DECISION-MAKING, GAlNoLOSS THEORY, AND THE
UTEUTY 0F SM~CDNSISTENCY

Thesis for the Degree of Ph. D.
MICHIGAN STATE UNWERSHY
WAYNE A. GUN
1972

Ihb.bl!

 

This is to certify that the
thesis entitled

DECISION—MAKING, GAIN- LOSS THEORY ,
AND THE UTILITY OF SELF-CONSISTENCY
presented by
Wayne A. Olin
has been accepted towards fulfillment

of the requirements for

Ph. D. degree in Sociology

“MW We;

hhmnpnmuux

 

Date May 5 , 1972

 

0-7639

 

 

 

 

 

ABSTRACT

DECISION-MAKING, GAIN-LOSS THEORY,.AND
THE UTILITY OF SELF-CONSISTENCY

By
Wayne A. Olin

An actor may hold beliefs about the relative abilities of himself
and the other actors involved in a decision-making task. These beliefs
or expectations may influence the decision process. At the same time,
an actor may attempt to maintain a consistent pattern in.making repeated
decisions. But if both expectations as to task relevant abilities and a
desire for self-consistency are simultaneously involved in the same
decisionqmaking process, how are these two factors interrelated? The
search for an answer to this question led to the development of the
research project which will be described in this dissertation. The
question emerged from a critical evaluation of the gain-loss theory of
decisionemaking as stated by Camilleri and Berger (1967). Their theory
is based on the utility assumptions that a loss avoided is a gain and
that a gain foregone is a loss. These ideas have been tested by them.in
an experiment, each subject is forced to choose between an action which
is self-consistent with his previous behavior and another action which
is inconsistent. The lack of agreement between predictions of the gain-
loss theory and the results obtained in one condition of their empirical
study suggests that there may be serious flaws in their utility

assumptions.

Kayne A. 01in

In the gain-loss theory the utility of self-consistency, called
ul, is assumed to be independent of the ability expectations (Camilleri
and Berger, 1967, p. 372). As the assumption of independence between
the ability expectations and the utility of self-consistency seemed to
be a particularly strong one, another experiment was designed to permit
the investigation of this assumed independence. Consequently, in this
new experiment the utility of self-consistency was incorporated as a
treatment factor with three levels. An analysis of the experimental
data indicate that, contrary to the gainqloss theory, u1 is dependent
on the ability expectations. It is concluded that the utility of self-
consistency must be reconceptualized in the gain-loss formulation of

the expectation process.

DECISION-MAKING, GAIN-LOSS THEORY, AND

THE UTILITY OF SELF-CONSISTENCY

BY

8
Wayne AJ'Olin

A THESIS

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

DOCTOR OF PHILOSOPHY
Department of Sociology

1972

Dedicated to Oscar Wilde
who said that,

"Consistency is the last refuge of the unimaginative."

ii

ACKNOWLEDGMENTS

Nothing is ever produced without the help of others, and I wish
to acknowledge the help of my committee members-éDrs. Santo F. Camilleri,
Thomas L. Conner, and James Phillips--and especially the Chairman,
Dr. Hans E. Lee, for their invaluable assistance and constructive (some-
times destructive) criticism, Thanks must go to Robert Shelly, who con-
ducted the experiments with me, and to Ann Newland, Mark Jurecki, Glen
Sanders, John Sorbet, and Cathy Phillips, the assistants in the experi-
ments, who somehow functioned admirably under the uncertainty of chaos.
Thanks also go to the subjects, those nameless faces, and to Steven
Crocker and Linda Lonning, whose names I cannot forget. I wish to
acknowledge the help of Fran Lammers, Linda Barnhouse, Gail Eitniear,
Phyllis, Wendy, Pamela, and Mary Jane. And I wish to acknowledge the
National Institute of Mental Health for financial support. The disser-
tation research itself was supported by a Training Grant in Formal
Theory in Social Psychology made to the Department of Sociology, Michigan
State University, by the National Institute of Mental Health (NIH MH-

11410-02).

iii

TABLE OF CONTENTS

List of Tables . . . . . . . . . . . . . . . . . . .
List of Figures . . . . . . . . . . . . . . . . . .
General Nature of the Problem . . . . . . . . . . . . .
General Theory and Research in Decision-Making
Gain-Loss Theory . . . . . . . . . . . . . . . . . . .
Expectation Experiment . . . . . . . . . . . .
GainéLoss Model of Expectation Experiment and Test
Replication Study . . . . . . . . . . . . . . . . . . .
An Application of the Theory to Questionnaire Data . .
Problems With the Design of the Expectation Experiment
Problems With the Gain-Loss Theory . . . . . . . . . .
Balkwell Modifications to MOdel . . . . . . . . . . .

Rqunalysis of the Model-Data Discrepancy --
Expectation Experiment . . . . . . . . . . . . .

Re-Analysis of the Model-Data Discrepancy --
Gain-Loss Model . . . . . . . . . . . . .

Theory Tested by New Experiment . . . . . . . . . .
Design of Experiment . . . . . . . . . . . . . . . . .
Results of Experiment . . . . . . . . . . . . . .

Interpretations of the Experimental Results--
Gain-LOSS MOdel o o o o o o o o o o o o o o o o o o 0

Interpretation of Experimental Results--
Other Models . . . . . . . . . . . . . . . . . . . .

Implications . . . . . . . . . . . . . . . . . . . . .

iv

vi

viii

10
16
26
26
32

35

41

42
45
48

51

54

62

65

Summary . . . . .

List of References .

Appendix . . . . .

10.

11.

12.

13.

14.

LIST OF TABLES

Predicted and Observed Mean Proportions of S Decisions
by Control and Ability Conditions . . . . . . . . . .

Observed Mean Proportions of S Decisions by Ability

Condition, Control Condition, and Sex for MSU Experiments

Rejection Rates by Ability Condition, Control Condition,
and Sex for MSU Experiments in Per Cent . . . . . . .

Predicted and Observed Mean Proportions of S Decisions
by Ability Condition for SU and MSU Experiments:
Equal Control Condition . . . . . . . . . . . . . . .

Predicted and Observed Percentages of A Decisions:
Questionnaire Data, Two Alternatives . . . . . . . . .

Predicted and Observed Percentages of A and B Decisions:
Questionnaire Data, Three Alternatives . . . . . . .

Percentage of Disagreeing Trials in Phase I and Phase II
by Ability Condition for SU and MSU . . . . . . . . .

Comparison of Estimates of a from.Phase I Based on All
Trials and Disagreeing Trials Only . . . . . . . . .

Comparison of Predictions Using [HH] and [LL] Conditions
for Estimating Parameters: Full Control Condition,
MSU Experiment . . . . . . . . . . . . . . . . . . .

Predicted and Observed Mean Proportions of S Decisions
for [LH] Ability Condition for All Degree of Control
Conditions . . . . . . . . . . . . . . . . . . . . .

Predicted and Observed Mean Proportions of S Decisions
by Ability and Control Conditions . . . . . . . . . .

Estimates of a_Based on MSU Data . . . . . . . . . . .
Estimates of R from Balkwell Article . . . . . . . . . .

Rejection Rates by Ability Condition and Level of
Self-Consistenqy . . . . . . . . . . . . . . . . . .

vi

15

22

24

25

27

28

31

34

36

37

4O

43

53

15.

16.

17.

Observed Mean PrOportions of S Decisions and Their

Standard Deviations by Ability Condition and Level

of Self-Consistency . . . . . . . . . . . . . . . . . . . . 55
Top-Bottom Preferences . . . . . . . . . . . . . . . . . . . 56

Joint Distribution of Right and Wrong Trials for Both
Low and High Ability Subjects in Phase I . . . . . . . . . 64

vii

LIST OF FIGURES

Utility Tree for the Gain-Loss Model . . . . . . . . .
Sample Stimulus for Meaning Insight Task: Phase I . .

Sample Stimulus for Meaning Insight Task: Phase II

Sample Stimulus for Spatial Judgment Task: Both Phases .

Sample Stimulus for Spatial Judgment Task: New Experiment

Observed Mean Proportions of S Decisions by Trial Blocks
by Ability Condition and Level of Self-Consistency

viii

12

18

20

21

50

61

General Nature of the Problem

Recently there was a television program entitled "Man's Thumb on
Nature's Balance." Man does affect his world through his behavior.

But so do all other organisms in nature. The difference is that man
can see alternative courses of behavior and can consciously choose
amongst the alternatives. Apparently other organisms cannot do this.
This phenomenon of consciously choosing among alternative courses of
behavior is called decision-making.

Decision-making is not a singular process, but is best considered
to be composed of a set of processes. Minimally, decisionrmaking may
be conceptualized as three processes. First, there is the process of
determining the set of alternatives. Next, the process of evaluating
these various alternatives as to the value or utility of each alter-
native; this is called the utility structure. The third process is
the mechanism by which one alternative is selected. This is called
the decisionemaking mechanism.

Any complete theory of decision-making must deal with these various
processes. However, the concern here will mainly be with the latter two
processes. Consequently, certain assumptions will be made and certain
steps taken to fix the other processes not being studied.

Camilleri and Berger (1967) have developed a gain-loss theory of
decision-making in which the selection of a given alternative is repre-
sented by a probability process. This theory assumes a fixed set of
alternatives. It posits a process of establishing the utility structure
that is based on the work of Homans (1961) on small group behavior and
on the ideas of Festinger (1957) on decision-making. An expectation

experiment was conducted to test the gain-loss formulation. The model

2
on the whole is supported by the data, but there are certain anomalies
in the fit.

In an attempt to deal with these anomalies, Balkwell (1969)
questioned an assumption of the attributed utility structure. Speci-
fically, he questioned whether the utility of self-consistenqy in the
behavior of the decision maker is independent of the abilities of the
decision maker and the others in his social environment, as had been
assumed in the experimental test of the gain-loss model. Balkwell
showed that by dropping this assumption and assuming that self-
consistency is related to the distribution of ability, the anomalies
in the fit of the model to the data can be greatly reduced.

In the research to be described, the assumed utility structure and
the dependency of self-consistent behavior are re-examined and certain
shortcomings in the design of the expectation experiment are discussed.
A new experiment has been conducted to study the utility of self-
consistency by bringing it under experimental control. This experiment
incorporates some modifications to the Camilleri-Berger design of the

expectation experiment.

General Theory and Research in Decision-Making

It is useful to conceptually divide decisionemaking into three
processes: (1) determining the set of alternatives, (2) attributing
the utility structure, and (3) the decisionemaking mechanism. In the
literature on decisionrmaking little attention is paid to the first
process. The set of alternatives are usually assumed to be obvious and
determined by theories of decisionemaking and, in the tests of these
theories, the alternatives are indeed obvious and determined. However,

social scientists who are not specifically interested in decision-making

3
do consider the determining of the set of alternatives as other than a
degenerate determined process. The process of discovering and developing
less obvious or new alternatives has often been an important concern in
social science (of. Peter Berger, 1963, pp. 136-150) and is becoming
more so.

Generally theorists in decision-making do not deal with the process
of determining the set of alternatives. An exception is Herbert.A.
Simon. Simon (1957, pp. 204-205) distinguishes two types of models of
decisionemaking; optimizing and satisficing. According to the Opti-
mizing or maximizing model, the decision maker determines the set of
alternatives, evaluates the utility of each alternative, then chooses
that alternative with the maximum utility. But according to the satis-
ficing model, the decision maker first determines a "threshold of
acceptability" that the chosen alternative must attain, then starts
evaluating the utility cf various alternatives. If an alternative is
"good enough",’that is, greater than the threshold of acceptability,
it is chosen and the process ends. With the satisficing model, a com-
plete set of alternatives is not developed unless all the alternatives
explored by the decision maker are not good enough. A short-coming with
the satisficing model is that apparently the process by which alterna-
tives are evaluated could make a difference as to what decision results.

The Gullahorns have compared the relative usefulness of the Opti-
mizing and satisficing models of decisiondmaking. They have devised a
representation of each of these models in their computer simulation model
of role conflict resolution. They conclude that the computer simulation
model incorporating the satisficing model more nearly represents actual

survey data concerning a decision situation than does the computer

4
simulation model incorporating the optimizing model (Gullahorn and

Gullahorn, 1965, p. 15).

Gain-Loss Theory;

 

The theory of decisionemaking which is of primary concern in this
paper is the gain-loss formulation developed by Camilleri and Berger
(1967). (In discussing this theory here, a notation system.will be used
different from that of Camilleri-Berger.) This theory assumes that the
set of alternatives is obvious and determined. In specifying the
utility structure this theory assumes that an alternative can be divi-
ded into various components each of which has some value or utility to
the decision maker. A numerical value (ui) can be assigned to a utility
component. A component of an alternative can either be a gain to the
decision maker or a loss. The utility of a gain can be represented as
ugi and of a loss uli' Both gains and losses are represented by posi-
tive numbers. They are distinguished by labeling them either gains or
losses. The utility of any component is assumed to be probabilistic
pi(ui), including certainty, i.e., 0‘5 pi S 1. If the utility to the
decision maker is conceptualized as having a source, the source can be
external, a thing or another person, or internal, the decision maker

himself.

According to the theory for an alternative Aj’ the gains are

assumed to be additive, i.e., GA. = E3pg,(u ). Likewise the losses
J i 1 gi
are assumed to be additive, i.e., LA» = 2 pl (ul ). The gains and
J i 1 1

losses of an alternative are kept separate and are not additive.
The gain-loss theory specifies that the expected gain of an alter-
native (E(Aj)) is the sum of the gains of that alternative and the losses

of all other alternatives, i.e., E(Aj) = GA. + 2 LA This assumption

3 17“] i.
is based on hypotheses of Homans and Festinger.

