnaire was analyzed and subjects were classified according
to their responses into whether or not they thought the experiment had
involved deception. Three categories of ”yes, uncertain and no” were
used to classify the subjects according to their response. The frequency

f the responses to each category for each experimental condition are
presented in Table 6. This frequency table was subjected to a method
developed by Sutcliffe (I957) for analysis of frequency data for multiple
classification designs. None of the chi-square values for the higher

order interaction was significant beyond the .25 level. However when

all the strategy conditions were classified according to whether they

l43
Table 6
Frequency of Deception Responses

to the Experimental Conditions

 

Noncontingent Contingent

 

No note Note No Note Note

 

 

Yes 2 4 4 6 3 5 l 2
Uncertain 4 2 3 3 l l 2 O
NO 9 9 8 6 l2 9 l2 l4

 

 

 

 

 

 

 

 

 

were contingent or noncontingent the chi-square value was 7.7 (p<.025,
df=2). This suggests that the frequency of ”yes, uncertain and no“
responses was not independent of whether the subject faced a contingent-
noncontingent strategy. Inspection of Table 6 snows that subjects had
more yes and uncertain responses to the noncontingent strategy (a total
of 27) than they did to the contingent strategy (a total of IS).

To ascertain the effects of the response to deception on the
subject's X choices the uncertain responses of the subjects were
reclassified into the yes and no categories on the basis of their
answers to the other deception questions. The subjects' explanations
of their uncertain response was used to reclassify ”uncertain” answers.
Thus an answer that indicated that all psychology experiments were
deceptive was reclassified into the no category whereas an answer that
mentioned the other subject changed his mind in a peculiar way was

placed in the yes category. For this analysis only those subjects who

I44
were used in the main analysis were retained. This meant dropping the
six subjects who were used to fill out the groups in the final stages
of the experiment, for a total of l20. This is presented in Table 7.
The chi-square value for this table was computed and foUnd to be equal to
5.2 (p<.025, df=l). As indicated by the test of significance the
frequency of yes and no responses was not independent of the strategy
that the subject faced. The effect of this lack of independence was
investigated by means of an analysis of variance of the data for Experiment
I. For computational reasons the trials effect was ignored and the

category of yes-no was added. Thus subjects were classified according to

Table 7
Frequency of Reclassified Reactions

to the Deception Questions

 

 

 

Contingent Noncontingent
Yes l5 27
No 45 33

 

 

 

 

the type of strategy they faced (contingent or noncontingent), the level
of clarity (high or low) with which the strategy was implemented,

whether or not a note was exchanged, and finally, the subjects postgame
response to the deception questions. The dependent variable was the
subjects' number of cooperative responses. Since the subjects' responses
to the deception questions were markedly unequally distributed among the
experimental conditions an analysis of variance technique employing the

least square algorithm to correct for unequal n was used. The results

l45

of this analysis are given in Table 8. Reaction to deception was found
to affect cooperation depending upon the type of strategy the other
adopted and the clarity with which he presented that strategy. For
subjects who faced an other who was clearly contingent subjects who

felt they were deceived chose cooperatively an average 94 times out of
I20 while those who felt they weren't deceived chose cooperatively 78
times. Those who faced a less clearly contingent other chose cooperatively
67 out of a possible 120 times when they responded ”yes” to the deception
questions while if subjects responded ”no'I they chose cooperative 58
times on the average. In the clearly noncontingent condition subjects
who responded ”yes” to the deception questions chose cooperatively 39
times. Subjects who responded ”no” chose cooperatively 57 times out of

a possible l20 choices. In the less clearly noncontingent condition
subjects who felt deceived chose cooperatively 72 times while the non-
deceived group chose cooperatively 5i times. When apprOpriate F tests
were run on the meaningful, simple interaction effects only the last
difference was found to be significant (F=4.27, p<.05, df=l/l04). That
is, in the less clearly noncontingent condition the l2 subjects who
responded ”yes” to the deception question cooperated more than the l8
subjects who responded ”no.“

It can be concluded then that the experimental deception affected
vubjects differently depending upon the type of strategy they faced.
Subjects who faced a noncontingent strategy were more suspicious than
that group who faced a contingent other. With respect to the effect
on the dependent variable of perceived deception it was found that
subjects in the low clarity, noncontingent condition who admitted

deception cooperated more than those who did not feel deceived.

I46

Table 8

Effect on C00peration Choices of the Various Experimental

Conditions and Subjects' Reaction to Deception

 

 

 

Source df Mean Square F
Contingency (A) 1 9095.9 12.2*
Note (9) 1 7281.9 9.7**
Clarity (c) 1 565.2
Deception (D) l ll46.5

A x B l 28l.4
A x c 1 7728.9 10.35**
A x D 1 743.0
8 x C l 890.]
B x D I 45.6
C x D l l583.9
A x B X C l ll09.8
A x c x 0 1 3168.1 A.2A*f
A x B x D l l2l.4
B x C x D l ll8.2
A x B x C x D l 320.2
rror Between lO4 746-9

 

 

 

 

“p<.001
xxp< . 002

‘{?{3:p< . OS

I47

Discussion

 

The effect of the note, the level of cooperation observed in the
noncontingent conditions of the experiment, and the level of cooperation
observed in this and other experiments where a contingent strategy was
used provides evidence that the imitation hypothesis of Rapoport and
‘ Chammah is inadequate to explain cooperative behavior in a wide variety
of PDG situations.

The effect of the noncontingent strategy points out that subjects
did not adopt any simple imitation rule in selecting their choices.
Rather than imitating an other who adopted a noncontingently cooperative
strategy subjects exploited the simulated other. There was no evidence
that the subjects even adopted a strategy of matching their exploitive
choices to the frequency of the other's cooperative choices since in
the clearly noncontingent no note condition the subjects' cooperative
choices were greater than the l0% a matching explanation would demand.
Furthermore in the less clearly noncontingent condition where no notes
were exchanged, subjects chose cooperatively at better than a 30% rate.

Within the four noncontingent conditions two of the conditions
called for the subject to exchange a note with the simulated other.
Compared to the no note exchange condition the level of cooperation
was increased for those groups who exchanged notes. This effect
cannot be explained by the imitation hypothesis since, at best, one
expects subjects to match their cooperative choices or their exploitive
choices to the noncontingent strategy. However the note had the effect
of having subjects choose c00peratively at much less than the frequency
with which the simulated other chose X. Subjects' exploitive choices

also decreased in frequency - a direction that was opposite to what

I48
would be expected if subjects were matching their exploitive choices
to the other's cooperative choices.

Rapoport and Chammah (I965a) have described just how the
contingent strategy condition fits their imitation explanation. For
a bona fide game they consider an automata with a perfect imitative
response tendency and show that, under the assumption that subjects want
to maximize their own payoff, they will end up in the mutual cooperation
cell. In their experiments, males in a real game situation (where
the other is a male) respond cooperatively to the extent that 26% of the
variance is accounted for by an index that the authors say measures tit-
for-tat tendency. In other experimental conditions the variance declines
until the imitative index explains just below l6% of the variance for
males facing women or women facing either sex. As a matter of fact,
Rapoport and Chammah limit their concluding conjecture to a statement
that the men's higher cooperative response frequencies were due to
their greater tendency to give tit-for-tat in social situations similar
to the PDG.

If this be so then some of the data presented in this experiment
are inexplicable. The imitation index of Rapoport and Chammah is in
reality the phi coefficient. It measures the extent to which the responses
of one player match the immediately preceding response of the other
, ayer. In contingent PDG experiments, including the present one, this
index is well controlled. The low clarity contingent condition
represented a subject or an automata whose propensity to imitate the
other (or phi coefficient) was set at .4. This is less than the average

propensity to imitate (.5) that Rapoport and Chammah found among 70 male

I49

pairs. Under the low clarity contingent conditions there was a decline
in cooperation from 7l% for trials 31 to 40 and 56% for trials 4i to
50, to 47%, 45% and 44% on the final three blocks of trials. This is
consistent with Rapoport and Chammah's theory. Inspection of their
graphs show that when the imitation index on the average is .5l cooperative
performance rose from 50% to between 60% and 70% for equivalent trial
blocks. However when males faced females the men encountered an imitation
index of .35 and their cooperation was found to rise from 4l% to just
over 50% by the final block of trials. That is, when faced with a .35
imitation index in the Rapoport and Chammah experiment, male subjects
increased their cooperation. When faced with a .40 imitation index in
the present eXperiment subjects decreased their cooperation from over
50% to about 46%. When the other had an imitation index of .8 in the
contingent high clarity condition, cooperation rose to about 74% by the
final block of trials.

In summary the relationship between the imitation index and
frequency of cooperation is complex. When the imitation index is zero
as it is for both noncontingent conditions cooperation is found to
decrease but in the low clarity noncontingent condition it remains at
about a 45% level of cooperation. This rate of decline is not that
deviant from one where the imitation index was set at .40 in Experiment
‘ The latter, in turn, is deviant from the pattern of cooperation that
Rapoport and Chammah found for males who faced an imitation index of
.35. Subjects in the RapOport and Chammah experiment increased their
cooperation to above the 50% level. When facing an imitation index of
.5] in a bona fide game subjects cooperated about 68% of the time by

the last block of trials. An increase in the imitation index to .8l in

150

the present experiment had increased the level of cooperation by only
6% to 74% by the last block of trials.