5

Homans (1961) deals with interaction. This means that the alter-
natives of the person's that Homans deals with are whether the person
will engage in or terminate social interaction with this person,
whether the person will engage in social interaction with that person,
and so forth. Therefore, the person contemplating a course of social
interaction is concerned with the rewards and punishments that he
expects from the person with whom he is contemplating that particular
course of social interaction. Although Homans discusses the situation
where the other person emits punishments to the person, he does not use
it: "In the exchange of hostilities only one side can win; the other
runs away, and it takes two to make social behavior" (Homans, 1961,

p. 57). For this reason Homans does not incorporate the emission of
punishment in his theory of social interaction. Homans does deal with
cost; the cost "of a unit of activity is the value of the reward obtain-
able through a unit of an alternative activity, foregone in emitting the
given one." (Homans, 1961, p. 60). Homans does not view an alternative
as possibly resulting in both reward and punishment. He views it as
resulting in either reward or punishment. Because of this Homans does
not postulate that a punishment avoided is a reward. Festinger does
break an alternative up into favorable and unfavorable components and

he proposes this postulate.

According to Festinger (1957, p. 40) an alternative may have
favorable characteristics and unfavorable characteristics. Festinger
says that when a person chooses between two alternatives, A and B, the
set of all the favorable characteristics of alternative A and of all the
tinfavorable characteristics of alternative B steer the person in the

(lirection of choosing alternative A. Similarly, the set of all the

6
favorable characteristics of alternative B and of all the unfavorable
characteristics of alternative A steer the person in the direction of
choosing alternative B. The person is pushed in two opposite directions
at once and is said to be in a state of conflict. Conflict is not to
be confused with dissonance; there is no dissonance in the pre-decision
process and therefore the person experiences no pressure to reduce
dissonance.

The implication is then that a gain foregone is a loss and that a
loss avoided is a gain. Therefore in the gain-loss formulation the
expected gain of an alternative are the gains (only) of that particular
alternative and the losses (only) of all other alternatives avoided.

The expected loss of an alternative is not a factor in the model. Since
the losses of all other alternatives must be known before the expected
gain of an alternative can be determined, the full set of alternatives
must be specified first and must not change throughout the decision
process.

The gain-loss formulation employs a probabilistic decision-making
mechanism. Its predictions are of the form, alternative x will be
chosen with probability y. The theory hypothesizes that the probability
of an alternative being chosen (P(Aj)) is proportional to its expected

galn, l.e.,

 

m1) ___ mg) 3 P013) z z POL“).
E(A1) E(A2) E(A3) ° ° ° E(An)

This is the basic probability assumption of the gain-loss theory.

Expectation Experiment

Berger and Camilleri carried out research to test their gainqloss
theory using the expectation experiment. The experiment was designed
by Berger and Snell (1951) to test their theory of self-other expecta-
tions. This basic expectation experiment has been used with minor
modifications as the decisionemaking situation in most studies related
to the gain-1053 model.

The subjects in the experiment are told that they are part of a
study designed to investigate how individuals (Phase I) and groups
(Phase II) make decisions. In the original test of the gain-loss model,
two subjects of the same gender were used. The subjects were students
at Michigan State University (MSU) and Leland Stanford Jr. University
(SU). There are two experimenters. The stimulus and the associated
ability were different at MSU and SH. These are described in detail
later in the section discussing the replication study.

The task involves a binary-choice stimulus. The subjects are led
to believe that for each stimulus one alternative is 'correct' and the
other 'incorrect'. Actually neither alternative is correct; there is
no right answer. The stimuli are designed so that the probability with
which either alternative is chosen is about equal, i.e., one-half. The
subjects are told that how well they do at the task is a measure of how
Inuch they have of a certain ability. The specific ability varies; the
£1bilities used will be described in the section below discussing the
Ireplication study. The subjects are given a series of trials to test
‘their task ability. The subjects make their choices individually,
separated by a partition using push buttons on a panel in front of them.

tTheyare given five seconds to study the stimulus. They get trial by

8

trial information on the supposed 'correctness' of their own and the
other's choices. Dependent upon the experimenter assigned condition,
a subject is either manipulated into a low or high ability state. For
low ability the subject gets about half of the trials ‘correct' which,
according to what he is led to believe are national standards, is very
poor. For high ability the subject gets almost all of the trials
'correct' which, according to the 'standards', is superior. With two
ability states and two subjects there are four ability conditions from
the perspective of one of the subjects:

[Hm] subject high, other subject low

[HM] subject high, other subject high

[LL] subject low, other subject low

[LE] subject low, other subject high.

Note that the four ability conditions are produced by three
different experiments. Each equal-high experiment produces two subjects
in the [HR] condition. Similarly for equal-low and [LL]. Each differ-
ential ability experiment produces one subject in the [ML] condition and
one in the [LH] condition.

In Phase II of the expectation experiment, the subjects are placed
in a two-stage decision-making situation. Each subject individually
makes an initial choice by means of buttons on the panel in front of the
subject. When both have made their choices each receives information on
the other's choice by means of lights on the panel in front of the sub~
ject. Then each subject makes a final choice, also by means of buttons.
The subject receives no information on the other's final choice. They
have five seconds to study the stimulus before making their initial choice
and the same for the final choice. The initial choice information on the

other subject is controlled by the experimenter on critical trials (which

9
are all or almost all of the trials) to be opposite that of the subject's
choice.

On a critical trial then a subject has the information that he has
chosen one alternative and the other subject has chosen the other alter-
native. In.making his final choice between the alternatives, the sub-
ject makes an implied decision to stay with his initial choice or to go
with the other's initial choice. For example, if the subject chooses
alternative A as his initial choice, on a critical trial he will receive
the information that the other subject chose alternative B as his initial
choice. If the subject then again chooses alternative A as his final,
he has made an implied self (S) decision. If on the other hand the
subject chooses alternative B, the other's initial choice, he has made
an implied other (0) decision. (Technically the initial choice is a
decision also, but it is primarily the result of perceptual factors. It
is the implied self-other decision that is affected by primarily social
and personality factors and that is the phenomenon in the expectation
experiment being represented by the gain-loss model.)

Besides the four ability conditions, there are three degrees of
control conditions. The degree of control is regulated by the way the
team score is computed. In the equal control condition, the team scores
one point for each correct final choice by either subject. In the
differential control conditions one subject has full control and the
other has no control. Subjects are assigned degree of control in the
differential control experiment by drawing slips of paper. One slip of
paper says "Team.Decision Maker" and the other says "Team.Advisor". The
slip drawn is a random process not controlled by the experimenter. The

final choice of the subject with full control is the team's decision.

10

If the subject is correct the team scores two points. Whether or not
the final choice of the subject with no control is correct does not
affect the team score. Thus, the three degree of control conditions

are produced by two different experimental settings (equal and differ-
ential). Two subjects each with equal control are produced by each
equal control experiment. One subject with full control and one subject
with no control is produced by each differential control experiment.

At the completion of the experiment each subject is interviewed
individually. There are three purposes served by this post-experiment
interview. One is to check for failure of ability manipulation-that
the subject doesn't have the intended ability expectation for himself
or for the other subject. Indicators include such things as prior
acquaintance and differences such as age, race, and style of life
indicated by dress and speech. If a subject does not have the intended
ability expectations for self and other, the subject is not included in
the experiment.

Another thing probed for in the post-experiment interview is
suspicion of the deceptions employed in the experiment-~the stimulus,
Phase I scores, and Phase II information on other's initial choice. One
indicator is prior acquaintance with deception experiments. If a subject
indicates suspicion, he is not included in the experiment. Finally, the
subject is debriefed. The full purpose of the experiment and all the

deceptions involved are explained to the subject.

Gain-Loss Model of Expectation Experiment and Test

 

In applying the gain-loss theory to the expectation experiment the
following gainéloss model was developed. Figure 1 shows the utility

tree of the gain—loss model of the expectation experiment. Self and

11
other make initial choices. Self receives information that other
disagrees with him on the initial choice. Self then either makes a
self decision by making the same final choice as his own initial choice
or makes an other decision by making the same final choice as other's
initial choice, i.e., Opposite his own initial choice. The probability
of a self decision is denoted P(S) and the probability of an other
decision is one minus the probability of a self decision, i.e., 1 - P(S).
Either decision that the subject makes results in a final choice that
may be either right or wrong. So there are four possible outcomes, and
associated with each outcome is a set of utilities representing gains
and a set of utilities representing losses.

In applying the gain-loss model to the expectation experiment,
three utilities are assumed to constitute the utility structure. They
are self-consistency (ul), reward from other (uz), and reward from
experimenter (u3). The first, ul, is utility to self from self for
self-consistency between the initial and final choices. This is a gain
if the initial and final choices are the same, i.e., a self decision is
made, and is a loss if an other decision is made. Whether it is a gain
or loss is determined by the decision, i.e., the utility has a probability
of one.

The second, uz, is the utility to self from other for making a cor-
rect final choice. This is a gain if the final choice is correct and a
loss if it is incorrect. Since whether the utility is a gain or a loss
depends on whether the final choice is correct or incorrect, a probabi-
lityg a, is defined. The parameter a_is the subjective probability
that the subject's initial choice was correct given the other subject

disagreed with him, i.e., a critical trial. So the gain if a self

12

Gains Losses
Final Choice + u + O
, a Is Right L11 2 “3
Self
Decision
Disagree l-a Final Choice u + u
ment between IS Wrong U1 2 3
Self and
gtfie? in Final Choice u + u
.232: 18mg“ 2 3 “1
Other
Decision
a Final Choice
Is Wrong O u + u2 + u
P(S) = probability of a self decision
a_ = probability that initial choice is correct, given
disagreement
= utility of self-consistency
u2 = utility of reward from other for making a correct
final choice
u3 = utility of reward from experimenter for making a correct

final choice

Figure l.--Utility Tree for the GainéLoss Model.

13

decision is made is au2 and the loss is (l-a)u2. If an other decision
is made the gain is (l-a)u2 and the loss is au2.

The third utility is similar to uz except for its source. It,
u3, is the utility to self from the experimenter for making a correct
final choice. Similarly the gain if a self decision is made is au3 and
the loss is (l-a)u3. If an other decision is made the gain is (l-a)u3
and the loss is au3.

The u's are assumed to be independent of the ability condition.
Also ul and u3 apparently are assumed to be independent of the degree of
control condition. It is assumed that “2 varies with the degree of con-
trol, greater for full than equal and zero for no control. The parameter
2.13 assumed to be independent of the degree of control, but of course
is assumed to depend on the ability condition.

The utility structure then is as follows:

 

Decision Gain Loss
Self ul + a(u2+u3) (l-a) (u2+u3)
Other (l-a) (u2+u3) u1 + a(u2+u3)

The expected gains for making a self decision (S) or an other decision

(0) are as follows,

E(S) 2(ul + a(u2+u3))

E(O)

2(l-a) (u2+u3).
Since there are two alternatives and the sum of the probabilities of each
must add to one, the basic probability assumption of the gain-loss theory

reduces to

_ E(S)
P(S) “ E(S)+E(0) °

l4

 

u
Substituting, simplifing, and letting R = ____i___. yields,
u2"‘13
R + a
P S =
(> 12+].-

In order to make a prediction in the expectation experiment using
the gain-loss model, R and a_must be estimated. According to the
utility structure assumptions, R is constant for any control condition
and a_is constant for any ability condition. The manipulation in

Phase I is used to estimate a_based on the following assumed equation:

no. correct by self .
no. correct by self + no. correct by other

R is estimated for each control condition by using the observed results
of one of the ability conditions. Using the [RH] condition results in
the predictions given in Table l. The results obtained from the test

of the gain—loss model with about 30 subjects in each cell are also given
in Table 1. The results differ slightly from those presented in the
Berger-Camilleri (1967, p. 375) article due to a more accurate computation
procedure used here.

Two approaches are used to compare the predictions of the gain-

1033 model with the results of the expectation experiment. First the
trends of the observations are as predicted in two directions. The
observed proportion of self decisions decreases as you go down the full,
equal, and no control columns in Table 1 from [HL] to [HH] and [LL] to
[L3] as predicted. Also the observed data increase as you go across

the row for each expectation from full to equal to no control as pre-
dicted (except for the slight reversal of [LH] equal and no control).
iIn comparing the predictions with the observations, it can be seen that

five of the nine are 'close'. Of the other four that are not 'close',

15

TABLE 1
PREDICTED AND OBSERVED MEAN PROPORTIONS OF S

DECISIONS BY CONTROL AND ABILITY CONDITIONS

 

 

 

 

 

 

Ability Full Controll Equal Control2 No Controll

Condition Pred. Obs. Pred. Obs. Pred. Obs.
[HL] .75 .73 .75 .78 .81 .82
[HH] x .61 x .67 x .71
[LL] .51 .52 .57 .55 .71 .74
[LH] .47 .24 .59 .44 .50 .43

 

Note.--Based on data from Berger and Camilleri (1967, p. 375)
N = about 30 subjects for each cell

lMSU experiment
2SU experiment

x[HH] condition used to estimate parameters

16
three are systematic, viz., observed considerably less than predicted
and all are [LH] ability condition. The fourth that is not 'close',

[LL] full control, does not systematically appear elsewhere.

Replication Study

The original expectation experiment testing the gain-loss model for
the equal distribution of control condition was carried out at Stanford
University (SU). Later an expectation experiment, similar to that
carried out at SU, but different in several details, was carried out at
Michigan State University (MSU) replicating the equal control condition
(Camilleri, Berger, and Olin, n.d.). Since it had the same design as
the full and no control conditions which were a part of the original test
of the gain-loss model, it also serves as a check on the validity of
comparing the results of the equal control condition (carried out at SU)
with the full and no control condition results (carried out at MSU) as
was done in the original test.

The experiments performed at SU and MSU differed to some extent in
both phases, the task, the experimenters, and the subjects. In the first
or manipulation phase both subjects were given trial-by-trial feedback as
to the 'correctness' of their own choice and that of the other subject.
The pattern of this feedback was designed to maximize the number of
disagreements, i.e., trials in which one subject was told that he was
'correct' and the other that he was ‘incorrect'. The SU and MSU experi-
ments were the same on these points. But they differed on other points.
The first phase consisted of 12 trials at SU and 20 trials at MSU.
Subjects manipulated to a high ability expectation state were told at SU

‘that they got 10 of 12 trials 'correct' and at MSU that they got 17 of 20

17
trials 'correct'. For low ability, it was 6 out of 12 'correct' at
SU and 8 out of 20 'correct' at MSU.

The second phase consisted of a series of trials in which each
subject made both an initial choice and a final choice. This was the
decisionamaking situation that was used to test the gain-loss model.

At SU the second phase consisted of 25 trials, 22 disagreement or
critical trials in which the subject found himself in disagreement with
his partner on the initial choice and 3 agreement or non-critical trials
interspersed on the 6th, 13th, and 20th trials. At MSU the second phase
consisted of 20 contiguous disagreement or critical trials and no
agreement or non-critical trials.

In both the first and second phases at SU and MSU, the subjects
were shown slides presenting two alternatives and were asked to make a
binary choice. The specific tasks, however, differed at SU and at MSU.
At SU the task involved something called "Meaning Insight". An example
of a first phase slide is shown in Figure 2. The subjects were instruc-
ted that the non-English word in the top row was a phonetic spelling from
a fictitious language unknown to them but that it had the same meaning as
one of the English words in the bottom row. They were told that by com-
paring the sounds of the non-English word with the meaning of the English
word they could decide which word was 'correct'. The ability to do this
‘was called "Meaning Insight Ability". Each word set was selected on the
Ibasis of a pretest so as to represent as ambiguous a choice as possible.
Chlly those word sets which elicited selection of one alternative 40-60%
of‘ the time when shown to 100 pretest subjects were used in the task

setquence. The order of presentation of the word sets was randomized.

l8

LU-BOYEL

(A) (B)
LOVE SOFTNESS

Figure 2.--Sample Stimulus for Meaning Insight Task: Phase I.

From Berger and Conner (1969, p. 195).

19

The slides used in the second phase at SU were somewhat different
from the first phase. In each of these word sets the role of the English
and non-English words was reversed as in Figure 3.

At MSU the task involved something called "Spatial Judgment". The
slides in both phases were similar and contained a rectangle as in
Figure 4. The large rectangle contains 100 small rectangles colored
either black or white forming black and white figures. The subjects were
instructed for each slide to determine which color - either black or
white - was more predominate. The ability to do this was called "Spatial
Judgment Ability". In actuality each rectangle was half black and half
white. Like the SU task, each alternative on each slide was intended to
be selected about half the time. For the first phase the same set of
slides were shown in the same order. For the second phase the same
slides were shown in cyclic order with the starting point randomly
determined.

A different set of experimenters carried out the experiments at
SU and MSU, but there was some overlap in personnel intimately involved
with the procedure of the experiments at both places. This included
Dr. Camilleri.

All of the subjects in the SU experiments were male. Roughly half
of the MSU subjects were female and half male. A pair of subjects in the
same experiment were of the same gender.

Traditionally the most salient difference between the SU subjects
and the MSU subjects is sex; the SU subjects were all male, while half
‘the MSU subject pairs were male and half female. However, as Table 2
anows, the effect of the sex of the subject pair on the proportion of

shelf responses was neither significant nor consistent. For any of the

20

YESTERDAY

(A) (B)
TA-KIN TU-SAK

Figure 3.--Sample Stimulus for Meaning Insight Task: Phase II.

From Berger and Conner (1969, p. 195).

21

 

 

 

 

 

Figure 4. Sample Stimulus for Spatial Judgment Task: Both Phases

 

22

TABLE 2
OBSERVED MEAN PROPORTIONS OF S DECISIONS BY ABILITY CONDITION,

CONTROL CONDITION, AND SEX FOR.MSU EXPERIMENTS

 

 

 

 

 

 

Ability Full Control Equal Control No Control

Condition F NF M NM F NF M NM F NF M NM
[HL] .75 18 .71 14 .77 19 .75 12 .82 18 .80 12
Inn] .53 20 .57 ll .52 15 .57 14 .71 23 .71 14
[LL] .50 20 .55 10 .58 10 .55 20 .72 24 .79 9
[LB] .27 14 .21 17 .43 12 .42 18 .44 21 .41 14

 

From Camilleri, Berger, and Olin (n.d.)

F = Mean proportions of S decisions for female groups
M.= Mean proportions of S decisions for male groups
NF = Number of Females

NM = Number of Males

23
degree of control conditions, there is an ability condition where the
proportion of S decisions is greater for females, but also an ability
condition where the reverse is true. For [HH] or [LL] ability conditions
the same is true for degree of control conditions. For [HL] there is a
cell where the difference is two percentage points. For [LH] it is one

percentage point.

In Table 3 are presented rejection rates for male and female sub-
jects by experimental condition for MSU. The rejection rates are not yet
available from SU. It can be seen then that the rejection rates for
females are less than those for males except for two conditions, the [LH]
full and equal control conditions. The rejection rates vary by condition,
but in no pattern obviously related to experimental condition.

The effects of the differences between the SU experiments and the
MSU experiments have not been separated out, but as Table 4 shows the
differences between the SU experiments and the MSU experiments (all
subjects) in the proportion of self decisions is again neither signifi-
cant nor consistent. The SU predictions given vary slightly from those
reported in the original article (Berger and Camilleri, 1967). Those
in the 1967 article were based on a slightly erroneous computation of a_
for the [BL] and [LH] conditions. The correction has been made in the
predictions presented here in Table 4.

The conclusion from these two sets of experiments is that the SU set
of experiments and the MSU set of experiments produce similar corrobora-
ting findings even though the procedures used at SU and MSU differ in
‘the trial schedule, the task, the experimenters, and the subjects. This
Ireplication adds considerable weight to the original findings. Also it

\ralidates the use of both MSU and SU data in the original test.

24

TABLE 3

REJECTION RATES BY ABILITY CONDITION, CONTROL CONDITION, AND SEX

FOR MSU EXPERIENTS IN PERCENT

 

 

 

 

 

 

Ability Full Control Equal Control No Control

Condition F M F&M F M F&M F M F&M
[ML] 25 35 30 17 33 24 10 29 19
[HH] 35 52 43 27 39 33 21 35 27
ulJ 29 47 35 9 25 21 8 35 18
[LH] 18 15 15 40 28 33 19 33 25

 

From Camilleri, Berger, and Olin (n.d.)
N's are given in Table 2

F = Female subjects
M = Male subjects
F&M = Female and Male subjects Combined

25

TABLE 4

PREDICTED AND OBSERVED MEAN PROPORTIONS OF S DECISIONS

BY ABILITY CONDITION FOR.SU AND MSU EXPERIMENTS:

EQUAL CONTROL CONDITION

 

 

 

 

 

SU MSU
Ability Expgriments Experiments
Condition Pred. Obs. N Pred. Obs. N
[ML] .75 .78 29 .77 .76 31
[HH] x .67 31 x .64 30
[LL] .67 .65 32 .64 .66 30
[LH] .59 .44 28 .52 .42 3O

 

From.Camilleri, Berger, and Olin (n.d.)

x[HH] condition used to estimate parameters

26

An Application of the Theory to Questionnaire Data

 

To demonstrate the applicability of the gain-loss theory of
decision-making to situations other than those presented in laboratory
experiments, the theory has been applied to questionnaire data (Olin,
1966). In a field study of social tensions in labor union relationships,
John T. Gullahorn (1956) gathered, among other information, questionnaire
data concerning a union stewardship vs. employees' club office dilemma.
Each of the respondents, 148 members and officers of a local, were
presented the dilemma in eight hypothetical situations and were given
three alternatives to decide between -- A, B, or C.

The gain-loss model was applied to the questionnaire data both as
a two-alternative simplified situation and as a three-alternative
situation. Four of the eight hypothetical situations were used to
estimate the parameters of the hypothesized utility structure. The pre-
dicted and observed percentages of.A decisions for the two-alternative
situation are presented in Table 5. Since the percentages for alterna-
tives A and B sum to 100%, only those for alternative A are given. As
can be seen, the predicted and observed percentages are quite 'close'.
The predicted and observed percentages for A and B decisions for the
three-alternative situation using a revised utility structure are pre-
sented in Table 6. Since the percentages for A, B, and C sum to 100%,
only those for alternatives A and B are given. Again the predicted and
observed percentages are quite 'close', but not as close as for the two-

alternative situation.

Problems With the Design of the Expectation Experiment
Let us return again to the situation in which the gain-loss model

was originally tested, the expectation experiment. Although the

27

TABLE 5
PREDICTED AND OBSERVED PERCENTAGES OF

A DECISIONS: QUESTIONNAIRE DATA, TWO ALTERNATIVES

 

 

 

Situation Predicted Observed
1 34.6 33.7
4 79.0 80.4
6 82.3 81.0
8 95.0 90.5

 

From Olin (1966, p. 35)
N varies between 83 and 116 respondents

28

TABLE 6

PREDICTED AND OBSERVED PERCENTAGES OF A.AND B

DECISIONS: QUESTIONNAIRE DATA, THREE ALTERNATIVES

 

 

 

 

 

Situation Alternative A Alternative B
Pred. Obs. Pred. Obs.

1 20.8 18.9 39.3 37.2

4 51.4 60.4 13.6 14.9

6 52.9 54.7 11.4 12.8

8 61.8 70.9 3.3 7.4

 

From Olin (1966, p. 48)

N = 148 respondents

29
replication study demonstrated that the basic design produces similar
results in spite of certain changes, this does not say that the design
is without flaw. Certain aspects of the design.may be having unintended
effects on the subjects.

The problems with the design which are important here fall into two
categories, shifts in the experimental setting from experiment to experi-
ment and the incredibility of Phase II disagreements in certain of the
experimental conditions. One way the setting does not remain constant
is in the experimenter effect. Who is experimenter changes and the same
experimenter can come across differently on different days. Theoreti-
cally, this should affect the subjects' responses because of u3, the
reward from experimenter. The duration of the stimulus presentation and
time allowed to study stimulus are also not standardized. The five
seconds for studying the stimulus before making a choice was judged with
and sometimes without a timepiece. The stimulus remained visible, in
Phase I, until the subjects had made their choices and whether they were
correct or incorrect was announced. 80 the latency of the subjects and
the speed of the experimenters also affected the duration of stimulus
presentation. Similarly in Phase II many factors affect this duration.
The stimulus was visible through a five second study period, initial
choices, another five second period, final choices, and time for responses
to be recorded.

A more important set of flaws has to do with the credibility of the
Phase II disagreements and the dissimilarity between Phase I and Phase II
in this regard. Subjects were told that the tasks in both phases dealt
with the same ability, but the phases differed in two key respects.

First, the percentage of disagreeing trials that the subject experienced

30
was different in the two phases as Table 7 shows. The difference is
especially evident for the [HE] ability condition. In this condition
it would seem that the subjects would find it incredible that, although
they both have high ability, they disagree with each other 88 or 100%
of the time in Phase II.

Second, the source from.which disagreements are inferred is
different in the two phases. In Phase I after a subject makes a choice
he receives information on the other's choice only through the experi-
menter who verbally reports and records on a chart who is correct and
who incorrect. In Phase II the subject sees what the other person chose
by means of a light on his own board. It would seem that the disagree-
ments would be more salient to the subject in Phase II than in Phase I.
In Phase I the concern is with the correct/incorrect evaluation.
Inferring disagreement must go through this evaluation which comes from
the experimenter. In Phase II there is no such trial by trial evaluation
and concern and the fact of disagreement is plainly displayed on the
board in front of the subject in there being a light representing the
subject's choice under one alternative and a light indicating the
other's choice under the other alternative.

Another facet of credibility is the manipulation phase used at MSU.
At SU a low subject got six out of twelve trials correct. This is half
and seems compatible with low ability. At MSU a low subject got eight
out of twenty correct. This is less than half and the subject might infer
a negative ability rather than low ability. Also in the [LL] condition,
the two subjects only agreed when both of them were wrong. This might

also lead to an inference of negative ability.

31

TABLE 7
PERCENTAGE OF DISAGREEING TRIALS IN PHASE I

AND PHASE II BY'ABILITY CONDITION FOR.SU AND MSU

 

 

 

 

 

Ability SU MSU
Condition Phase I Phase II Phase I Phase II
[HL] 57 88 75 100
[RH] 33 88 30 100
[LL] 100 88 80 100

[LH] 57 88 75 100

 

32

Other differences between the two phases involved the stimulus.
On each trial the stimulus was shown for about twice as long in Phase
II as in Phase I. Also at SU different stimuli were used in the two
phases, although the subjects were told the two tasks involved the same
ability.

Thus there are several aspects of the design of the expectation
experiment which can be interpreted as flaws. There are also several

flaws in the gain-loss formulation of decisionemaking.

Problems With the Gain-Loss Theory

 

The gain-loss theory of decision-making can be discussed at a
general theoretical level, at the level of its application to the expec-
tation experimental situation, and at the level of experimental results.
In general a crucial step in applying the theory to any decision situa-
tion is specifying the assumed utility structure. Assumptions must be
made about the social environment of the individual and his internal
make-up, about what are the relevant utilities, and about what their
probabilities are. It has been mentioned that in applying the theory to
questionnaire data, three-alternatives, using a revised utility structure,
the model fit the data quite well as shown in Table 6. A revised utility
structure was used because the originally hypothesized structure for
three alternatives failed miserably. The point is that the theory pro-
vides little in the way of guidelines for its application. This can be
viewed as a shortcoming since a failure of the predictions to fit the
data can be blamed On its application rather than the theory itself. It
can also be viewed as asset. The theory can be used as an heuristic in
the finding of additional relevant utilities as was done in specifying

the revised utility structure.

33

The gain-loss theory has been experimentally tested in situations
of two-alternatives only. Fbr three or more alternatives there is a
possibility that the decision process is a two or more step process.

This would change the predictions of the theory and therefore warrants
investigation.

In applying the theory to the expectation experiment there are some
specific problems dealing with §_and the other utility parameters. The
subjective probability of being correct given a disagreement, g, is
estimated on the basis of Phase I. Is this valid since the phases are
different as noted earlier. Also §_is applied to disagreeing trials only
in Phase II, but is estimated from both agreeing trials and disagreeing
trials in Phase I. If based only on disagreeing trials in Phase I, the
estimates are different as can be seen in Table 8.