While the data do not support the imitation hypothesis the data
are consistent with the idea that subjects form impressions about the
other's strategy early in the game. First, the responses of the subjects
were in a direction consistent with what would be expected of subjects
who were prepared to maximize their gains in the face of a contingent
or a noncontingent strategy. That is, the noncontingent strategies
were expected to reflect a martyrlike strategy and thus elicit less
c00peration than the contingent strategies which were expected to reflect
a cooperative but nonmartyrlike approach. As each of these two basic
strategies became clearer it was postulated that subjects show less
confusion in assigning the noncontingent strategy to the martyr status
and the contingent strategy to the basically cooperative status. Thus
the strategies of the other as they were constituted for the present
experiment can be ordered along a dimension of strategies that
increasingly demand the cooperative response in order to ensure that
the person is successful at winning money in the PDG. The rank order
among the groups of the average number of cooperative responses was in
accordance with this ordering. However with respect to the two non-
contingent conditions this effect was only present in a statistically
.3iiable way where no notes were exchanged between the players.

The subject's ratings of the person they played with supports
the ordering of the other's strategies. The subjects discriminated
along two dimensions among the various strategies they faced. In
addition, the content of the adjectives suggests that the hypothetical

strategies underlying the four stimulus conditions had merit in terms

lSl
of content validity. Thus where the simulated other employed an
unconditional strategy he was seen as ”believing, yielding, formed,
weak, meek, calm and bad.” These words are consistent with the
martyrlike strategy which three of the four groups above the median on
this function were predicted to perceive. The fourth noncontingent
group that was below the median on this function was the low clarity
noncontingent group that also exchanged a note. The effect of the
simulated other's note seems to have been to raise the probability
of the subject entering the CM state. That is, he enters a state
inconsistent with seeing the other as “meek, believing, yielding etc.“
This same relationship that exists between the noncontingent low
clarity note and no note groups holds between the two high clarity
noncontingent groups. Thus, the high clarity group who received a
note saw the other as less ”believing, yielding, etc.” than the comparable
group that received no note.

Why the low clarity, contingent, note group was rated highly on
the meekness dimension is not entirely clear. However it may be noted
that for the low clarity contingent group the information provided by
the choices of the other was probably minimal since the other matched
the subjects' choices only 70% of the time if they were cooperative and,
30% of the time if they defected. Subjects would receive the least
.nformation about the strategy of the other if they were choosing
randomly. 0n the six trials just preceding the note exchange the low
clarity contingent note group was choosing X about 45% of the time.

On the six trials following the note exchange but just prior to the
rating, subjects were choosing cooperatively 67% of the time. This

increase had the effect of increasing the subject's earnings and

l52

decreasing his losses. At the same time the other was perceived in

a more favorable light since his cooperation was perceived to be
increasing. Finally, the note, a fairly clear statement emphasizing
cooperation appeared in a situation where the subject faced maximal
uncertainty until the note came. The ratings were then done within
seven trials of the note. The s-rong effect of the note and the initial
promise it offered in increased earnings is verified by the score of 4.3
on the ”cooperative” adjective that the contingent, low clarity, note
group gave the other as opposed to the score of 2.7 that the contingent,
low clarity, no note group received for the same adjective. While in
the high clarity condition the note increased information and earnings,
the net effect was less than in the low clarity conditions.

The second function separated the eight groups of subjects into
six groups who saw the other as “faithful, offensive, feminine, good,
etc.” versus two groups who saw the other at the opposite extreme for
these words. The common feature of the latter two groups was that given
the strategy of the simulated other there was no way that these subjects
could make very much money. In terms of the model the second discriminant
function separated those subjects who faced a CM strategy from those
subjects who faced either a CR or a CM strategy. While the low clarity
contingent conditions had been expected to reflect a C8 strategy, in

etrospect they were more representative of a CM strategy. To see this
one can consider a subject who chose cooperatively l00% of the time when
faced with a contingent strategy. This is the one strategy that
maximizes gains for a subject. Even so the subject would make only l¢
in an average ten trials because on three trials he would lose a total

of 6c while on the remaining seven trials he would make 7c. (This

153

would account for the unexpected decline in cooperation over trials
of the low clarity contingent groups.) The two words ”ungrateful and
untrustworthy” load heavily on this discriminant function. In terms
of content it is difficult to reconcile these two adjectives with
”reputable and altruistic” which are also present in the second function.
However both of these troublesome words may reflect perception of the
two contingent groups who employed a CM because one would be unlikely
to trust very much a person who would drive such a hard bargain as
represented by a predominantly tit-for-tat strategy. Nor would one
expect to find such a person grateful for any favors done for him.

To conclude, relatively early in the game a subject comes to form
impressions of the other just from the strategy with which he makes
in his PDG choices. The relative values of the eight groups on the
two functions are consistent with the hypothetical strategies underlying
the choices of the other. The majority of the words that have the
highest function weights on each function explain the discrimination
made among the groups. The interpretation of each of the two functions
is imperfect because of two words ”ungrateful and untrustworthy“
both of which have high weights on the first two functions. Other
than this the most heavily weighted words on the first function described
an essentially cooperative passive other. The second function included
9rds that were more indicative of a cooperative, as opposed to a
destructive, other.

The model as it is constituted in Chapter One had mixed results.
It fitted not too badly in the four contingent conditions both with
respect to the difference between the predicted and the observed mean

level of cooperation and with respect to the relationship among the

l54

value of s. It is notable that in both of these respects the model
falls short in the noncontingent condition. First Of all it predicted
a lower level of cooperation than the subjects actually attained

in the noncontingent condition. Second the values of 5 do not have
sufficient ”power” in this condition to withstand the tendency of

the model to predict noncooperation. The values 5 can take on range
from 0 to I. It was observed in the present eXperiment that the note
exchange increased cooperation even for the noncontingent conditions.
To account for this moderate increase the values of 5 underwent an
increase that spanned the range of possible values. Thus in the low
clarity noncontingent condition where c00peration increased on the
average from 44% for the no note group to 54% for the note group,

the s values increased from .025 to .975. in the high clarity noncontingent
condition an average increase from 27% in the no note condition to 53%
in the note condition brought about an increase in s values from .325
to .975. The assumption is that the model, in trying to predict a
higher level of cooperation forces the subject into the CM state which
is the one state from which a cooperative response can be made.

These two facts suggest that subjects in the CM state either
because of guilt, passivity, imitation, response style, or some unknown
reason choose C with some probability not equal to zero or one. This

aceives empirical support from Rapoport's (I968) report of an unpublished
finding by Chammah to the effect that about half the subjects who faced
a l00% unconditionally cooperative strategy adopted a cooperative
strategy themselves while the remainder exploited such a strategy.

This points out a major gap in Mead's theory. There is no
specification of the responses that the subject will select in any

particular situation. The theory states only that the subject is

ISS

capable of forming a relationship between the early part of the other
person's behavior and what will occur later by means of the inferences
that the subject makes about the earlier behavior. The nature of
those inferences and the type or frequency of the responses can be
ascertained only from an investigation of the particular situation in
which they occur. This is not necessarily a bad feature of a theory
especially one which claims to have generality apart from any given
situation. It can be expected that research about the situation itself
will supply information sufficient to fill in the gaps of the theory.

To return to the problem at hand there is reason to believe that
the response matrix needs to be modified. One practical modification
would be to have the subjects respond cooperatively with some probability
qn when they are in the CM state. When in the CM state subjects would
not respond cooperatively with some probability qn but with probability
one. This would be consistent with the present finding that q3o was .9
or better for seven of the eight groups. The net effect of this change
would be to make it easier for a cooperative choice to follow from the
CM state without substantially changing the rate of cooperation that
the CM state would produce. This would also reflect Chammah's finding
that a certain proportion of subjects responded cooperatively when
faced with a pacifist strategy. The CM state would remain the same with
He probability of a cooperative response being zero as it was with the
old model.

This brings up one other difficulty with the model or rather the
theory behind it. It is still unspecified how independent variables
like type or clarity of strategy, sex of the player, and message

exchanges operate on the two parameters 5 and q. In the analogous

156
perceptual decision model, the parameter equivalent to s is assumed
to be determined by the physical aspects such as the signal to noise
ratio. The parameter equivalent to qn on the other hand is felt to
be determined by experimental conditions such as the presentation
schedule, information feedback and the outcome structure. However the
exact relationships between these two parameters and the independent
variables remains to be proven. While the present model is similar
in form to the signal detection model it is similar in content to a
model deveIOped by Rapoport (l967) and Rapoport and Cole (l968).
It will be recalled that in their model subjects estimate their opponent's
propensities on the basis of their own and the opponent's joint choice
on any given trial. The prOpensity is assumed to be determined by the
policy of the other player. Unlike the present model Rapoport speculates
that subjects forosee that their choice on any particular trial affects
the policy of the other player. As a consequence of this assumption
Rapoport's model produces a cumbersome mathematical formulation.

While not using this feature of the Rapoport model the present
model incorporates one of Rapoport and Cole's post hoc findings that
subjects form their estimate of the policy of the other on the basis
of the choices that are made by the other one trial after their own.
Their original model called for the estimate to be made from the joint
occurrence of their own and the other's choice on the same trial.