In applying the theory to the expectation experiment a utility
structure is assumed. Whether the particular utilities assumed and
their probabilities is correct is always open to question. Also what
they are assumed to be independent of or dependent on is open to question.
These types of dependency assumptions are not easily tested.

At the level of experimental results the model does not fit the data
for four cells as described earlier and indicated in Table 1. One of
these is the [LL] full control cell. This appears to be an anomaly since
it is not the case for the other [LL] or [HR] (discussed later) cells
(equal and no control) nor the other full control cells except for [LB]
which will be discussed later. Because of this it is suspected that it
is an anomaly in the data rather than a fault in the theory that is the

problem.

34

TABLE 8
COMPARISON OF ESTIMATES OF a_FROM PHASE I

BASED ON ALL TRIALS AND DISAGREEING TRIALS ONLY

 

 

 

 

 

SU MSU_
Ability All Disagreeing All Disagreeing
Condition Trials Trials Only Trials Trials Only
[HL] .53 .75 .68 .80
[HH] .50 .50 .50 .50
[LL] .50 .50 .50 .50

[LH] .38 .25 .32 .20

 

35

The theory predicts that the mean for [BB] and [LL] for a control
condition will be the same. Since [HR] was used to estimate parameters
it is the [LL] full control that is the problem, If [LL] is used to
estimate parameters, it is [HH] full control that is the problem. Since
any ability condition can be used to estimate parameters, the one that
gives the worst predictions for the other cells is the best candidate for
being an anomaly. .As Table 9 shows, using the [RH] condition to make the
estimates gives the worst predictions. This implies that it is the
anomaly, that the data from the [BB] full control condition somehow are
at fault.

The other three cells in Table l where the predictions of the gain-
loss model do not fit the data from the expectation experiment are the
[LH] full, equal, and no control conditions. As Table 10 shows, the MSU
equal control condition and two more conditions(high and low control)
reported by Balkwell (1969), to be discussed in the next section, give
similar results. The gain-loss model consistently over-predicts by an
average of 16 percentage points. From this fact it would appear that
the fault is in the gain-loss model and not in the design of the

expectation experiment.

Balkwell MOdifications to Model

 

With it being obvious that the gain-loss model does not fit the data
in certain respects, Balkwell (1969) using the data as a guide and based
on certain theoretical considerations related to self-esteem.maintenance
modified the model so that it would better fit the data. The heart of
the modification is 3a below. The rest of his hypothesized utility

structure are also presented.

36

TABLE 9
COMPARISON OF PREDICTIONS USING
[HR] AND [LL] CONDITIONS FOR ESTIMATING PARAMETERS:

FULL CONTROL CONDITION, MSU EXPERIMENT

 

 

 

 

Ability Predicted Using: Observed
Condition [THU ILL]
[EL] .75 .73 .73
[HH] x .51 .61
[LL] .61 x .51

[LH] .47 .43 .24

 

N = about 30 subjects per cell

x this condition used to estimate parameters

37

TABLE 10
PREDICTED AND OBSERVED MEAN PROPORTIONS OF S DECISIONS
FOR [LR] ABILITY CONDITION FOR ALL DEGREE OF CONTROL CONDITIONS

 

 

 

 

Degree of Control
Equal
Full High1 SU MSU Lowl No
Observed .24 .32 .44 .42 .38 .43
Predicted .47 .54 .59 .52 .52 .60

 

1This control condition is discussed in the next section

38
l. u3, reward from experimenter, is independent of
control and ability conditions.
2. uz, reward from other, is directly related to the
degree of control of self (inversely related to the
degree of control Of the other, control being a
zero-sum quantity) and independent of ability
condition.
3. ul, self-consistency, is related to both ability and
degree of control
a. it is directly related to degree of equality of,
ability, i.e.,
u1(HH) = ul(LL) > ul(HL) = ul(LH)

b. it is inversely related to degree of control, i.e.,
ul(full) < u1(equa1) < u1(no)

c. it is directly related to degree of equality of

control, i.e., the relationship in 3b is not

R(full) + R(no);

uniform, but R(equal) > 2

rather than being equal.
It is assumption 3a that is the big change. It makes the utility ratio,
R, dependent on the ability condition. In the original gain-loss model
as it is applied to the expectation experiment, the utility ratio is
assumed to be independent of the ability condition.
values for the parameters are determined by a combination of

assumptions and estimation procedures using the original expectation
experiment data.

1. “2 is assumed to be 2 for full, 1 for equal, and O for

no :ontrol conditions.

39

2. §_is assumed to be 0.25 for [LH], 0.50 for [LL] and [HH],
and 0.75 for [BL] ability conditions.

3. making these assumptions, u3 is estimated from.the data
to be 0.85.

4. similarly ul is estimated to be for [BB] and [LL] ability
conditions 0.388 for full, 0.871 for equal, and 0.668 for
no control conditions, and for [BL] and [LH] ability con-
ditions 0.000 for full, 0.483 for equal, and 0.280 for no
control conditions.

Since the estimations are based on the original expectation experi-
ments, it would be expected that the predictions should be quite good.
And they are with a maximum.discrepancy ([HH] full control) of 5 percen-
tage points for all nine cells and a maximum of 3 percentage points for
the three [LH] cells where the problem.lies.

Balkwell applied his fOrmulation to data from an expectation experi-
ment that had two conditions of partial control, high, which is between
full and equal in degree Of control, and low, which is between equal and
no control. The results are presented in Table 11 for approximately 20
subjects in each cell. The fit looks quite 'close' for the eight cells
except possibly for the two [LL] cells. But the fit for the [LH I
ability conditions is a definite improvement over the fit of the original

gaineloss model to the original expectation experiment.

40

TABLE 11
PREDICTED AND OBSERVED MEAN PROPORTIONS

OF S DECISIONS BY'ABILITY AND CONTROL CONDITIONS

 

 

 

 

 

Ability High Control Low Control

Condition Pred. Obs. Pred. Obs.
[EL] .78 .81 .80 .80
[HR] .52 .55 .57 .55
HIJ .62 .56 .67 .60
[LH] .35 .32 .41 .38

 

From Balkwell (1969, p. 472)

N = about 20 subjects per cell

41
Re-Analysis of the MedelAData Discrepancy -- Expectation Experiment

It is obvious that the (original) gainéloss model and the data
from the expectation experiment do not coincide in all respects, there
is a certain systematic difference. (The anomaly of the [HM] full con-
trol data is not of concern here.) There are two possible reasons for
this. The data may be faulty due to factors in the experimental design,
or the model is inadequate.

The first can be reduced by eliminating many of the faults in the
experimental design which were discussed in the section labeled problems
with the design of the expectation experiment. Shifts in the experimen-
tal setting from experiment to experiment can be reduced by standardizing
the setting. Tape recorded instructions can be used in both phases. The
stimulus can be presented for an exact, fixed duration, say two seconds,
using an interval timer. The two phases still would probably'have to
differ on this point. Stimulus would be shown once in Phase I and twice
in Phase II on each trial.

The similarity of Phase I and Phase II in other respects can be
increased. The proportion of disagreeing trials can be made the same
in Phase II as in Phase I by interspersing additional agreeing trials,
especially for the [BB] ability condition. In order to get the same
number of critical trials, the total number of trials would be different
for the several ability conditions.

Another way to increase similarity is to give the subject informa-
tion on the other's choice in Phase I in the same manner as in Phase II,
viz., by lights on his panel. This will make disagreements more salient

in Phase I. However, since the correctness of the choices are still

42
reported after each trial in Phase I, the salience of the disagree-
ments might be reduced by a concern for being correct.
The negative ability problem can be eliminated by making low
ability getting one half of the Phase I trials correct as it was in the
SU experiments. Also the stimuli should be the same type in Phase I and

Phase II as it was in the MSU experiments.

Re-Analysis of the ModeléData Discrepangy--Gain-Loss Model

If the fault is assumed to be in the gain-loss model, then some
aspect of the hypothesized utility structure must be dependent in ways
not hypothesized or some basic tenets of the model are incorrect.

The probability, a, could be dependent on the degree of control.
If the observed proportions are substituted in the prediction equation,
R, the utility ratio, is assumed to be independent of ability condition,
and a_for [HH] is assumed to be 0.600 for all control conditions, esti-
mates are obtained for gias presented in Table 12. These estimates
indicate that g could be inversely related to degree of control in that
3 tends to increase as degree of control is decreased from full to
equal to no cOntrol.

The utility ratio, R = E§:%_E§, could be dependent on the ability
condition. This would be so if ul and/or “2 and/or u3 were dependent
upon the relative ability. According to Balkwell's analysis R.wou1d
have to have values like those in Table 13. Here §_is assumed to be
independent of control condition and to have the values 0.75 for [HL],
0.50 for [HH] and [LL], and 0.25 for [LH].

Balkwell hypothesizes that it is u1 that is dependent on the ability

condition and he makes a theoretical case for this, but it could be uz

TABLE 12

ESTIMATES OF a_BASED ON MSU DATA

 

 

 

Ability Full Equal No
Condition Control Control Control
[HL] .72 .73 .75
[HH] .50 .50 .50
[LL] .50 .52 .54
[LH] .22 .35 .21

 

Estimates made under following conditions:

observed proportions are substituted in the
prediction equation

R is assumed to be independent Of ability condition

a_for [HH] assumed to be 0.60 for all control
conditions

44

TABLE 13

ESTIMATES OF R FROM BALKWELL.ARTICLE

 

 

 

Ability Full Equal No
Condition Control Control Control
[HL] .000 .251 .329
[RH] .135 .471 .785
[LL] .135 .471 .786
[LH] .000 .251 .329

 

From Balkwell (1969, p. 461)

is assumed to be independent of control condition

[KL] = 0.75, a = 0.50, a. = 0.25

a
‘3‘- -[HR] = 2int] [LH]

45
and/or us that is dependent upon the ability condition. In the present

research, ul is assumed to be dependent on the ability condition.

Theory Tested by New Experiment

 

The following theoretical development is an attempt to explain why
the gain-loss model of the expectation experiment does not fit the data.
Recall that the utility of self-consistency, denoted ul, is defined in
the gain-loss model by the way ul enters into the utility structure of

the decision situation and by the way ul enters into the prediction

equation for a self decision, viz.,

 

“2+ u3

It is here hypothesized that the conceptualization of the utility of
self-consistency as 111 is inadequate.

Let there be a utility called the true utility of self-consistency
in the expectation experiment and denote it by c for consistency. It
is assumed that c is independent of the relative ability of the subject
as indicated by a: The gain-loss model posits that ul = c, that is,
ul, as it enters into the gain-loss model, is the same as c, the true
utility of self-consistency in the expectation experiment. Contrary to
the gain-loss formulation, it is here hypothesized that ul and c are not
the same, but rather ul is some function of c and a:

According to the gain-loss model, ul, uz, u3, and a_are all indepen-
dent of each other. If u1 is varied but uz, u3, and a left unchanged,

P(S) according to the gain—loss model will vary directly with ul. Under

46
these conditions variations in ul will be indicated by variations in
P(S). Since according to the gain-loss model u1 = c, variations in c
also will be indicated by variations in P(S) for a given u2, u3, and a,
any given as So it is predicted by the gain-loss model that variations
in c will be reflected in similar variations in P(S) for the [EL]
ability condition where a_is large; and also, these same variations in
c will be reflected in similar variations in P(S) for the [LH] ability
condition where a_is small.

But it is hypothesized here that this will not be the case. If u1
is a function of both c and a, then variations in c will be reflected
differently in the variations of P(S) for the [BL] ability condition
where a_is large than in the variations of P(S) for the [LH] ability
condition where a_is small. The hypothesis tested, then, is as follows:

Variations in c, the true utility of self-consistency, will

affect P(S) differentially, depending on whether the ability

condition is [HL] or [LH] .

If the experimental results show this hypothesis to be true, then it is
indicated that u1 is a function of a_as well as c; and the conceptuali-
zation of the utility of self-consistency in the gain-loss model of the
expectation experiment is inadequate. If the hypothesis is shown to be
false and P(S) for [HL] and [LH] are similarly affected by variations in
c, then ul is a function only of c and the conceptualization of the
utility of self-consistency in the gain-loss model is appropriate.

This theoretical development of the hypothesis tested here can be
formalized as follows.

Definitions:

1. P(S) is the observed mean proportion of self decisions.

47
u1 is a parameter in the gain-loss model, supposedly
a conceptualization of the utility of self-consistency,
but probably also a function of a,
c is the true utility of self-consistency in the
expectation experiment.
u2 is the utility of the reward from other for making
a correct final choice.
u3 is the utility of the reward from the experimenter
for making a correct final choice.
a'is the subjective probability that the initial choice

is correct given a disagreement.

Assumptions:

1.

E can be estimated from the manipulation phase as follows:

number right by self
number right by self and other

 

For both the Camilleri-Berger MSU experiments and this
experiment, a for the [BL] ability condition is 0.68 and
a_for the [LH] ability condition is 0.32.

c, u2 + u3, and a_are all independent of each other.

u1 is an unspecified function of c and a.

u1 and u2 + u are independent of each other.

3
Instructions can be added to the original instructions
given in the expectation experiment to vary c and leave

“2 + u3 and a_unchanged.

48

6. The relation between the parameter ul and P(S) is

___“L.
u2"“3
___"_}_
u2 + u3

+2].

= P(S)

 

+1

Derivations:
1. Variations in ul will be indicated by variations in P(S)

for any u2 + u and a,

3

2. Variations in c will be reflected by variations in u1
for any a,

3. Variations in c will be indicated by variations in P(S)
for any u2 + u3 and a,

Hypothesis:
Variations in c will be indicated by variations in P(S)
differentially, depending on whether a_is large ( [HL]
ability condition) or small ( [LH] ability condition)

for any uz + u3.