The strong effect of a note whose referent was the strategy of the
programmed other was established in this experiment. Gahagan and
Tedeschi (I968) rightly point out that note effects had been previously
established only for very short games. They ran subjects in a rather

lengthy game where the referent to a note was the play on the next

157
trial of a programmed other. Specifically the other promised to
cooperate on the next trial and followed through on that promise 30%,
50% or 90% of the time. They found that the strategy of the other
elicited differential overall cooperation only when the credibility
of the other's promises were at their two lowest values. That is,
the note was involved in an interaction with the credibility of the
other's strategy. In the present experiment where the note's referent
was the strategy of the programmed other no such interaction was observed.
That is, the note raised cooperation independently of the clarity or
credibility of the other's strategy and without regard to the type of
strategy the other employed. This effect was postulated for the con-
tingent condition but was unexpected for the noncontingent condition.
This could be considered as evidence against the relationship between
the note and its referent. However for the noncontingent conditions
it was possible for subjects to enter the CM state. In addition it
is also possible under the proposed new model for the note to operate
on qn indirectly by the fact that subjects in the CM state can choose
cooperatively.

As has been explained previously an imitation explanation could
not explain the note effect. The model as formulated in Chapter Two
could have had the effect of increasing the probability of entering the
E state for all conditions. In the contingent conditions this explana-
tion was borne out by an increase in s which, for the most frequent
stimulus occurrences in the contingent conditions, (C /C and

E,n S,n-l

DE nIDS n-l)’ increases the probability of entering the CM state.

There was an increase in s in the noncontingent conditions which is

explicable for one of the relevant stimulus conditions for the

l58

noncontingent case (CE

probability of entering the CM state. However for the other stimulus

n/CS n-l) since an increase in 5 increases the
’ ’

condition (CE’n/Ds,n'l) which occurs frequently under the noncontingent
strategy an increase in 5 increases the probability of the subject
entering the CM state. As the original model stands the subject is

not able to respond cooperatively from this state.

Some modification of the model might bring the predicted level of
cooperation closer to the noncontingent empirical data in which case the
parameter 5 could reflect the effect of the note. The possibility exists
that the note as it was constructed conveyed information that prevented
subjects from entering the CM state and thus exploiting the pacifist
other. After all, the other person's note could be divided into two
parts. The first part indicated that the other would play X while
the second part gave information about the conditionality for doing so.
In the noncontingent note condition the note expressed an unconditional
plan to choose X. If subjects could react to the note according to
a primacy effect than the second piece of the note's information could
be expected to have little or no effect on subjects' choices. The
results from Experiment 3 which are discussed in the next chapter argue
against such an interpretation since subjects who played against a
machine that adopted a noncontingent strategy chose cooperatively less
often when a note was exchanged as opposed to when no note was exchanged,
while for those who played with a male the reverse was true.

One other finding is of interest with respect to the high and
low clarity noncontingent conditions. When no note was exchanged the
amount of cooperation was less for those subjects who faced an other

playing a clearly noncontingent strategy than it was for those who faced

159
a similar but less clear strategy. This parallels the finding of Gahagan
and Tedeschi (I968) that subjects react differently to different
noncontingent strategies but only under conditions of relatively low
credibility. Thus both Experiment I and the Gahagan and Tedeschi
finding call into question Rapoport's (I968) contention that a differential
noncontingency does not elicit a differential response from subjects.

Any conclusion drawn about this or any other Prisoner's Dilemma
study must take into account the effects of the subjects' response to the
deception. Subjects in the present study were more suspicious of a
noncontingent than a contingent strategy. A post hoc analysis of these
deception responses found that the effect of the response to the
deception was related in a reliable way to the choice responses of
the subject. The relationship was complex however in that both the
frequency and the type of contingency interacted to affect the player's
responses. When examined, this interaction was found to be determined
by one meaningful comparison. Facing an unclear, noncontingent strategy,
subjects who felt deceived cooperated more as compared to subjects who
did not. It is difficult to understand why subjects who were responding
cooperatively to a cooperative other felt that they were deceived by
the experimenter. It is possible that such subjects adopted a coopera-
tive response set themselves and when the other doublecrossed them
“Jr a large percentage of the time they realized that their expectations
of earning $2.50 were being thwarted. It could be that these subjects
then felt deceived in terms of their failure to carry away from the
experiment the money that was promised to them.

This line of reasoning leads to an important point about any
inferences which are to be made about the effects of the response style

upon subject' reaction to deception. As in any post hoc analysis there

I60

is the question of the directionality of the cause effect relationship.
That is awareness of the deception could have brought about a shift in
the_manner in which the subjects chose. But it is equally plausible
that the subjects' level of cooperation caused them to be more aware of
the other's programmed strategy. This would be especially the case in
the contingent case where the further the subjects' responses deviated
from a random pattern the clearer was the strategy of the other. This
explanation would fit both the contingent conditions and the clearly
noncontingent condition. In the contingent case subjects who felt
deceived chose more cooperatively than those who did not feel deceived
although this difference did not reach an acceptable level of statistical
significance. This argument is strengthened by the finding in Experiment
3 (to be discussed in the next chapter) that subjects who knew they
were playing against a machine strategy cooperated less than those
subjects who thought they were facing a male other.

Any analysis of the response to deception must go lacking in
the present experiment for want of an appropriate control group. That
is, one must consider what proportion of subjects would have responded
affirmatively to the deception question even if they were playing with
a real other person. This could arise from the demand characteristics
of the questionnaire, the general suspicion surrounding the psychological
-.periment, disappointment over not making $2.50, as well as other
similar reasons. ~The intent of the present experiment was to assess
the differential effect of the two strategy styles on the responses to

deception and the effect this had on the subjects' choices.

CHAPTER FOUR
Results and Discussion: Experiments 2 and 3

At the end of Chapter One it was pointed out that only in a
very few studies was the nature of the opponent in PDG investigated.
Consequently in Experiment 2 the effect on both males and females
of having a contingent or noncontingent opponent who is either a male
or female is looked at. In addition, Experiment 3 investigates the
effect, on males only, of playing against a machine that adopts a
contingent or noncontingent strategy.

Experiment 2: Results.

 

The effect of the experimental variables on the c00perative
choices of the subjects in Experiment 2 is presented in Table 9. In
this experiment the simulated other chose according to either a clearly
contingent or clearly noncontingent strategy (Factor A). Also the sex
of the subject (Factor B) was varied. In addition the subject was
faced with an opponent of either the same or opposite sex (Factor C)
as determined by the instructions. From inspection of Table 9 the type
of strategy that the other player employed was found to be only
marginally significant. However when the sex of the player is considered
the contingent-noncontingent strategy has differential effects as
.1dicated by the significant A x 8 interaction. Males playing against
a contingent strategy chose X an average of 65% of the time while an
independent group of males who faced a noncontingent other chose X 37%
of the time. Females who played a contingent other chose X 48% of the

time whereas females who played a noncontingent other chose X, on the

I6l

I62

average, 6l% of the time. When the appropriate tests of simple effects.
are made it is found that males who faced a contingent other were more
cooperative than males who faced a nOncontingent other (F=2l.5, p<.0l,
df=l/ll2). Females were more COOperative with the noncontingent other
(F=4.7, p<.05, df=l/ll2) than were females who faced a contingent other.
Males were more cooperative than females facing a contingent other
(F=8.0, p<.0l, df=l/ll2) while the reverse was true when the two sexes
faced a noncontingent other (F=l5.6, p<.0l, df=l/ll2).

Disregarding the type of strategy of the other player, females
were more cooperative when playing against a supposed male than were
males (F=4.2, p<.05, df=l/ll2). There was only marginal evidence that
males playing against a female other were more c00perative than when
they played a male other (F=3.l, p<.lO, df=l/ll2).

There was a difference in cooperation across the trial blocks.
Inspection of the average amount of cooperation in blocks of trials
suggested that cooperation oscillated about the 50% level with four of
the blocks having cooperation just below this level and eight of the
blocks being above. With the exception of the first block, subjects
chose the cooperative response less than 55% of the time. In the first
block subjects chose X 63% of the time.

The level of cooperation across trials was contingent upon the
Sex of the player and the type of strategy he or she was faced with
(F=5.35, p<.0005, df=ll/l232). Males were most responsive to the
opponent's strategy. They predominantly chose X when faced with a
contingent other and they increased their cooperation as the game
progressed. In their final block of trials they were cooperating 7l%

of the time. in contrast males faced with a noncontingent strategy

I63

Table.9

Analysis of Variance of X Responses for Experiment 2

 

 

 

Source df MS F
Between Subjects

Contingency (A) l 207.03 3.04***
Sex of Player (8) l 4l.34

Sex of Opponent (C) l l0.00

A x B 1 1579.21 23.16*
A x C l 96.lO

s x c 1 304.34 A.A6**
A x 8 x C l 25.07

Error Between ll2 68.l9

Within Subjects

Trials (T) 11 20.11 5.1A*
A x T ii 5.79

B x T ii 2.27

C x T II 3.32

A x B x T 11 20.96 5.35*
A x C x T II 4.59

B x C x T ii 2.45

A x B x C x T ll 2.4l

Error Within l232 3.92

 

 

 

 

"p<.0005
:t..p<.05

as I\

"p<.l0

I64

ended up cooperating 30% of the time. Females facing a contingent
other chose cooperatively about 45% of the time on the last block of
ten trials. In the same block of trials, females playing against
the noncontingent other chose X 63% of the time.

Experiment 3: Results

 

The F values for the factors and their possible interactions are
presented in Table ID. The dependent variable is the amount of
cooperation. Factor A represents the high clarity contingent or non-
contingent strategy of the other. Factor B stands for the nature of
the other in terms of whether subjects faced a man or a machine. Factor
C indicates whether or not a note was exchanged between the dyad.

The first two main effects are significant and indicate that the subjects
are more cooperative when they are playing against a contingent strategy
as opposed to a noncontingent strategy. For the former on the average
they chose X 62% of the time while for the latter they chose X 30% of

the time. They are also more cooperative when playing another man than
when playing a machine (55% versus 38%). There was only marginal
statistical evidence that an exchange of notes increased cooperation.

On an average block of ten trials subjects increased their cooperation
from 42% to SI% when they exchanged a note. However it would appear

that the nature of the opponent is important in determining the effect

of the note. Thus for both the note and no note conditions subjects
playing against a machine chose cooperatively only 38% of the time.

When a note was exchanged with a human opponent the X button was

chosen 64% of the time as opposed to only 45% of the time when no note
exchange took place. This latter difference is statistically significant

at better than the I% level.

I65

Table l0

Analysis of Variance of X responses for Experiment 3

 

 

 

Source df MS F
Between Subjects

Contingency (A) l 3657.66 48.20*
Man-Machine (B) 1 962.03 12.68**
Note-No Note (c) 1 290.70 3.83****
A x B l 23.26

A x C l 22.25

B x c 1 318.28 A.i9***
A x a x c 1 323.95 h.27***
Error Between Il2 75.88

Within Subjects

Trials (T) ll l3.59 3.69*

A x T ll l3.88 3.77*

B x T II 2.29

C x T II 2.37

A x B x T ll 4.l2

B x C x T ll l.67

A x C x T ll 6.6l l.80***
A x B x C x T II 2.04

Error Within l232 3.68

 

 

 

 

it
p<.OOS

**p<.001

**p£.05
it *7:
* p-.os3

3':

I66

_The significant three way Interaction of type of contingency
by type of OppOnent by note conditions indicates that the type of
strategy of the opponent and whether or not the OppOnent exchanged
notes with them, were both important in determining how subjects
responded to the nature of their opponent. Faced with a contingent
strategy, subjects who were led to believe they were facing another
male became more cooperative when notes were exchanged. Subjects who
faced a machine playing contingently likewise became more cooperative
when notes were exchanged. The former group of subjects Increased
their cooperation from 64% to 75%. The latter group increased their
response level from 50% to 6i%. Subjects who were led to believe they .
were facing a male other and the other played a noncontingent strategy
chose X 53% of the time when a note was exchanged but chose X only 27%
of the time when no note was exchanged. Likewise subjects faced with
an unconditionally cooperative machine chose X 27% of the time when no
note was exchanged but chose X only l5% of the time when a note was
exchanged. 'Thus it appears from inspection that an exchange of notes
for subjects facing a noncontingent other Increased cOOperatIon if
subjects felt that they were playing with a male but decreased it or
had no effect if subjects felt they were playing against a machine.
When no notes were exchanged In the noncontingent condition subjects
on the average chose X with exactly the same frequency whether or not
they played against a man or a machine.

The significant Interaction of type of contingency and note
conditions with blocks of trials suggests that the subjects' performance
curves took a different form depending upon their experimental conditions.

The contingent note group chose X with relatively the same frequency

167

across blocks of trials. The other contingent group gradually increased
their cooperation as the game progressed but their average level of
cooperation was always below that of the note group. Similarily the
note noncontingent group reached their asymptotic level of cooperation
by about trial 50 and stayed at that level for the remainder of the
game. The group who faced a noncontingent other but did not exchange
notes continued to choose X less and less as the game progressed.
Furthermore, on the average the latter group cooperated less than the
noncontingent group that exchanged notes.

Discussion. The results of Experiment 2 suggest that there is
no unequivocal way of asserting that females are more or less cooperative
than males. Thus the finding of Rapoport and Chammah (l965a) that
females are less cooperative than males is true in this study but only
when the other is contingently cooperative. When the other is non-
contingently cooperative females are observed to be more cooperative
than males. Vinacke's (I969) assertion that the laboratory renders
males less exploitive and females more exploitive has to be qualified
since the demand characteristics of the PDG coupled with unconditional
cooperation would invite exploitiveness. If as Vinacke suggests,
females are cooperating with the demand characteristicsset up by the
experimenter by becoming exploitive then there is no reason to see why
.cmales would not exploit the simulated other when It appeared
unconditionally cooperative. As Tedeschi, Lesnick and Gahagan (l969)
postulated Experiment 2 showed that under intermediate levels of 1
conflict intensity, as measured by CI, sex differences among subjects
become paramount. These differences probabily become'even more apparent

when the strategy of the other Is more “rational” or predictable as the

l68

strategy of the other was in the present experiment. The strategy
of the other in the Tedeschi et al study was to randomly choose the X
button 50% of the time.

Contrary to the findings of Rapoport and Chammah, the sex of
the other had no appreciable effect either alone or in combination with
other effects on the way that subjects selected from their two choices.
The above two authors found that in a real two person game males competed
more against a female than when they faced a male other. The reverse
relationship held for females. In the present study it was found that
males cooperated with males significantly less often than females
cooperated with males. There was no statistically significant difference
between groups faced with the same or opposite sex when the sex of
the player is held constant. Such a finding must be interpreted with
caution since a major difference between the Rapoport and Chammah
experiment and the present one is that the other was simulated. Further-
more it was simulated at a high level of clarity. This may have mitigated
the effects that the other's sex could have under conditions of lower
clarity. This would be implied by Jones and Davis' contention that the
individual characteristics of the other person cannot convey more informa-
tion about the Other than the intentions of the other. Thus if the
intentions of the other were less clear from his choices then one would
c.0ect that information about the other would have proportionately greater
effect. Thus in a much shorter game where the clarity of the other
person's strategy was less than in the present study, Marlowe, Gergen
and Doob (l966) found that manipulated attributes of the other had an

effect on subjects' choices in the PDG.

I69

In Experiment 3 it appeared at first glance that whether the

opponent was a computer or a human affected the way subjects responded.

In the contingent, no note condition subjects chose X 64% of the time

when they faced a male and 50% of the time when they faced a machine.

This difference was not statistically significant. When the note exchange
condition was added to the experimental treatments of these latter two
groups it had the effect of increasing their levels of cooperation by

ll%, to 75% for the male other group and to 6l% for the machine other
group. The group who faced a noncontingent male other chose X 27% of the
time as did the group who faced a noncontingent machine other. When

notes were exchanged the noncontingent male other group increased their
COOperative choices to 53%. When notes were exchanged with a noncontingent
machine cooperative choices decreased from the 23% level to l5%. Thus
under noncontingent conditions a note had the opposite effect of decreasing
or holding constant c00peration when a machine was the Opponent and of
increasing COOperation when a male was the opponent.

In terms of the second exploratory hypothesis, playing against a
male other led to a counterstrategy which was nonsignificantly different
from the counterstrategy adopted against a machine when the other
adopted a contingent strategy. This is contrary to the hypothesis as
formulated in the introduction. As postulated in the explanation given
..ove the strategy was sufficiently clear about the intentions of the
other that any information gained from a knowledge of the characteristics
of the other would be superfluous. In the noncontingent no note
condition there was also no difference in the way subjects played
against a male other and a machine other. Again it can be expected

that the behavior of the subject was dictated by the high level of

I70

clarity with which the strategy was played. In the noncontingent,
note condition cooperation was higher for those who faced a male other
than for those who faced a machine other. Thus, for the noncontingent,
note, male other group the hypothesis was verified. That is knowledge
that the other was a human subject under the note exchange condition
led to a less effective counterstrategy.

Across all conditions subjects played less cooperatively against
a machine than they did against a male other (F-l2.68, p<.00l). Thus
three of the four groups who faced a machine other cooperated on the
average at least I4% less than comparable groups who faced a male other.
Disregarding specific experimental treatment combinations, it appears
that the type of opponent is important in ascertaining how subjects
will choose. Subjects cooperate more with humans than they do with a

machine.

CHAPTER FIVE
Summary and Conclusions

This dissertation set out to investigate the cognitive components
of social interaction. A class of theories appropriate to such an
investigation has recently been advanced under the general rubric of
attribution theories. Three of these formulations were outlined and
discussed in Chapter One.

The first of these formulations, Heider's Theory of Interpersonal
Relations (Heider, l958) is really an attempt to bring some order to
a variety of interpersonal examples drawn from everyday situations.
He observed that people over a wide range of varying social situations
are interested in attributing to others such qualities as ”can,” ”trying,”
and ”want.” He drew attention to the analogy between the social attribu-
tion problem and the phenomenon known as perceptual constancy in the
psychology of perception. That is in perception a large number of
proximal stimuli are capable of mediating the perception of a distal
stimulus. Similarly a wide variety of social situations are capable
of rendering the same attribution. While the theory is comprehensive
enough to be classed as a methatheory it seems to lack sufficient
specificity to test its major tenents.