Design of Experiment

 

There were three main experimental conditions in the experiment
conducted, varying c, the true utility of self-consistency—-c standard,
c high, and 0 very high--and two ability conditions--subject high and
other low [HL], and subject low and other high [LH]. In all the
conditions the experimental design was basically that of the expecta-
tion experiment used by Camilleri and Berger for the equal control
condition, but incorporating some of the improvements discussed here in
the analysis of the expectation experiment. Improvements introduced

included tape recorded instructions, stimulus presentation controlled by

 

49
an interval timer, and keeping the stimuli of the same type in Phase I
and Phase II. In order to retain comparability with previously con-
ducted expectation experiments, some aspects in the design of the
expectation experiment which I have pointed out as being possible flaws
have been retained in the experiment reported here. These are the
proportion of disagreeing trials not being the same in Phase II as in
Phase I, information on other's choice in Phase I being implied only
through the trial by trial report of the experimenter of who is right
and who wrong, and low ability being produced by a score of less than
50 per cent correct in the manipulation phase, Phase I.

The ability involved in the experiment conducted is Spatial
Judgment Ability, but the task was slightly different from that in the
expectation experiment. The stimuli were of the sort shown in Figure 5.
Six different slides were presented according to a fixed random schedule,
one upside down. The alternatives were top or bottom, which has more
white. These stimuli eliminate the bias introduced by any black-white
color preferences the individual subject may have. There were twenty
trials in Phase I. As in the previous MSU expectation experiments, low
ability was indicated by getting eight 'correct' and high ability by
seventeen 'correct'. Phase II had twenty-five trials with manipulated
agreements on trials six, thirteen, and twenty, and disagreements on all
other trials. There were twenty-two critical trials.

In the 0 standard condition, the rest of the instructions were the
same as in the original equal control expectation experiment. The
instructions and other details of the experiment are given in the
Appendix. In the c high and 0 very high conditions, 0 was increased to

two different levels by means of experimental instructions. These are

50

 

 

 

 

 

 

 

 

 

 

 

Figure 5. Sample Stimulus for Spatial Judgment Task: New Experiment

51
given in the Appendix. The added instructions in both cases consisted
of a paragraph inserted in the instructions for the standard condition.
The paragraph for c very high was basically the paragraph for 0 high
with several statements strengthened and several sentences added. In
principle, the instructions were meant to define a group of people as
conformers and to portray them in a bad light. The intent was to in-
crease the utility of anti-conformity, which is assumed to be an impor-
tant component of self-consistency. Behaviorally, anti-conformity and
self-consistency are the same in a situation where the choices are
binary and the information from other indicates disagreement, which was

the situation in the experiment.

Results of Experiment

 

The experiment was conducted Spring 1971 at Michigan State
University. The subjects were male MSU students recruited from
University Social Science and Introductory Sociology classes, primarily
first and second year courses. Participation was voluntary, and pros-
pective subjects were told that they would receive at least $2.00 for
participation in a study that would take about an hour on how individuals
and groups solve certain types of simple problems. They were paid $2.00
for participation. Prospective subjects were screened on the basis of a
prelimdnary questionnaire. In order to reduce suspicion, those who had
had courses where deception experiments probably were discussed, such as
the courses in social psychology taught in the Sociology Department and
the Psychology Department, were eliminated. And in order to get some age
and education homogeneity, those older were likewise eliminated. A total

of 214 subjects were run. At the end of an experiment each subject was

52
interviewed individually to determine if he should be eliminated from
the sample. (See the Appendix for the interview schedule.) The
following were the reasons used to eliminate subjects from the sample:
1. Deliberately making wrong initial choices.
2. [Misunderstanding instructions.
3. Prior acquaintance with other subject which interferes
with process (change of expectation manipulation, friendship
determining acceptance of influence, etc.)
4. Status differences based on physical characteristics which
interfere with process.
5. Suspicion:
a. Volunteered information that exchange of information
was "rigged" (Phase I or Phase II).
b. Read previously about deception experiments and
thought present study was similar.
c. Heard from others that there was deception in
present study.
d. Previous participation in deception study and
belief that present study was similar.
Forty-six subjects were eliminated from the sample and their data are
not included here. See Table 14 for a breakdown on the distribution of
eliminated subjects. After the elimination of these subjects from the
sample, there remained twenty-eight subjects in each of the six conditions
of the experiment. Subsequent analysis was based only on the twenty-
eight subjects in each of the six conditions.
On each of the twenty-two critical trials in Phase II, for each

subject it was recorded whether he made a stay, or self, decision

53

TABLE 14
REJECTION RATES BY ABILITY CONDITION AND

LEVEL OF SELF-CONSISTENCY

 

 

Level of Self-Consistency

 

 

Ability 6 Standard c High 0 Very High
Condition
No. % No. % No. %
No. Re- Re- No. Re- Re- No. Re- Re-

Run ject ject Run ject ject Run ject ject

 

[BL] 33 5 15 37 9 24 38 10 26

[LH] 29 1 3 38 10 25 39 11 28

 

The true level of self-consistency in the present
experiment is represented by c and was manipulated by
verbal instructions.

54
(initial and final choices the same) or a change decision (initial
and final choices opposite). The total number of changes by each
individual was determined and those were grouped by condition. The
average proportion of self or stay decisions and its standard deviation
are reported in Table 15 by condition.

Also recorded for each subject was his initial choice on each of
the twenty-five trials in Phase II and each of the twenty trials of
Phase I. The top-bottom preferences for the six slides and one upside
down for the six conditions of this experiment and two conditions of a

similar one (Shelly, 1972) are presented in Table 16.

Integpretations of the Experimental Results--Gain-Loss Model

It is hypothesized that in the gain-loss model ul, the utility of
self-consistency, is dependent on a, the subjective probability of being
correct given that a disagreement has occurred. Therefore, as the sub-
jective probability of being correct given a disagreement is assumed to
be a function of the manipulated ability levels of the two subjects,
different levels of c, the true utility of self-consistency, will produce
different values of the observed P(S), the observed mean proportion of
self decisions in the [BL] ability condition (a large) from those produced
in the [LH] ability condition (a_small).

The results of the experiment, as shown by Table 15, indicate that
there is essentially no difference in P(S) for the [BL] ability condition
for the three levels of self-consistency--c standard, c high, and c very
high (P(S) = 0.77, 0.80, and 0.81 respectively). And there is essentially
no difference in P(S) for the [LH] ability condition for c high and 0 very
high (P(S) = 0.50 and 0.48 respectively). However, there is a great

difference for the [LH] ability condition between P(S) for c standard and

55

TABLE 15
OBSERVED MEAN PROPORTIONS OF S DECISIONS
AND THEIR STANDARD DEVIATIONS BY ABILITY CONDITION

AND LEVEL OF SELF-CONSISTENCY

 

 

Level of Self-Consistency

 

 

 

Ability 0 Standard 0 High c Very High
condition P(S) S.D. P(S) S.D. P(S) S.D.
[HL] .77a .086 .80a .114 .81a .114
[LH] .37b .182 .50C .168 .48C .177

 

Cells having a common subscript are not significantly
different at the 0.05 level using student's t test for
comparisons (two-tailed, df = 54).

N = 28 subjects for each cell.

56

TABLE 16

TOP-BOTTOM.PREFERENCES

 

 

 

Slide Type Per Cent Top Times Shown
F 35 7
F upside down 71 l
H 60 9
I 55 7
0 62 7
Y 66 7
OR 36 7
All 53

 

57
0 high (P(S) = 0.36 and 0.50 respectively). It is therefore concluded
that the distribution of P(S) over the three levels of self-consistency
is different for the two ability conditions, and that different levels
of c will distribute P(S) differently depending on a_because ul is
dependent on a, So ul, the conceptualization by the gain-loss model of
the utility of self-consistency, in the expectation experiment is in-
adequate.

The data can be used to get some notion of the way that ul is
dependent on c and a, As 0 is increased by going from the standard
condition to the high to the very high, P(S) either increases or remains
essentially the same for either ability condition. So u1 is apparently
a non-decreasing function of c. If it is assumed that in the very high
condition 0 could not or was not increased (this will be discussed shortly),
then the relationship between ul and c is decreased as a_is increased.

That there is a dependency of u1 on a was preposed in the section
re-analyzing the model-data discrepancy in terms of the model as a
possible explanation for the failure of the gain-loss model to fit the
results of the expectation experiment for the [LH] ability condition. In
order for the gain-loss model to better fit the data, the gain-loss model
needs to be reformulated under the assumption of dependence between “1 and
a.

The results of the experiment support the hypothesis concerning the
gain-loss model; what the results say about the experiment itself will now
be examined. The two sets of instructions designed to produce high and
very high levels of self-consistency did not produce differences in P(S)
either for [HL] ability condition (0.80 and 0.81) or [LH] ability condi-

tion (0.50 and 0.48). Two explanations can be offered. One, the

58
instructions for c high have produced a level of utility of self-
consistency that is at or near its limit, and stronger instructions
cannot raise the level of this utility. Two, the instructions for
0 very high were not sufficiently stronger than for 0 high to produce
a difference in the level of c, the utility of self-consistency. whether
the very high condition could not or did not raise the level of 0 above
that of the high condition, the point is that c was not raised. But,
the hypothesis that P(S) is differentially affected by a still is sup-
ported, since P(S) for 0 standard and high are essentially the same for
the [HL] condition, but greatly different for the [LH] condition.

The experimental design of this experiment is quite similar to that
of the Camilleri-Berger experiments, so it can be of interest to compare
them. A comparison of the results of this experiment for c standard
(see Table 15) with the results for the equal control experiments con-
ducted at Stanford University and Michigan State University (see Table 4)
reveals that there is very little difference for the [HL] ability condi-
tion (observed P(S) of 0.77 as compared with 0.78 and 0.76), but a great
difference for the [LH] ability condition (observed P(S) of 0.36 as
compared with 0.44 and 0.42). This difference could be due to differences
in the experimental design, viz., tape recorded instructions, experimenter
not present, and stimuli presented with an interval timer. In terms of
the gain-loss model, the different results might be accounted for by
differences in a_or in R = ;_E%_El. for the two experimental designs.

It would seem reasonable to :rgue3that with the experimenters not present,
u3, the reward from experimenter for making a correct choice, would be
smaller. But this would result in a P(S) for the [LH] condition for this

experiment greater than that for the Camilleri-Berger experiments,

59
exactly opposite of what was actually found. Or it could be argued
that instructions given by a machine have more presence than an experi-
inenter being present. This would argue for the observed results.

It is an assumption of the gain-loss model that each of the 22
critical trials presents the subject with the same decision situation.

.A necessary condition for this assumption is that the decisions a person
makes in the experiment are independent, i.e., the probability that a
person makes a self decision on a given trial does not depend on whether
he made a self or other decision on the previous trial. To check this,
the one-step transition matrix for the 22 critical trials was computed
for each of the six conditions. Whether the previous decision was self
or other made a maximum difference in the probability of a self decision
of 0.08 for the [HL] standard condition and less of a difference for each
of the other five conditions. The direction of difference was not sys-
tematic for the six conditions. The one-step transition matrix was also
computed for the first half and the second half of the critical trials
for the six conditions. The dependency was less in the second half for
all conditions except the [HL] high condition. It is concluded that the
dependency tends to be low and to decrease as the experiment progresses.
Thus, the independent trials assumption of the gain-loss model is
approximated in the experiment.

Another necessary condition for the assumption of the gain-loss model
that all the critical trials present the same decision situation is that
the proportion of self decisions be the same for each trial for each
condition. The proportion of self decisions was computed for each of
the four contiguous blocks of critical trials (trials 1 to 5, 7 to 12,

14 to 19, and 21 to 25) for each condition. This is presented in Figure 6.

 

60
For the [HL] ability condition for all three levels of self-consistency
the proportion of self decisions remains quite stable. For the [LH]
ability condition for all three levels of self-consistency the proportion
of self decisions decreases in a relatively consistent fashion over the
four blocks of trials by an average of 0.15 from first to last block and
the decrease over the four blocks of critical trials is quite similar for
the three levels of self-consistency. This failure of the [LH] ability
condition to meet the assumption of stability over trials of the propor-
tion of self decisions may affect the accuracy of the prediction of P(S),
which is not tested in this experiment. But since the decrease over the
four blocks was quite similar for the three levels of self-consistency
for the [LH] ability condition, it can still be concluded that different
levels of self-consistency will affect P(S) differentially depending on
whether the ability condition is [HL] or [LH], which is the hypothesis
tested in this experiment.

There is evidence then that P(S) remains relatively stable through-
out the 22 critical trials for the [HL] ability condition, but decreases
over the 22 critical trials for the [LH] ability condition, regardless of
level of self-consistency. A possible explanation for this is that 3
remains stable throughout Phase II for a large ([LH] ability condition),
but decreases in Phase II for a_small ([LH] ability condition). The
stability of a_would depend on its magnitude. This could explain why
the gain-loss model predicts a P(S) greater than observed for the [LH]
ability condition in the Berger-Camilleri experiments. For the [LH]
ability condition a_at the beginning of Phase II has the value estimated
from Phase I. As Phase II progresses 3 decreases to a value at the end

of Phase II which is much less than that estimated from Phase I. Thus

1.00
m
z 0.90
9
92 0.80
8
o 0.70
m
u_ 0.60
o
z 0.50
9
+..
0: 0.40
E
m 0.30
a.
z 0.20
35

0.10
2

0.00

 

61

 

STANDARD
HIGH SELF-CONSISTENCY

VERY HIGH SELF-CONSISTENCY

 

l-5 7-12 I449 2l-25
TRIAL BLOCKS

Figure 6. Observed Mean PrOportions of 5 Decisions by Trial Blocks by

Ability, Condition and Level of Self-Consistency (Trials 6, l3, and 20
are non-critical trials)

62
the observed mean proportion of S decisions should be less (because of
a decreasing 2) than the predicted mean proportion of S decisions
(based on a constant a). 80 not only does the effect of the utility
of self-consistency on P(S) depend on the value of a, but also the
stability of a_possibly depends on the value of a. The parameter a

and its estimation definitely needs investigating.