A variation of Heider's theory has been presented by Jones and
Davis (I965). Their main assumption is that the social perceiver seeks
to find sufficient reason why an action on the part of another person
took on a particular form. In particular the person tries to establish
links between stable individual dispositions and observed action.

Jones and Davis introduce the term ”correspondence” to refer to the

l7l

I72

extent that the intended actions and the underlying characteristic

or attribute are similarly described by the inference of the perceiver.
A particular version, called “hedonic relevance,“ of social perception

occurs when the actions of the other affect the perceiver in a positive
or negative way.

Kelley's theory (Kelley, I967) was put forward as the third
attribution theory. For Kelley, attribution refers to the process
of inferring or perceiving the dispositional properties of entities
in the environment by one or more observers observing at any one time
or over periods of time. Kelley also details sources of systematic
errors that can be made in attributions.

A fourth account of cognitive behavior, Meads theory of social
interaction (Mead, I934), was introduced as an alternative to the general
class of attribution theories. While it contains some of the same
features as the attribution formulations it introduces the notion of
the act, the gesture and the significant symbol. The present dissertation
was directed towards a test of Mead's theory using a standard laboratory
situation the well known Prisoner's Dilemma Game.

The PDG is well suited to test out theoretically derived hypotheses
since the degree of communication between two interactants can be well
controlled. Furthermore compared to other social situations a large
.umber of facts have accumulated which enable one to control for certain
conditions. For example, as was shown in Chapter One, paper and pencil
measures of personality differences tend not to predict behavior in
PDG experiments. Yet real life variables such as the sex of the
participants influences such behavior considerably. Finally it was

shown in Chapter One that no adequate explanation of peoples' behavior

’ I73
in PDG situations has yet been advanced. A number of studies especially
ones which utilize a simulated other whose choices are determined by‘
a preset strategy have hypothesized that impressions of the other are
important in determining choices. However until now few attempts were _
made to measure these impressions and no attempts were made to systematically.
relate the Impressions to response choices. 7‘

The present thesis postulated that broad classes of simulated
strategies can be placed into two specific classes. A noncontingent.
strategy was postulated to be associated with impressions of martyrdom
on the part of the simulated other. A contingent strategy was supposed'
to signify an other who was cooperative but unwilling to be a martyr.—

If this impression of the other was correct then subjects motivated to
win money were expected to choose cooperatively to the extent that the
other's strategywas considered to be clearly contingent. The clarity
of the Other person's strategy was manipulated by changing the frequency
with which the choices of the other were consequent upon the choices of
the subject. In addition clarity was manipulated in Experiment I by
having the simulated other pass a note to the subject which expressed
the essential contingency or the essential noncontingency of the
strategy employed.

To verify the postulated relationship between the simulated
other's strategy and the impressions that the subject formed about
the other, a-30 Item bipolar semantic attribution scale was administered
to all subjects In Experiment l. The Items were selected to reflect
a basic martyrdom orientation on the part of the other as well as to

be able to assess a cOOperatIve but unmartyrlike approach.

I74

Finally in Experiment I a specific model was advanced based on
a decision theoretic approach. This model postulated that, dependent
upon what contingent event appeared on any two adjacent trials, the
subject thought the simulated other was either a martyr, or cooperative
but not a martyr, or finally was not at all cooperative. The latter state
was assumed to be necessary since at low frequency levels of the contingent
strategy the subject is exploited on a number of occasions throughout
the game.

The responses of the subject were assumed to be meaningfully
related to the impressions formed about the other. Specifically,
motivated subjects were expected to choose noncooperatively to the
extent that the other impressed the subject as being exploitive. To
the extent that the subject inferred the other to be cooperative but
not a martyr then the model of the subject's behavior hypothesized'
that the subject chooses cooperatively with some positive probability.

To implement the experimental design l20 male subjects were
assigned to Experiment I which had a total of eight different experimental
conditions. These conditions varied the contingency or noncontingency
of the other simulated person's strategy, the clarity, high or low,
with which that strategy was carried out, and whether or not the
simulated other exchanged a note with the subject. Fifteen subjects
were randomly assigned to each combination of experimental conditions.

The PDG was then played for ISO trials with the simulated other. The
other person was in fact a preprogrammed electronic machine.

Subjects were found to choose more cooperatively if they faced
a contingent than If they faced a noncontingent strategy. Moreover

the clearer the strategy of the other the stronger was the reaction

ail-i —

I75
of the subjects to that strategy with the exception of the noncontingent
note condition where the response to the note was nondiscernable from
that where no note was exchanged. The rate of change in cooperation
over the last l20 trials was dependent upon both the strategy of the
other and the clarity with which it was played. Finally the exchange
of notes had the effect of increasing cooperation across all other
conditions. While this effect was expected for the contingent conditions
it was unexpected in the noncontingent ones.

The data for both the noncontingent and contingent strategy was
more consistent with a cognitive approach to PDG interaction than it
was with an imitative explanation. In particular the other person's
strategies which a priori could be classed as eliciting an impression
of the other as a martyr or not, elicited cooperation in accordance
with these hypothesized impressions.

That these two dimensional impressions had some basis in fact is
made clear by a discriminant function analysis of the subjects' impressions
of the other early in the game. The first two functions were found to
discriminate among the groups. The first function separated the contingent
from the noncontingent conditions (with the exception of contingent,
low clarity, note condition). The second function separated the two
low clarity contingent conditions from all others. This latter finding

hile not expected, made sense in that, in retrospect, these two
conditions prevented subjects from winning money. They would thus be
more consistent with a CM strategy.

The model did not explain the behavior of the subjects with the
exception of the two low clarity, contingent conditions. Decline of

cooperation for all the noncontingent conditions as predicted by the

176
model was too abrupt and drastic to fit the data. In the high clarity
contingent conditions the model predicted a level of cooperation in
the same general region as the data provided. However there was a
predicted gradual decline in the subjects' cooperative choices which
was not at all evident in the data. In the light of some recent
research results of Chammah's a proposed modification of the model was
made.

Subjects were asked to indicate in a postexperiment questionnaire
the extent to which they felt they were deceived. A post hoc analysis
of the subjects' responses to this deception question indicated that
they were more likely to be suspicious if they faced a noncontingent
strategy. Further analysis of the cooperative choice data taking into
account whether the subject felt deceived or not, revealed that
affirmation of deception affected COOperation. However this relationship
held only for one meaningful comparison. That is reaction to deception
only made a difference in the low clarity noncontingent condition.
Subjects who felt they were deceived responded more cooperatively than
those subjects who did not. An explanation for this finding centered
around the fact that subjects expected they would receive $2.50. If
subjects adopted a cooperative strategy in the low clarity noncontingent
condition their expectation would be far from being met.

All of the findings from this first study were limited to a
population of male subjects. The literature revealed that the sex of
the opponent and the sex of the player were important determinants of
how the PDG is played. These two findings had not been previously
established for simulated other PDG's which ran for a large number of

trials. Accordingly, the second experiment using the same procedure

I77
as the first, varied the sex of the player and the sex of the simulated
opponent. While the sex of the player was found to be an important
factor in how the game was played the sex of the opponent played little
or no role in the outcome of the game. Specifically males were found
to react to the strategy of the other more than females. Furthermore
the direction of the reaction was dependent upon the type of strategy
the other adopted and the sex of the player. Males coOperated more when
facing a contingent strategy than when facing a noncontingent one.
Females cooperated more while facing a noncontingent other.
The third experiment found that the nature of the other was important
in determining male subjects' reactions to it. Males across a variety
of experimental conditions, cooperated more with a human than with a
computer. However this difference while present in every condition
reached statistical significance only when the other adopted a noncontingent
strategy and told the subject he was doing so by way Of a note.
In retrospect several conclusions can be drawn from the three
eXperiments and the theory that underlies them. No evidence was found
that an imitation explanation was sufficient to explain PDG choice.
There is evidence that subjects form multidimensional impressions
relatively early in a repeated PDG based on the strategy that the other
chooses. The nature of these impressions is not excessively strong

5 one would predict from the decision theoretic model which in turn
was based on Mead's theory of the significant symbol.

The specific predictions generated by Mead's theory resulted in
a model of the PDG which fit the observed data well in two out of eight
conditions. In two out of the remaining six it fit only moderately well.

But in all four noncontingent conditions the model did not represent the

I78

data at all. Suggestions were taken from the recent experimental
literature to improve the model. The choices of the subjects in each

of the experimental conditions were consistent with an a priori

ordering of the strategies in terms of the impressions these strategies
must have exerted. That is subjects who faced a clearly noncontingent
other played more exploitively than a group of subjects who faced a less
clearly noncontingent other. This latter group in turn cooperated about
as much as a group that faced a contingent other who played a relatively
unclear strategy. The group that faced a clearly contingent other
cooperated more than the group that faced the less clearly contingent
other.

The experimental procedure used to vary the clarity of the other
person's strategy was to increase the frequency with which the simulated
other matched the choices of the subject in the contingent condition or
the frequency with which he chose cooperatively in the noncontingent
condition. The exchange of notes had been eXpected to clarify the
basic strategy of the other. This effect is present for the contingent
condition but is not for the noncontingent condition where cooperation
is observed to increase. It is concluded, however, that subjects are
capable of reacting to the underlying PDG strategy of the other when
it is verbally communicated even though that reaction may be inapprOpriate.

In two ancillary experiments it was found that the sex of the
opposing player in a simulated game is unimportant in determining
c00peration although the nature of the other plays a role in promoting
- c00peration. That is, neither females nor males change their level of
cooperation when they face a male or female. Females are more cooperative

with a noncontingent other as opposed to a contingent other whereas

I79

males are more COOperative with a contingent other. This calls into
question Vinacke's assertion that females are cooperating with the
demand characteristics of the experiment whereas males are resisting
these same demands.

Finally the first experiment shows that male subjects are more
likely to be suspicious of a noncontingent strategy than they are of a
contingent one. The effect of this suspicion on their PDG choices is
not altogether clear since closer inspection of the data reveals that
a reaction is present for only the group who faced a less clearly
noncontingent strategy. Furthermore the subjects that admit awareness

of deception cooperate more than the subjects who make no such assertion.

BIBLIOGRAPHY

BIBLIOGRAPHY

Axelrod, R. An axiomatic approach to the Prisoner's Dilemma. Journal
of Conflict Resolution, I967, 11, 87-99.

 

Bandura, A., 8 Walters, R. Social learning theory and personality
development. New York: Holt, Rinehart 8 Winston, I963.

 

 

Baxter, J., Winters, E., 8 Hammer, R. Gestural behavior during a brief
interview as a function of cognitive variables. Journal of Personality

and Social Psychology, I968, 8, 303-307.

 

 

Bixenstlne, V., 8 O'Reilly, E. Money versus electric shock as payoff
in a Prisoner's Dilemma Game. The Psychological Record, l966, l6,
251-264. —

 

Bixenstlne, V., Potash, H., 8 Wilson, W. Effects of level of cooperative
choice in a Prisoner's Dilemma Game: Part I. Journal of Abnormal
and Social Psychology, I963, 66, 308-3l3.

 

 

Bixenstlne, V., 8 Wilson, W. Effects of level of cooperative choices
in a Prisoner's Dilemma Game: Part II, Journal of Abnormal and
SoCial Psychology, I963, 61, l39-l47.

 

 

Bobbitt, R. Internal external control and bargaining in a Prisoner's
Dilemma Game. Dissertation Abstracts. I967, 21, 9B, 3266-3267.

 

Cooley, W., 8 Lohnes, P. Multivariate procedures for the behavioral
sciences. New York: Wiley, I962.

Crowne, D. Family orientation, level of aspiration and interpersonal
bargaining. Journal of Personality and Social Psychology, l966, 3,
64l-645.

Deutsch, M. The effect of motivational orientation upon trust and
suspicion. Human Relations, I960, 13, l23-l39. (a)

 

Deutsch, M. Trust, trustworthiness, and the F scale. Journal of Abnormal
and Social Psychology, I960, 61, l38-I40. (b)

 

Dolbear, F., 8 Lave, L. Risk orientation as a predictor in the Prisoner's
Dilemma. Journal of Conflict Resolution, I966, 19, 506-5l5.

 

Druckman, D. Dogmatism, prenegotiation experience, and simulated group
representation as determinants of dyadic behavior in a bargaining
situation. Journal of Personality and Social Psychology, I967, 6,

279-290.

I80

I8l

Duncan, S. Nonverbal communication. Psychological Bulletin, l969,

_7__2_, 118-137.

 

Evans, G. Effect of unilateral promise and value of rewards upon
cooperation and trust. Journal of Abnormal and Social Psychology,

1961+. 99. 587-590.

 

Evans, G. 8 Crumbaugh, C. Effects of Prisoner's Dilemma format on
cooperative behavior. Journal of Personality and Social Psychology,

1966, _3_, 486-488. (a)

 

Evans, G. 8 Crumbaugh, C. Payment schedule, sequence of choice and
COOperation in the Prisoner's Dilemma Game. Psychonomic Science,

I966, 2, 87-88. (b)

 

Faucheux, C. 8 Moscovici, S. Self esteem and exploitive behavior an
a game against chance and nature. Journal of Personality and Social
Psychology, I968, 8, 83-88.

 

 

Gahagan, J. 8 Tedeschi, J. Strategy and the credibility of promises
in the Prisoner's Dilemma Game. Journal of Conflict Resolution, I968,
12, 224-234.

 

Gallo, P. 8 McClintock, C. Cooperative and competitive behavior in mixed
motive games. Journal of Conflict Resolution, I965, 9, 68-78.

 

Gouldner, A. The norm of reciprocity: a preliminary statement.
American Sociological Review, I960, 25, l6l-l78.

 

Grant, M. 8 Sermat, V. Status and sex of others as determinants of
behavior in a mixed-motive game. Journal of Personality and Social
Psychology, I969, 12, l5l-l57. .

 

Greenhouse, 5., 8 Geisser, S. On methods in the analysis of profile
data. Psychometrika, 1959, £11. 95-112.

 

Gumpert, P., Deutsch, M. 8 Epstein, Y. Effect of incentive magnitude
on cooperation in the Prisoner's Dilemma Game. Journal of Personality,
and Social Psychology, I969, 11, 66-69

 

 

Guyer, M. Response-dependent parameter changes in the Prisoner's Dilemma
Game. Behavioral Science, I968, 12, 205-219.

 

Haas, H. 8 Maehr, M. Two experiments on the concept of self and the
reaction of other. Journal of Personality and Social Psychology,
I965, l, IOO-IOS.

 

Halpin, s., a Pilisuk, M. Probability matching in the Prisoner's
Dilemma. ngchonomic Science, I967, Z, 269-270.

 

Ifiarford, T., 8 Solomon, L. I'Reformed sinner“ and ”lapsed saint” strategies
in the PrisOner's Dilemma Game. Journal of Conflict Resolution,

1967,1_1_, 1011-109.

 

l82

Harford, T. 8 Solomon, L. Effects of a “reformed sinner” and ”lapsed
saint” strategy upon trust formation in paranoid and nonparanoid
schizophrenic patients. Journal of Abnormal Psychology, l969, 4,
498-504. . '-

 

Harris, R. Note on ”Optimal policies for the Prisoner's Dilemma.“
Psychological Review, l969, lg, 363-375.

Harrison, A., 8 McClintock, C. Previous experience within the dyad_
and cooperative game behavior. Journal of Personality and Social
Psychology, I965, l, 67l-675.

 

Heider, F. The psychology of interpersonal relations. New York:
Wiley, I958:

Heller, J. The effects of racial prejudice, feedback, strategy, and
race on cooperative-competitive behavior. Dissertation Abstracts:

I967, 21, 78, 25078-25088.

 

Horai, J. 8 Tedeschi, J. Effects of credibility and magnitude of
punishment on compliance to threats. Journal of Personality and
Social Psychology, I969, 12, I64-I69.

 

 

Israel, J. Experimental change of attitude using the Asch-effect.
Acta Sociology, I963, Z, 95-lO4.

 

Johnson, T., Feigenbaum, R., and Weiby, M. Some determinants and
(consequences of the teacher's perception of causation. Journal of
Educational Psychology, I964, 55, 237-246.

 

 

Jones, E. 8 Davis, K. From acts to dispositions. In L. Berkowitz (Ed.),
Advances in Experimental Social Psychology. Vol. 2. New York:
Academic Press, I965, 2I9-266.

Jones, E., 8 Harris, V. The attribution of attitudes. In H. Kelley,
Nebraska Symposium on Motivation. Lincoln, Nebraska: University
of Nebraska Press, I967. '

Jones, E., Steele, M., Gahagan, J., 8 Tedeschi, J. Matrix values and
c00perative behavior in the Prisoner's Dilemma Game. Aggrnal of
Personality and Social Psycholpgy, I968, 8, l48-I53.

.:.Aan, J. Effects of level of aspiration in an experimental bargaining
situation. Journal of Personality and Social Psychology, I968, 8,
l54-l59.

 

Kanouse, 0., 8 Wiest, W. Some factors affecting choice in the Prisoner's
Dilemma. Journal of Conflict Resolution, l967, ll, 206-2l3.

 

Kelley, H. Attribution theory in social psychology. In D. Levine
(Ed.), Nebraska Symposium on Motivation. Lincoln Nebraska: University
of Nebraska Press, I967.

 

I83

Kelley, H., Thibault, J., Radloff, R., 8 Mundy, D. The deveIOpment
of cooperation in the minimal social situation. Psychological
Monographs, I962, l-l9, No. 538.

 

 

Knapp, W., 8 Podell, J. Mental patients, prisoners, and students with
simulated partners in a mixed motive game. Journal of Conflict
Resolution, I968, 12, 235-24l.

 

 

Komorita, S. Cooperative choice in a Prisoner's Dilemma Game. Journal
of Personality and Social Psychology, I965, 2, 74l-745.