Interpretation of Experimental Results-~0ther Models

 

Decision-making theories such as that of Festinger (1957, p. 40)
and the maximizing theory of Simon (1957, pp. 204-205) are deterministic,
that is, given a specific utility structure, the choice of only one
alternative is predicted. Obviously deterministic theories are not
supported by the results of this experiment or other expectation experi-
ments; P(S), the mean proportion of S decisions, for any condition is
never anywhere near 0.0 or 1.0, as would be predicted by a deterministic
theory. It is the same if responses are looked at individually rather
than aggregated. Only 2 subjects out of 168 chose the same alternative
throughout the 22 critical trials of Phase II for the 6 conditions of
this experiment.

So, deterministic theories do not fit the results of this experiment.
Simon's (1957, pp. 204-205) satisficing decision-making theory can give
probabilistic results. But in the expectation experiment where there are
only two alternatives and one of the alternatives (depending on the con-
dition) always has a greater utility, the probability mechanism is the
order in which the alternatives are considered. This mechanism is not
described by Simon, and so satisficing theory is of little help here. The

probability mechanism must be central to the theory of decision-making

 

63
itself, as it is in the gain-loss theory and other probabilistic theories
such as those of Bush and Mosteller (1955) and Estes (1950 and 1954.).

The primary importance of the expectation eXperiment has been its
use as an experimental setting for the testing of gain-loss theory.
Obviously other theories can be applied to the expectation experiment
in an attempt to explain the results. Such a set of theories that could
be applied are those making probability matching predictions (e.g., Bush
and Mosteller, 1955, and Estes, 1950 and 1954). For the type of situa-I
tion where the subject makes a binary response to a set of stimuli and
each of the two responses is reinforced with its own probability, proba-
bility matching theories arrive at the prediction that the probability
with which a response is given approaches an asymptote at the probability
value with which that response is reinforced. The expectation experiment
can be viewed as providing this type of setting. In Phase II on dis-
agreeing trials the subject must choose between the response that self
is right and other is wrong and the response that other is right and self
is wrong. In Phase I, on trials where one person is right and one wrong,
if the subject is told that he is right, then this can be viewed as
reinforcing the response that self is right and other is wrong; or if the
subject is told that he is wrong, then this can be viewed as reinforcing
the response that self is wrong and other is right.

The joint distribution for right and wrong trials for both the low
and the high subjects in Phase I used in this experiment and in the
Camilleri-Berger MSU experiment are given in Table 17. One subject is
right and one is wrong on fifteen trials. Of these fifteen trials the
low subject is right on three trials. This means that the response that

self is right and other wrong is reinforced for the low subject with a

64

TABLE 17
JOINT DISTRIBUTION OF RIGHT AND WRONG TRIALS

FOR BOTH LOW.AND HIGH ABILITY SUBJECTS IN PHASE I

 

 

High Subject

 

Right Wrong

 

Right 5 3
Low Subject
Wrong 12 O

 

65
pnfljbability of 0.20 in Phase I. 0f the fifteen disagreeing trials in
Pruise I, the high subject is right on twelve trials. So the response
that:self is right and other wrong is reinforced for the high subiect
xmith a probability of 0.80 in Phase I.

Thus, probability matching theories would predict that the high
snibject would make a self right, other wrong response (self decision)
‘with.a probability of 0.80, which is quite close to the observed propor-
tions for the three conditions of this experiment (viz., 0.77, 0.80, and
0.81). For the low subject, probability matching theories would predict
a self right, other wrong response (self decision) with a probability of
0.20, which is quite different from the observed proportions (viz., 0.36,
0.50, and 0.48).

Although probability matching theories do not fit the data of this
experiment as well as the gain-loss model, this does not mean that they
should be disregarded. The probability of a self right, other wrong
response in the probability matching model sounds like the definition of
a.in the gain-loss model, viz., the subjective probability of self being
right given that self and other disagree. Both quantities are asymptotes,
as R is reduced in the gain-loss model, P(S) approaches a: And the two
quantities are computed similarly except that where the probability
matching model uses only the disagreeing of Phase I, the gain-loss model
uses all Phase I trials. Thus, the probability matching model raises

again the question of how a_should be estimated.

Implications

This examination of the gain-loss model has cleared up some things,

left some questions unanswered, and raised some more questions. The

 

66
results of this experiment indicate that the utility of self-consistency
can be experimentally increased through instructions, but the results
are not always as intended. The dynamics are yet to be understood. it
is indicated that, contrary to the assumption of the gain-loss model, the
utility of self-consistency is dependent upon the ability condition.
Something has been said about the nature of this relationship, but the
exact relationship is not known, nor has it been determined how the
utility of self-consistency should be reconceptualized in the gain-loss
model. The dependency of the gain-loss model conceptualization of the
utility of self-consistency may account for the failure of the model to
fit the data for the [LH] ability condition; but it is not known whether
other explanations are possible, nor whether other modifications to the
gain-loss model are necessary. Little has been done in determining the
adequacy of the conceptualization of a, the subjective probability of
being correct in the face of a disagreement, nor in determining the
adequacy of its estimation procedure. Perhaps a_should be estimated
from the responses in Phase II, rather than the manipulations of Phase
I. Possible flaws in the expectation experiment and their remedy have
been suggested; but while these remedies incorporated in this experiment
may have had the desirable result of reducing suspicion rates, they also
may have had the unexpected result of decreasing P(S) for the [LH]
ability condition. The effect of the remaining proposed remedies is

not known.

Summagy

Decision-making behavior in a social environment is influenced by
numerous factors, two of which are the person's desire for self-

consistency and the person's expectations for his own ability relative

67
to the ability of others. These two factors may often compete with
one another. The person may be torn between the alternatives of being
consistent and possibly wrong or having to be inconsistent in order to
jprobably be right. In the theory developed here, it is hypothesized
that the two factors of self-consistency as conceptualized in the gain-
loss model and relative ability do not influence decision-making inde-
pendently, but that the effect of each is dependent on the other.

The expectation experiment provides such a situation for study
(Camilleri and Berger, 1967). The experiment employs two-person groups
making binary choices. On each critical trial each person is forced to
make a decision whether to stay with his own choice and be consistent
but possibly wrong or to go with the other person's Opposite choice and
possibly be right but inconsistent. The expectation experiment presents
this situation in a setting where the relative abilities of the decision
maker and the other person are explicitly manipulated by the experimenter.

The gain-loss theory applies to this situation of conflict between
being consistent or being right. The gain-loss theory is based on the
utility assumptions that a gain forgone is a loss and that a loss
avoided is a gain. The expected gain of an alternative is the gains
associated with that alternative plus the losses associated with all
other alternatives. Gain-loss theory states that decision-making is a
probabilistic process, and that the probability of the choice of any
alternative is proportional to the expected gain of that alternative.

The gain-loss model of the expectation experiment postulates a
utility for being self-consistent and two utilities for being right.
The magnitude of the utilities for being right obviously is dependent

on the relative abilities of the decision maker and the other person

68
as set by the expectation experiment. However, in the gain-loss model
the utility of self-consistency is denoted ul and is assumed to be
independent of the relative ability of the decision maker. As a compar1~
son of the predictions of the gain-loss model with the experimental
results shows a failure of theory to predict the obtained values in four
of the nine cells (see Camilleri and Berger, 1967), this assumption of
independence between relative ability and utility of self-consistency is
questioned. Consequently, an expectation experiment was designed in
which the utility of self-consistency was incorporated as a treatment
factor with three levels. An analysis of the experimental results
indicates that, as is to be expected, u1 is dependent on the utility of
self-consistency; but, contrary to the gain-loss model, the results also
indicate that 111 is dependent on the relative ability of the decision
:maker. It is concluded that the utility of self-consistency must be

reconceptualized in the gain-loss model of the expectation experiment.

LIST OF REFERENCES

LIST OF REFERENCES

Balkwell, James W. "A Structural Theory of Self-Esteem Maintenance."
Sociometry, 32 (December, 1969), 458-473.

Berger, Joseph, and Conner, Thomas L. "Performance Expectations and
Behavior in Small Groups." Acta Sociologica, 12, 4 (1969),
186-198.

Berger, Joseph, and Snell, J. Laurie. "A Stochastic Theory of Self-
Other Expectations." Paper circulated from.Department of
Sociology, Stanford University, Stanford, California, 1961.

Berger, Peter L. Invitation to Sociolggy: A Humanistic Perspective.
Garden City, New York: Doubleday, 1963.

Bush, Robert R., and Mosteller, Frederick. Stochastic Models for
Learning. New York: John Wiley & Sons, 1955.

Camilleri, Santo F., and Berger, Joseph. "Decision-Making and Social
Influence: .A Medel and an Experimental Test." Sociometry,
30 (December, 1967), 365-378.

Camilleri, Santo F.; Berger, Joseph; and Olin, Wayne A. "Decision-
Making and Social Influence: A Replication of Some Experimental
Results." Forthcoming.

Estes, William K. "Toward A Statistical Theory of Learning."
Psychological Review, 57 (1950), 94-107.

Estes, William.K. "Individual Behavior in Uncertain Outcome
Situations: An Interpretation in Terms of Statistical
Association Theory." Decisions Processes. Edited by
R. M. Thrall, C. H. Coombs, and R. L. Davis. New York:
John Wiley &:Sons, 1954.

Festinger, Leon. A Theory of Cognitive Dissonance. Stanford,
California: Stanford University Press, 1957.

Gullahorn, Jehn T. "Measuring Role Conflict." American Journal
of Sociology, 61 (January, 1956), 299-303.

Gullahorn, John T., and Gullahorn, Jeanne E. Computer Simulation of
Rple Conflict. System.Development Corporation Paper No. SP-2261.
Santa Monica, Calif.: System Development Corporation, 1965.

 

69

7O

Homans, George Casper. Social Behavior: Its Elementary Forms.
New York: Harcourt, Brace, & World, 1961.

 

Olin, wayne A. "Rational Decision-Making: Three Models."
Unpublished M.A. thesis, Michigan State University,
East Lansing, Michigan, 1966.

Shelly, Robert. "Social Influence and Decision-Making: Monetary
and Non-Monetary Payoffs." Unpublished Ph.D. dissertation,
Michigan State University, East Lansing, Michigan, 1972.

Simon, Herbert A. Models of Man: Social and Rational. New Yerk:
John Wiley &:Sons, 1957.

 

APPENDIX

APPENDIX

PROCEDURES MANUAL FOR THE EXPERIMENT

Operations in the Experiment

 

Sequence of Events

 

The subjects are first met in the waiting room and then escorted
to the activity room, where they are seated in either chair and
receive the initial verbal instructions. The tape is put on for the
instructions (each tape is marked for the appropriate condition). If
the subjects have questions, answers should be rephrasings of the
instructions. During the first and second phases of the experiment, one
of the experimenters should monitor the subjects to make sure that they
do not talk to one another. Following the end of the second phase the
subjects are escorted to the interview rooms and the interview is con-

ducted. 23 no: use the terms "subject" or "experiment" in any conver-

sations with the subject.

The Eqpipment and Its Operation

 

The equipment employed in the experiment consists of the ICOM, the
slide projector and Hunter timer, the tape recorder, the event recorder,
and the buzzer. Each piece of equipment should be checked out before an
experimental session begins. This check-out includes being sure that all
switches are turned on, that the Hunter timer is set properly, and that

the slide projector's lamp is not burned out.

7I

72

Tpg_IQQM_has two switches at the top of the control panel which
must be turned on. One of these is the "Relay" and the other is
"lights." The switches marked "veridical" and "normal" are to be in
the "normal" position for the instructions and during the first and
second phase "tests." The switches marked "Agree" and "Disagree" are
to be positioned according to the schedules provided for the second
phase. The red "Relay Release" is pushed at the conclusion of each
trial, after the data have been recorded, to reset the machine for the
next trial. If this does not work, shut the machine off and on once
to release the relays. Should this fail to reset the machine (turn the
lights off), abort the session.

222.§$i§2 projector has two switches on it: one, on the left side
of the rear projection box, is the power switch; the other, on the right
side of the slide projector, is the switch for the lamp. Both must be
on for a slide to be shown; turn on the power first. Always turn off
the lamp first when turning off the projector, and allow it to cool
until the air coming from the vent at the rear of the projector no
longer feels warm (about five minutes). Turn off the lamp between
subjects.

To change a burned-out lamp remove the slide tray, unplug the
power and shutter connections to the projector, and then move the pro-

jector out of the box. Turn the prgjector over. If you don't, the

 

lenses which focus the lamp will drop out and likely break--they are
nearLy irreplacable. Remove the screw on the small trap door and
replace the bulb. Reverse the procedure to reinstall the projector.
Any session in which a lamp burns out during either of the trials
phases should be noted on the data envelope. Include the trial number

and phase.

73

The Hunter timer has a power switch, a start switch, and four

 

rotary interval switches. Turn the power switch on before the experi-
ment begins. The top rotary switch sets the interval in tens of
seconds, the second in seconds, the third in tenths of a second, and
the bottom in hundredths of a second. The first exposure is a demon-
stration slide and the duration is 12.09 seconds (top switch set to 1,
second to 2, third to O, and fourth to 9); after that all exposures are
for 2.09 seconds (top switch set at 0, second at 2, third at 0, and
fourth at 9). Turning the start switch on starts the exposure (start
light comes on, at end of interval finish light comes on). After the
finish of that exposure and before the start of the next, the start
switch must be turned off.

The control for changing slides is a grey box with a three-position
lever. Only the center position and the spring return down position are
used during the experiment. The up position opens the shutter and keeps
it open. This can be used as a check on the projector before the
experiment. With the shutter in the center position, the timer con-
trols the Opening of the shutter. In the spring return down position,
the slides are advanced. To advance the projector one slide, pull the
lever down, then let it up--don't hold it down. If the lever is held
down, the slides will keep advancing.

The tgpe recorder has an off-on switch to be turned on before the

 

session begins. The volume should be checked before each session. The
slide switch which resembles a trigger on the right side of the machine
is pulled in to stop the tape during the tape for the "live" instruc-

tions. Use the tape for each condition which i§_marked for that

 

 

condition.