 

Komorita, S. 8 Mechling, J. Betrayal and reconciliation in a two-person
game. Journal of Personality and Social Psychology, I967, 8, 349-353.

 

Komorita, S., Sheposh, J., 8 Braver, S. Power, the use of power, and
cooperative choice in a two-person game. Journal of Personality_and
Social Psychology, I968, 8, l34-I42.

 

 

Krausse, R., Vivekananthan, P., 8 Weinheimer, S. ”Inner speech” and
”external speech”: Characteristics and communication effectiveness
of socially and nonsocially encoded messages. Journal of Personality
and Social Psychology, I968, 2, 295-300.

 

Loomis, J. Communication, the development of trusts, and cooperative
behavior. Human Relations, I959, 13, 305-3l2.

 

Lumsden, M. Perception and information In strategic thinking. Journal'
of Peace Research, l966, 8, 257-277.

 

Lutzker, D. Internationalism as a predictor of cooperative behavior.
Journal of Conflict Resolution, l960, fl, 426-43l.

 

Marlowe, 0., Gergen, K., 8 Doob, A. Opponents personality, expectation
of social interaction, and interpersonal bargaining. Journal of
Personality and Social Psychology, l966, 8, 206-2l3.

 

 

Marwell, G., Ratcliff, K., 8 Schmitt, D. Minimizing differences in a
maximizing difference game. Journal of Personality and Social
Psychology, I969, 12, l58-l63.

 

 

Mead, G. Mind, self and society. Chicago: University of Chicago Press,
I934.

 

Messé, L., Dawson, J., 8 Lane, I. Equity as a mediator of the effect of
reward level on behavior in the Prisoner's Dilemma Game. Unpublished
manuscript. Michigan State University, I970.

Messick, D. Interdependent decision strategies in zero sum games:
A computer controlled study. Behavioral Science, I967, 12, 33-48.

 

McClintock, C., 8 McNeel, S. Reward and score feedback as determinants
of cOOperatIve and competitive game behavior. Journal of Personalipy
and Social Psychology, I966, fl, 606-6l3.

 

I84

Mckeown, C., Gahagan, J., 8 Tedeschi, J. The effect of prior power
strategy on behavior after a shift of power. Journal of Experimental.
Research in Personality, l967, 2, 266-233. -

Miller, R. No play: A means of conflict resolution.. igurnal of
Personality and Social Psychology, l967, 8, l50-i56. ‘

Minas, J., Scodel, A., Marlowe, D., 8 Rawson, H. Some descriptive
aspects of two person non-zero-sum games ll. Journal of Conflict
Resolution, I960, 8, l93-l97.

Murdock, P. Exploitation, accommodation and social responsibility in~
a bargaining game. Journal of Personality, I968, 88, 440-453.

Natsoulas, T. Locus and orientation of the perceiver (ego) under
variable, constant, and no perspective instructions. Journal of
Personalipy and Social fsychology, I966, g, I90-l96.

 

Orne, M. On the social psychology of the psychological experiment:
With particular reference to demand characteristics and their implica-
tions. American Psycholggist, I962, 11, 776-783.

Osgood, C., Suci, G., 8 Tannenbaum, P. The measurement of meaning.
Urbana: University of Illinois Press, I957.

Oskamp, 5., 8 Perlman, 0. ‘Factors affecting cooperation in a Prisoner's
Dilemma Game. Journal of Conflict Resolution, I965, 2, 359-374.

Oskamp, S., 8 Perlman, 0. Effects of friendship and disliking on
cooperation in a mixed motive game. Journal of Conflict Resolution,»,
I966, 18, 22l-226.

O'Toole, R., 8 Dubin, R. Baby feeding and body sway: An experiment in
George Herbert Mead's “Taking the role of the other.” gournal of
Personality and Social Psychology, I968, 18, 59-65. -

 

Pillsuk, M., Potter, P., Rapoport, A., 8 Winter, J.‘ War hawks and peace
doves: Alternate resolutions of experimental conflicts. Journal of
Conflict Resolution, I965, 2, 49l-508.

Pruitt, D. Reward structure and cooperation: The decomposed Prisoner's
Dilemma Game. Journal of Personality and Social Psychology, l967,
Z, 2I-27.

Pylyshyn, Z., Agnew, N., 8 Illingworth, J. Comparison of Individuals
and pairs as participants In a mixed motive game. Journal of Conflict
Resolution, I966, 18, 2ll-220. -

 

Quigg, H., 8 Pitt, L. Inquest into a photograph. United Press International,
The Stateggournal, Lansing, Michigan, June 29, I969. '

Radlow, R. An experimental study of ”cooperation” In the Prisoner's
Dilemma Game. Journal of Conflict Resolution, l965, 2, 22l-227.

I85

Radlow, R., 8 Weidner, M. Unenforced commitments in ”cooperative”
and ”noncooperative” non-constant-sum games. Journal of Conflict
Resolution, I966, 18, 497-505.

 

Rao, C. Advanced statistical methods in biometric research. New York:
Wiley, I952.

Rapoport, A. A note on the index of cooperation for Prisoner's Dilemma.
Journal of Conflict Resolution, I967, 11, lOO-l03. (a)

Rapoport, A. Optimal policies for Prisoner's Dilemma. Psychologjcal

Review, 1967, 28, I36-I48. (b)

 

Rapoport, A. Editorial comments. Journal of Conflict Resolution, I968,
l2, 222.

 

RapOport, A., 8 Chammah, A. Prisoner's Dilemma. Ann Arbor: University
of Michigan Press, I965. (a)

 

Rapoport, A., 8 Chammah, A. Sex differences in factors contributing to
the level of c00peration in the Prisoner's Dilemma Game. Journal
of Personality and Social Psychology, 1965, _2_, 831-838. (b)

Rapoport, A., 8 Cole, N. Experimental studies of interdependent mixed
motive games. Behavioral Science, I968, 12, l89-204.

 

Rosenberg, S., 8 Cohen, 8. Referential processes of speakers and
listeners. Psychological Review, I966, 22, 208-23l.

 

Rosenthal, R., Experimenter effects in behavior research. New York:
Appleton Century Crofts, I966.

Sampson, E., 8 Kardush, M. Age, sex, class and race differences in
response to a two person non-zero-sum game. Journal of Conflict
Resolution, I965, 2, 2l2-220.

 

Schacter, S., 8 Singer, J. Cognitive social and psychological
determinants of emotional states. Psychologjcal Review, I962,

$3.9.» 379-399.

 

Scheff, T. A theory of social coordination applicable to mixed-motive
games. Sociometry, I967, 88, 215-234.

 

Sermat, V. The effect of an initial cooperative or competitive
treatment upon a subject's reSponse to conditional cooperation.
Behavioral Science, l967, L2, 30l-3I3. (a)

Sermat, V. The possibility of Influencing the other's behavior and
cooperation: Chicken versus Prisoner's Dilemma. Canadian Journal
of Psychology, 1967, 21, 204-2l9. (b)

 

 

Shure, G., Meeker, R., 8 Hansford, E. The effectiveness of pacifist
strategies in bargaining games. Journal of Conflict Resolution,

1965, _9__, 106-117.

 

l86

Solomon, L. The influence of some types of power relationships and
game strategies upon the deveIOpment of interpersonal trust.
Journal of Abnormal and Social Psychology, I960, 81, 223-230.

Steele, M., 8 Tedeschi, J. Matrix indices and strategy choices in
mixed motive games. Journal of Conflict Resolution, l967, 11,

l98-205.'

 

Sutcliffe, J. A general method of analysis of frequency data for
multiple classification designs. Psychological Bulletin, I957,
iii: I3“”‘37.

 

Swensen, R. Cooperation in the Prisoner's Dilemma Game I: The effects
of asymmetric payoff information and explicit communication.
Behavioral Science, I967, 12, 3l4-322.

 

Swingle, P. Effects of the emotional relationship between protagonists
in a two person game. Journal of Personality and Social Psychology,

1966, 5, 270-279.

Swingle, P. Effects of prior exposure to cooperative or competitive
treatment upon subject's responding in the Prisoner's Dilemma.
Journal of Personality and Social Psychology, I968, 18, 44-52.

 

Swingle, P., 8 Coady, H. Effects of the partner's abrupt strategy
change upon subject's responding in the Prisoner's Dilemma. Journal
of Personality and Social Psychology, l967, 8, 357-363.

 

Swingle, P., 8 Gillis, J. Effects of the emotional relationship
between protagonists in the Prisoner's Dilemma. Journal of Personal1§y
and Social Psychology, l968, 8, l60-l65.

 

 

Tedeschi, J., Aronoff, D., Gahagan, J., 8 Hiesler, D. The partial
reinforcement effect and the Prisoner's Dilemma Game. 89urnal of
Social Psychology, I968, 28, 209-2l5.

 

Tedeschi, J., Lesnick, S., 8 Gahagan, J. Feedback and ”washout”
effects in the Prisoner's Dilemma Game. Journal of Personality
and Social Psychology, l968, 18, 3I-34.

 

 

Terhune, K. Motives, situation and interpersonal conflict within
Prisoner's Dilemma. Journal of Personality and Social Psychology,
l968, 8, (Monogr. suppl. 3)

 

Toch, H. Violent men. Chicago: Aldine Publishing Co., l969.

 

Thomas, S. Discrim 2. Technical Report No. 33. Computer Institute
for Social Science Research, Michigan State University, I968.

Valins, S. Cognitive effects of false heart-rate feedback. Journal
of Personality and Social Psychology, l966, 8, 400-408.

 

Vinacke, W. Variables in experimental games: Toward a field theory.
Psychological Bulletin, I969, 21, 293-3l8.

 

I87

Wilson, W. 8 Insko, C. Recency effects in face-to-face interaction.
Journal of Personality and Social Psychology, l968, 2, 2l-23.

 

Winer, B. Statistical principles in experimental design. New York:
McGraw-Hill, I962.

Wrightsman, L. Personality and attitudinal correlates of trusting
and trustworthy behavior in a two-person game. Journal of
Personality_and Social Psychology, l966, 2, 328-332.

 

 

APPENDICES

Appendix A

SemantichfittributighvScaje
The purpose of this scale is to measure the communication effect
in experiments. Please base your judgments on the X and Y choices of
the other person and how much he has made from the way he has played.
Immediately after these instructions there is a set of 30 scales. You
are asked to rate the other person on each of these scales in order.
Here is how you are to use these scales.

If you feel that the other person is very closely related to one

end of the scale, you should place your check mark as follows:

 

beautiful x / / / / / / ugly
or
beautiful / / / / / / x ugly

 

If you feel that the person is closely related to one or the
other end of the scale (but not extremely), you should place your

check mark as follows:

 

beautiful / x / / / / / ugly
or
beautiful / / / / / x / ugly

 

If the concept seems slightly related to one side as opposed

to the other (but is not really neutral), then you should check as

 

follows:
beautiful / / x / / / / ugly

or
beautiful / / / / x / / ugly

 

I88

l89

The direction toward which you check, of course, depends upon
which of the two ends of the scale seem most characteristic of the
other person.

If you consider the person to be neutral on the scale, both
sides of the scale equallyyassociated with the concept or if the scale
is completely irrelevant, unrelated to the other person, then you should
place your check mark in the middle space.
Important:
(I) Place your check-marks in the middle of the space.
(2) Be sure to check every scale.
(3) Just check once for every scale.
(4) Do not try to remember how you checked earlier items.

Make each scale a separate and independent judgment.

(5) Do not puzzle over individual judgments. Give your first impressiOns
based on the person's choices. Remeber, think about his choices as
he played them in this game then check the words in a way that best

describes the kind of person he is.

 

 

 

 

 

 

 

 

l. good / / / / / / bad

2. trusting / / / / / / skeptical
3. independent / / / / / / . yielding

4. optimistic / / / / / / pessimistic
5. sociable / / / / / / unsociable
6. unfair / / / / / / fair

7. meek / / / / / / aggressive
8. disreputable / / / / / / ’ reputable
9. passive / / / / / / active

 

20.
2i.
22.
23.
24.
25.
26.
27.
28.
29.
30.

I90

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

. grateful / / / / /
. pleasureable / / / / /
. severe / / / / /
irrational / / / / /
. believing / / / / /
. offensive / / / / /
. trustworthy / / / / /
. stingy / / / / /
. strong / / / / /
. masculine / / / / /
altruistic / / / / /
calm / / / / /
formed / / / / /
unsuccessful / / / / /
wise / / / / /
kind / / / / /
changeable / / / / /
meaningless / / / / /
cautious / / / / /
competitive / / / / /
faithful / / / / /

 

ungrateful
painful
lenient
rational
skeptical
defensive
untrustworthy
generous
weak
feminine
self-serving
excitable
formless
successful
foolish
cruel

stable
meaningful
rash
c00perative

unreliable

Appendix 8

Final Questionnaire
In order to pay you by check and to explain the experiment to
you we need to know how to reach you by mail. Please fill out
that information below.

Name

 

Address

 

 

Telephone No.

 

Social Security

 

In order to analyze the data from these experiments some more

information is needed.

I. I am ' ' male

f—"| female

2. I am | I freshman

I I SOphomore
I I junior
| I senior

3. I have taken the following number of credits in psychology
(approximately).
0 - l2
l2 - I8

I8+

l9l

I92
4. I have participated in an experiment similar to this one on
previous occasions
Yes

No

5. If the answer to 4 was yes briefly describe how it was similar.

C. Now we would like to ask you some questions about how the game wa5'
played.‘ 4
I. What percentage of the time do you think most accurately
describes how often the other person pushed the X button.
‘252
45%
55%
70%
90%

2. Briefly describe the way you played the game.

3. Estimate how much money you think that he made.
less than 75¢

75¢ to $l.25

$l.25 to $l.50

$l.50 to $l.75

$l.75 to $2.00

$2.00 or more

. I93
4. Estimate how much money you made.
less than 75¢
75¢ to $l.25
$l.25 to $l.50
$l.50 to $l.75
$l.75 to $2.00
____ $2.00 or more
5. Some people adopt a particular pattern of ”idea“ in choosing

X or Y. Some people don't. Describe as best you can how you

chose X or Y throughout the game.

6. Do you think he adopted some pattern or strategy throughout the

game. Describe what it was if there was one.

7. Do you think the other player saw you as having a strategy for
playing X or Y.
Yes

No

8. If the answer to 7 was ”yes“ please describe how he might
have seen you, pointing out any differences between your own

strategy and how he might have perceived it.

ll.

I2.

I4.

I94
Below are ten squares. Fill in the squares with an X as you
think the other player would have played X during an average

IO trials.

 

 

 

 

 

 

 

 

l 2 3 4 S 6 7 8 9 I0
Below are ten squares. Do the same for yourself. That is
put in an X where you feel you would have chosen X for an

average l0 trials.

ILII I i

l 2 3 4 5 6 7 8 9 l0

 

 

 

 

 

 

 

What did you think the experiment was trying to accomplish?

Do you think the other player had any special advantage over
you?
Yes

No

Did you feel that your partner played the game in any unusual way?

Yes

No

If the answer to l3 was yes, briefly describe what was unusual

about it.

l5.

I95
Subjects feel that because experiments are on human beings
that they are being secretly watched or subtly manipulated or
deceived in any number of ways. 00 you feel now that this
experiment was designed to deceive you in any way?
Yes

No

If your answer to IE was ”yes” please explain.

Appendix C

Letter
To the people who took part in the experimental game:
The following is a brief explanation of the three experiments
that were conducted in the past month. You participated in one of
the three. These belong to a class of experiments which go under the

name of the Prisoner's Dilemma Game. Each of the three experiments

 

had several different conditions within it so it will be impossible to
go into the details behind the experiment. However I hope from the
general description given here that you will be able to tell by yourself
what type of experimental condition you were playing under.

The name of Prisoner's Dilemma is given because of the real life
example used to explain the experimental situation. (Similar examples
can be constructed around conflicts between married couples, de-escalation
in the Viet Nam war and other conflicted social situations.) ‘The
example of the Prisoner's Dilemma is as follows. Two criminals have
committed a robbery. However the police have no evidence against
either one. The police know they must obtain a confession from at
least one of the suspected men in order to secure a conviction. They
bring the two in for questioning and place them in separate rooms.

They offer each one a deal. If he confesses he gets no prison
sentence and also a small reward. The one who doesn't confess gets
a maximum jail sentence. If both confess both get a light jail
sentence. Both men know that if neither confess they both get off

free.

I96

I97

In the experimental situation you will remember that both people
could make either an X or Y choice. The Y choice corresponds to
”confession” on the part of the men. If you chose Y you could make
as much as 2¢ or lose l¢ depending upon what the other person did.

If you chose X (not to confess) then you could make as much as l¢ or
lose 2¢. One major difference between the experiment and the example
is that the prisoner's had an opportunity for only one choice whereas
the experimental game allowed a large number of repeated choices.

When there are such a large number of choices a form of communica-
tion is possible. That is, subjects can form ”expectations” of how
the other person will behave, what kind of person he is etc., just
from the way he makes his choices. These ”impressions” or ”expectations”
of the other person will then determine the course of the game. In
order to control for the way the other person played a machine was pre-
programmed to adOpt two basic strategies with several variations on the
two strategies. In the one strategy the machine was set to play X a
large part of the time regardless of what button the subject pressed.
Under this condition the subject was expected to exploit the machine
by preSsing Y a majority of the time. In the second major strategy the
machine was programmed to make the choice that the subject made on the
last trial. Thus if the subject played X for l00 trials the machine
nould choose X an average of 80 times. If the subject chose Y I00
times the machine would choose Y an average of 80 times. Under this
strategy it was predicted subjects would gradually learn to push X.
Since some of the strategies were more unfair than others it was
decided to give all subjects the maximum amount $2.50, that could have

been earned under the best of conditions in this game. Enclosed is a

I98

check for this amount. It was impossible to tell subjects that they
would make the full $2.50 since it was necessary to maintain the
incentive value of the game as much as possible. Similarly in order
to maintain the simulation to a realistic game as much as possible we
did not tell the subjects in the first two experiments that they were
playing against a machine.

Dr. Phillips and I would like to thank all the subjects for

their participation in the experiment. We both wish all of you success

in your studies.

Paul O'Grady (Department of Psychology)