71+

The Esterline Angus event recorder records the responses of the

 

subjects on a moving chart by means of ink pens. There is no main power
switch; it must be plugged in and unplugged. Be sure to unplug it wnen
it is not being used. There is a rotary switch on the left that turns
on the chart drive motor for in/min and in/hr. We use in/min. When the
motor is turned on, a light comes on; this is an indication light for

the motor and. not the main power supply. If it is plugged in, the main

 

power supply is on even though the motor light isn't lit. A rotary
switch on the right controls the speed: 12, 6, 3, 1.5, and 0.75

min
in/hr . We use 12 in/min. At the beginning of the day, the inkwell

 

must be filled and the pens primed. (See inking and priming instruc-
tions, pp. B-3 and -4 of instruction manual for portable model.)

Before each experiment and possibly during, the inkwell must be
filled. At the end of the day, the inkwell and pens must be emptied
and cleaned. (See instructions on cleaning, pp. B-11 and -lO.)

Before each experiment, the chart must be 1abelled--XG number,
ability condition by subject, experimental condition--and again at the
end of the chart run at the end of the experiment. (See pp. C-6, -7,
and -10 in Esterline Angus Instruction Book for mounting and removing
charts.)

Eight pens are used to indicate subject responses, and have these

functions:

75

Subject Initial/Final Top/Bottom
1 1 I T
2 1 I B
3 1 F T
4 1 F B
5 2 1 T
6 2 I B
7 2 A F T
8 2 F B

Manipulated other's information is not recorded. Pen nine indicates
when slide is finished until start switch is turned off.

During an experiment, the event recorder has to be watched for
several problems: inkwell running dry--use a toothpick to check and
fill when necessary; pens running dry or clogging-~(See Instructions
p. Ball) or reprime; pens sticking--jiggle.

Each piece of equipment should be checked before starting a

session.

Control Sheets

 

The following page is a copy of the control sheet for each experi-
ment. It is to be completely filled out for each subject who partici-
pates. Make every effort to get all the information. At the end of the
session, place the control sheet and data records in one of the larger
envelopes provided. Be sure to put the subject's name, the name of the
host, the condition of the experiment, and the date and time on the

outside of the envelope.

 

76

Interview Abstract

 

 

 

 

 

 

 

 

Name John Subject ___n__" ”1.
Age 19 Manipulation I+ QIControl
Major Sociology Included?

School Interview Tape No.: 00
Interviewer __Shelly» Group X01 1-".
Date May 1, 1971 Time 1:50 Seating Position 1

 

 

 

No. of Changes 00

° 65 63 it it it it it it it it it it it it it it it it it 63 it it it it

Was there prior acquaintance? No

Were any of the experimental manipulations
unsuccessful? No

Did the subject ever become suspicious? No
WOuld you include the subject in the sample? No

Did the subject change states during the
experiment? No

Did the subject give an unusual response to
any of the interview questions? No

Explanations and other things to be noted:

I {I \I \I \I
it 7: 7: 7: 7: a:

 

B Yes (explain)

B Yes (explain)
B Yes (explain)

B Yes (explain)

B Yes (explain)

B Yes (explain)

 

77

Instructions fer the Data Records

The hand-recorded responses will be collected on the data sheets
marked Phase I or Phase II for each subject. The information requested
on the form.is to be completed for each subject, and the subject's
choices recorded by circling either the T or B for Top or Bottom for
each trial. At the completion of the second phase this information is
to be placed in the envelope for the experimental session. The
following two pages are sample completed forms.

The machine-recorded responses are collected on the event recorder.

 

A line should be drawn across the paper before you start each session,
and the date, time, and condition noted on the paper directly below the
pens. The person who records the data should check the pens by turning
on the machine before the actual data collection begins. At the end of
the second phase a line should again be drawn across the page to indi-

cate the end of the session.

Experimenter's Roles

 

There are three roles, or sets of tasks to be carried out in the
experiment. The Host greets the subjects, seats them, briefly intro-
duces the study, and leaves for the first portion of the Phase I in-
structions. The Host then returns for the test part of Phase I and
records the subjects' responses on the display panel. "Slides" has
responsibility for the presentation of the instructions and the slides.
"ICOM" has the responsibility of recording the data and operating the
ICOM.

(Slides): Slides is responsible for the presentation of the
slides and instructions. He is directly concerned with the operation

of the tape recorder, the slide projector, the Hunter timer, and the

 

78

Choices for Phase I

 

 

 

 

 

 

 

Experimental Group XGl Manipulation i: -]Control
Name of Subject John Subject Seat Number 1
Date May 1L1971 Time 1:50
Trial Number m
l T B
2 T B
3 T B
4 T B
5 T B
6 T B
7 T B
8 T B
9 T B
10 T B
1.1 T B
12 T B
13 T B
14 T B
15 T B
16 T B
17 T B
18 T B
19 T B

20 T B

 

79

 

 

 

 

 

 

 

 

Phase II
Name of Subject John Subject Group Number “EC-1“” ._...-..-
Date May 1, 1971 Condition I+ -]Control
Time 1:50 Status Manipulation _
Seat Number 1 Host Shelly
Number of Changes 00 Included Yes No
Trial I F Trial I F hi
1 T T A 13 T T
B B C B B Agree
2 T T 14 T T
B B C B B C P‘
3 T T 15 T T
B B C B B C
4 T T 16 T T
B B C B B C
5 T T 17 T T
B B C B B C
A 6 T T 18 T T
B B Agree B B C
7 T T 19 T T
B B C B B C
8 T T A 20 T T
B B C B B Agree
9 T T 21 T T
B B C B B C
10 T T 22 T T
B B C B B C
11 T T 23 T T
B B C B B C
12 T T 24 T T

80

intercom. The tape for the instructions should be put on the tape
recorder and checked, so that no lag exists once the machine is turned
on for the subjects. The tapes are labelled according to condition and
the conditions should be run according to the schedule in the back of
the manual. Slides is also responsible for giving the appropriate
feedback on each trial of the first phase. (See the notebook marked
Phase I manipulation for this schedule.)

(ICOM): ICOM is responsible for the operation of the ICOM, the
event recorder, and the buzzer in the first phase. He is responsible

for recording the data and resetting the ICOM after each trial.

8]

A General Outline of the Experiment

 

Slides' Job

Recorder's Job

 

Tape recording of in-
structions started

Demonstration slide
presented

Test of Spatial Judgment
Ability begun

Feedback to subjects
after each trial

Subjects told final tally
of right-wrong by Host

Phase II begun

Demonstration slide for
Phase II

Phase II trials
administered

#3me

ICOM set in "Normal" mode
Buzzer cues for slides
Subjects respond

Data recorded

Note on Esterline-Angus
end of Phase I

Buzzer cues for slides
Data recorded for Phase II

 

82

ICOM Job
Phase I

Job Description

Comments

 

Beginning of tape instructions

1. Set ICOM st "Normal"

1 . Sound buzzer for Demo
slide

2. Show Demo slide #1 when
indicated by tapes. Demo
slide choices made by
subjects

3. Test part of Phase I

a. Buzzer sounded

b. Subjects respond

6. Responses recorded

d. Displays cleared by
pushing "Relay
Release" on ICOM
panel

4. At conclusion of test
phase set ICOM to
"veridical" for Phase II
demonstration

During first phase instruc-
tions ICOM operator has to
monitor instructions for
occurrence of Demonstration
slides

 

Turn on event recorder

ICOM operator must monitor
the event recorder and co-
ordinate with the actions Of
the Slide Operator

Turn off event recorder at
end of test part of Phase I

 

Job Description

83

ICOM Job
Phase II

Comments

 

1.

Demonstration slide

a. Sound buzzer

b. Subjects respond on
initial choice

0. Sound buzzer for
second presentation

d. Subjects respond

Set ICOM to "Normal"

Test part of Phase II

a. Set switches for
agree/disagree

b. Sound buzzer

c. Subjects respond on
initial choice

d. Record data

e. Sound buzzer for
second presentation

f. Subjects respond on
final choice

g. Record data

h. Push "Relay Release"

Don't fall asleep

Listen to instructions for
occurrence Of Demonstration
slide

Coordinate actions with
"Slides" and monitor event
recorder

 

 

Job Description

84

Slides Job
Phase I

Comments

 

Start tape after Host has
entered Observation room

Present Demonstration
slide after buzzer sounds

Test part Of Phase I

a. Present slide after
buzzer sounds

b. Give who correct/
incorrect after sub-
jects respond

0. Change slide after
who correct/incorrect.
Reset timer.

d. Repeat 20 times.

Listen to tape for cues for
live instructions

Coordinate actions with ICOM
operator. Be sure you press
lever on intercom when you
talk to subjects

Host gives results and
signals start of Phase II

 

Job Description

85

Slides Job
Phase II

Comments

 

Start tape for Phase II
after Host has changed
display boards and
returned to Observation
room

Present Demonstration
slide after buzzer sounds
for both initial and final
choice

Present slides for test

part of Phase II

a. Present slide after
buzzer for initial
choice

b. Reset timer

0. Present slide after
buzzer for final
choice

d. Change slide after
timer shuts off.
Reset timer

e. Repeat 25 times

Listen to instructions for
cues for live parts Of
instructions

Don't fall asleep

 

 

Phase I

86

Instructions

 

(Observe waiting room. Try to enter just as second subject enters.

Have subjects leave coats, books, etc., in waiting room. Begin by

saying:)

(Host)

(tape)

Let me introduce myself. I'm Mr. .
Take either chair. The instructions for this study will be
given by recording over this speaker. By means of this
intercom we can hear you and talk to you. SO be sure to
tell us if you have any questions or there is a mechanical
failure such as the sound being too low, a slide not pro-
jecting properly, and so on. Otherwise, please do not talk
during the study. As you can see, there is a number on the
machine in front of you, Number 1 or Number 2. During the
rest of the study each of you will be referred to by that
number. We will start the tape now.

First, let me ask you not to push any Of the buttons on the
panel until I give you instructions for their use. Feel

free to smoke if you wish.

(Host leaves study room)

(Slides:

(Host)

Start tape)

I'd like to thank you for being able to join us today. We
think you'll find this to be an interesting as well as a

rewarding experience.

We are members of a research team of social scientists who
are interested in studying the way in which individuals and

groups solve certain types of problems. Furthermore, we are

87
interested in studying the ways these problems are solved
in different kinds of situations. Our work will be divided
into two phases or parts, Phase I and Phase II. In each of
these phases you will be asked to solve problems but under

different conditions. I will explain the nature of these

 

problems and conditions as we go along.
Let us now turn to Phase I of our work.

Within the last few years social scientists have found in
their studies that individuals differ in their ability to

accurately perceive the spatial relationships between fig-

 

ures. More simply, it has been found that when some indivi-
duals are presented with a set of figures, they are able to
make accurate judgments about how those figures are placed
in relation to one another. Other people do not seem to

have this ability to the same extent. This ability'Ip make

 

accurate jpdgments about §patial relationships social

scientists call Spatial Judgment Ability.

At this time we frankly do not know much about why some

peOple have this ability more than others.

One thing we do know about this pgrceptual ability is that

 

it is not necessarily related to other specialized skills
that a person might possess. This means that people with
high mathematical or artistic skills, for example, d2 not

necessarily have high Spatial Judgment Ability.

 

 

88
Because of the importance of this Spatial Judgment Ability,
social scientists are engaged in an extensive set of studies
to examine this ability among college students such as

yourselves here and elsewhere.

What we are going to do today is to give you an especially
prepared test which is extremely accurate in measuring an
individual's Spatial Judgment Ability. That is, this test
distinguishes those who have a great deal of this ability

from those who do not.

The test consists of a series of pairs of rectangular
figures. Each rectangle is composed of smaller black and

white figures.

We will proceed as follows: I will present to you on the
screen above the speaker a slide containing one pair of the
rectangular figures.

(Slides: Put Demonstration slide #1 on screen for 12 seconds)

In each pair, one figure has more small white rectangles

 

than the other. That is, the color white will cover more
of the area Of one rectangle in the pair than it does of the
other. Your task is to determine which of the two rectangles
is more white in area. I will present a slide such as this
for two seconds for you to study.

(Slides: Rgpgpt Demonstration slide #1 for 2 seconds. Continue when

slide is through. Change slide)

During the actual test, you will indicate which rectangle

you think is correct by pressing the button labeled either

 

89
"tOp" or "bottom" below the statement on your board which

reads "Final Choice." That is the bottom row of buttons

on your board.

YOur decision will then be registered on my board. Face
decision you make constitutes one round or trial. In this

phase, you will be asked to make twenty such decisions.

(Emphasize: ) h
These decisions which you make will enable us to measure

your Spatial Judgment Ability.

 

After both of you have made your decisions as to which of
the two rectangles is more predominantly white in area, I
will announce and your host will record whether each of your
answers is correct or not. That is, whether or not white
does in fact cover more Of the area of the rectangle you
selected than it does of the other rectangle in the pair.

In this way, you'll be able to tell how well you're doing as
you go along. At the end of the test, your host will tally

up your scores.

‘ You may find that some of the slides will seem difficult to
judge as the difference between the two rectangles in the
area covered by white is sometimes quite small. It was
found in previous studies that some individuals are able to
make correct judgments on the basis of very slight, almost
intuitive, cues and feelings. In general, we have found

that people with high Spatial Judgment Ability consistently

 

90
make correct decisions and those with low Spatial Judgment

Ability usually make incorrect decisions.

SO that you have some idea of how well you might do, we have
put on the Chart standards based upon previous stubies of

college students here and elsewhere.

This test has been administered to college students of yuur

level in this part of the country and elsewhere. The
standards are based on those studies. As you can see, a
score of

13 - 16 correct out of 20 is a good score;

 

l7 - 20 correct out of 20 is a rare Occurrence
and represents a superior individual
performance;
9 - 12 correct is a poor score; and
O - 8 right is also quite a rare occurrence and

represents a very poor individual performance.

In general, the characteristics of this test are that peOple
with high ability will usually score in the good or superior
category. Likewise, people with low ability will score in
the poor or very poor category. You can also see that
although a person might get nine or ten out of twenty
correct by merely guessing, this is a poor score. We find
that most people who guess score about the same as people

with low ability.

Before beginning the test, we'll go through a practice

(Recorder:

(Slides:

9l
slide so that you can get familiar with the procedure we

are going to use in this first phase of our work.

At the beginning Of each trial I will sound a buzzer and
then present a slide containing two of the rectangular
figures. You will see it for two seconds. As soon as the
slide goes Off, you are to choose which rectangle, either
top or bottom, is more white in area. As soon as the slide
goes Off, you are to press the appropriate button under tie

words "Final Choice" on your panel.

Now this trial will not count on your Spatial Judgment

score. It is just for practice.

I will first announce the presentation Of the slide. The
next slide is Demonstration slide Number 2.

Sound buzzer)

Put on Demonstration slide #2)

(Tape: Allow 2 seconds after slide goes off)

(Slides:

(Recorder:

Change slide)

All right. You should have made your choice by now. When
you've pressed your button, your decision is registered on
your own board and on my board. During the test I will
announce whether each of your decisions is right or wrong
and your Host will record the result on the board. When
this has been recorded, a button will be pressed.

Press "Relay Release")

92
The lights on the boards will go off and we will be ready
for the next trial; we will repeat this procedure for the

20 slides in this phase.

Let me summarize these important points before we begin.

1. Each decision constitutes one round or trial.

2. In this phase of our work, you will be asked to make
twenty such decisions.

3. You will have two seconds to judge the slide. Please
make your response as soon as possible after the slide

has been taken Off the screen.

(Pause)
During this phase you should not in any way communicate

with one another.

(Pause)
1122.5. 1_s_ pp: if; E 31313. I suggest you study the
rectangles carefully.

(Host re-enters)

(Host) Is everything clear? Number 1? Number 2? Okay. We will
begin with slide Number 1.

(Recorder: Sound buzzer)

(Slides: Display slide)

(Slides via intercom:)
Number 1, you are right/wrong. Number 2, you are
right/wrong.

(Recorder: Push "Relay Release")

(Slides: Change slide, repeat for all 20 slides)

 

 

93
(Wait)
(Host) As you can see, for the entire test Number 1 got 9/18
correct and 11/2 wrong. 'Number 2 got 9/18 correct and

11/2 wrong.

As you can see, we have had two unusual performances today.
Number 1 is in the Superior/Poor category and Number 2 is

in the Superior/Poor category.
If everything is clear . . .

(Pause)
. . we'll go on.
(Host: Take down Phase I standards and Phase I scoreboard. Give

subjects Phase II standards; leave study room.)

Phase II
We are now ready to turn to the second part of our work.
In this phase we are going to ask you to work together
under a different set of conditions. In this situation we
are interested in seeing how well you can work together as
a tggp. We are going to allow you to exchange information
with each other on what you think is the correct choice
before you make your final decision. That is, we are going
first to allow each of you to make a preliminary choice
which will be communicated to the other person. Then,
after a short period, you will be asked to make a fipal

choice between the two rectangles. We will be concerned with

 

 

9h
your team getting as many correct final choices as you

possibly can.

After I present a slide, each of you is to make an initial
ghpigg as to which you think is the correct answer--the top
or the bottom.figure. This is for the purpose of letting
the other person know what ypthhink is the correct choice.
After ERIE of you make your initial choice, you wiLl
receive information on your board as to what the other
person thinks is the correct answer. OnLy after'you see
the other person's initial choice will I repeat the slide
for you to make your final choice. Shortly thereafter I

will clear your boards.

You are to indicate your initial choice by pressing the
appropriate button immediately below the words Initial
Choice. This is the top row of buttons on your’panel.

Once you make your initial choice, this choice will be com-
municated to the other person and you will be able to see
the other person's initial choice on the panel marked "His
Choice." That is, the bulb:marked "top" or "bottom"

corresponding to the other person's choice will light up.

However, this i§_important. You will not receive information

 

on the other person's initial choice until you have made
221.13: _o_w_n initial choice. After you both make your initial
choice, the slide will be presented again for 2 seconds.
Immediately after the second presentation, you are to indi-
cate your final choice by pressing the button marked "Final

Choice."

 

95
(Pause)
Remember only your final decisions are scored in this

phase of our work.

(Pause)

Let's try this out. I will present a third demonstration
slide so that you can practice with this procedure. This
will p23 count. It is just for the purpose of becoming

more familiar with the procedure.

Make your choice immediately after the slide has been
presented.

(Recorder: Press buzzer)

(Slides: Present Demonstration slide Number 3)

(Wait 2 seconds after slide)
All right. Now you should have made your initial choice.

(Recorder: Press buzzer)

(Slides: Rppgp£_Demonstration slide Number 3)

(Wait 2 seconds after slide)

(Tape) All right. Now you should have made your final choice.

(Recorder: Push "Relay Release")

(Slides: Change slide)
During the actual test, you will be presented with 25
slides. The procedure for all of them will be as was
demonstrated. We will pp: tell you after each slide which
is the correct answer. we will record for each slide

whether your final choices are correct or incorrect, and at

 

the end Of the test we will tell you how many correct and

incorrect final choices the team made.

 

 

 

 

96
ifli§.i§ important. We are pplggy interested in your’making
as many correct IEEEL choices as you possihky can. The only
answer that is recorded is your fippl.choice, and you should
not hesitate for any reason to change your initial choice in
order to make a correct final choice. Let me repeat. Try

to make as many correct final choices as you can. DO not

  

worry whether your initial choice and final choice are the

 

A...

same. Let me caution you, however, to make your initial
choices with care so as to provide your partner with the

best information you can.

 

F”;

I have already mentioned that in this phase Of our work we
are interested in Eppp performance; that is, in how well
two people working together can do a spatial judgment task.
We have found from previous studies that the most efficient
way for two people to work together on this type Of task is
to give each member of the team 3933.1. regponsibility for
making the final choices for the team. Regardless of
whether'your scores in the first test were alike or dif-
ferent, it is our standard practice to give each of you

equal responsibility for the team.score.

In our work today each member of the team will have egual

responsibility for the team's score.

 

In this phase we are interested primarily in seeing how well
you can work together as a team. Therefore, we are allowing
you to exchange Opinions with one another as to what you

think is the correct answer before you make your final

97

choice. Only your final decision will count on your team's

spatial judgment score.

Let me explain how we score final decisions in this phase.

Since you have equal responsibility) each time a person

 

makes the correct final decision, your team will get one
point. If an individual makes an incorrect final choice,
then his final choice adds nothing to the team's score on
that trial. Since there are 25 trials in this phase, this
means that each person's maximum.contribution to your teamIs
spatial judgment score is 25 points and his minimum.contri-
bution zero. This means that the maximum score that the
team.can achieve is 50 and the minimum.is zero. Each of you

has an equal Opportunity to contribute to the teamis score.

On the sheet Of paper handed to you is a table Of team
standards for this situation. These standards are also
based on previous studies that have been done with college

students like yourselves here and elsewhere.

What we find for this set of conditions-that is, where you
can exchange preliminary opinions before making your final
choice, and where each of you has equal responsibility for
the teamls score-~is that:
31 - 40 is a good team score;
41 - 50 out Of a possible 50 is a rare occurrence
and clearly constitutes a superior team

performance;

 

 

 

98
21 - 30 is a poor team.score; and
O - 20, which doesn't Often occur, would clearly

constitute a very poor team performance.

Remember: We are interested in seeing how well you can
work together as a Egpp.
(Manipulations)

Before we begin, let me summarize several important points.

1. Each final choice you make constitutes one round or
trial.

2. In this phase, you'll be asked to make 25 final choices
in all.

3. Since you have pgppl_responsibility for the team’s
score, for every correct final decision each of you
makes the team will get one point. If either Of you
makes an incorrect final choice, then his final choice
contributes no points to the teamis score on that trial.
This means that each Of you has an equal opportunity to
contribute to the team's score.

(Turn Off tape)

(Slides via intercom)
Is everything clear? Number 1? Number 2? Okay. We will
begin with slide Number 1.

(Recorder: Push buzzer)

(Slides: Put slide on screen)

(Recorder: Record responses)

(Recorder: Push buzzer)

(Slides: Repeat slide)

 

 

 

99

(Recorder: Record reSponses and push I'Relay Release“)

(Recorder and Slides: Repeat for all 25 slides)

(Host: Enter study room).
The study is now completed. Before we discuss your scores,
we would like to talk to each of you individually to get a
further elaboration of your feelings and opinions about the

study. Mr. will Speak to you,

 

Number 1; and I will Speak to you, Number 2.

 

lOO

Manipulation for c High and c Very High Conditions

for c High Condition Only

As we all know, some peOple just go along with the group without
much thought. They tend to be conformers in many situations. This
can happen here. For example, if you and your partner do not make
the same initial choice--you disagree--you might be tempted to conform
by going along with your partner, by giving his initial choice as
your own final choice whether he is right or not. Of course, feel
free to change your choice when you feel his answer is right. Avoid

conforming by thinking carefully about each choice.

For c Very High Condition Only

As we know, some people just go along with the group without much
thought. They tend to be lazy or to have serious personality problems
and tend to conform in many situations. This can happen here. For
example, if you and your partner do not make the same initial choice--
you disagree--you might be tempted to conform by going along with your
partner by giving his initial choice as your own final choice whether
he is right or not. If you tend to be the kind of person who conforms,
try to overcome this. Of course, feel free to change your choice when

you feel his answer is right.

Remember:
Think carefully about each choice.

Do not conform just for the sake of conforming.

«r. d. "J: y

 

lOl
Completion of the Experiments

 

The following pages contain a listing of the order in which the
experimental sessions were completed. In the columns headed
"Included in Analysis" the following code is used: Y means the
subject is included, N means the subject is not included, and C

means that a confederate participated in that position.

 

102

ORDER IN WHICH EXPERIMENTS WERE COMPLETED

 

 

 

Included

Group Ability Condition Level of in Analysis
Number No. 1 No. 2 Self-Consistency HL LH
1 LH HL Very High N Y
2 LH HL High N N
3 HL LH Standard Y Y
4 HL LH High Y N
5 LH HL Very High N Y
6 HL LH Standard Y Y
7 LH HL High Y Y
8 LH HL Standard Y Y
9 HL LH High N Y
10 LH HL Very High Y Y
11 HL LH Very High N Y
12 LH HL High Y Y
13 HL LH Very High N N
14 HL LH High N N
15 LH HL Standard Y Y
16 LH HL High N Y
17 HL LH Standard Y Y
18 LH HL Standard Y Y
19 LH HL Very High Y N
20 HL LH Standard Y Y
21 LH HL Standard N Y
22 HL LH Very High N Y
23 HL LH High Y Y
24 HL LH very High Y Y
25 LH HL Very High Y Y
26 HL LH High Y N
27 LH HL Very High N Y
28 LH HL Standard Y Y
29 LH HL High Y N
30 HL LH Standard Y Y
31 HL LH Very High Y Y
32 HL LH High Y Y
33 LH HL High' N Y
34 LB HL High Y Y
35 HL LH Standard Y Y
36 LH HL Standard Y Y

\s. '4!“ ‘ .t'. . ”J -

 

l03

 

. Included
Group Ability Condition Level of in Analysis
Number No. 1 No. 2 Self-Consistency HL LH
37 HL LH Very High Y Y
38 LH HL Very High N Y
39 LH HL Very High Y Y
40 LH HL High Y N
41 BL LH Standard Y Y
42 HL LH High Y Y
43 LH HL Very High Y N
44 HL LH Standard N N
45 LH HL High N N
46 LH HL High Y ‘Y
47 HL LH High Y Y
48 LH HL Very High Y Y
49 HL LH Very High N N
50 LB HL - High Y Y
51 HL LH Very High N N
52 HL LH High Y Y
53 HL LH Very High Y Y
54 LH HL Standard Y Y
55 LH HL High Y Y
56 HL LH Very High Y Y
57 LH HL Standard Y Y
58 LH HL Very High Y N
59 HL LH High N Y
60 LH HL Very High Y Y
61 HL LH Very High N Y
62 HL LH High Y N
63 HL LH Very High Y Y
64 LB HL Very High Y Y
65 HL LH High N Y
66 HL LH Very High Y N
67 LH HL Very High Y N
68 LH HL High Y Y
69 HL LH Standard Y Y
70 HL LH Standard Y N
71 HL LH High Y Y
72 LB HL High Y Y

 

10h

 

Included
Group Ability Condition Level of in Analysis
Number No. 1 No. 2 Self-Consistency HL LH
73 LH HL High N Y
74 BL LH Standard N Y
75 LH HL Standard N Y
76 HL LH Very High Y Y
77 LH HL Very High ‘Y Y
78 LH HL High Y Y
79 LH HL High ‘Y Y
80 HL LH Standard Y Y
81 HL LH High Y Y
82 LH HL Very High Y N
83 HL LH Standard Y Y
84 LH HL High Y Y
85 LH HL Standard Y Y
86 HL LH High Y N
87 LH HL Very High Y N
88 HL LH Very High Y Y
89 LH HL High Y Y
90 HL LH Very High Y ‘Y
91 HL LH High ‘Y Y
92 LH HL Standard Y Y
93 LH HL High Y NY
94 HL LH Standard Y Y
95 LH HL Standard Y' Y
96 LH HL Very High C Y
97 HL LH Standard Y ‘Y
98 LH HL Standard Y Y
99 HL LH Very High Y Y
100 HL LH High Y N
101 HL LH Very High ‘Y Y
102 LH HL Very High Y Y
103 HL LH Standard Y C
104 LH HL Standard N C
105 LH HL Very High C Y
106 LH HL Standard Y C
107 LH HL Standard N C
108 HL LH Standard Y C
109 EL LH Standard Y C
110 HL LH Standard Y Y
111 LH HL Very High Y C
112 LH HL High N Y

 

l05

Ehe Stimulus

 

The stimuli in the experiment consist of a series of slides which
contain two rectangular figures. Each figure is subdivided into a grid
with approximately half of the grid colored black, the remainder white.
Figure Al contains a sample slide. The slides have been pre-tested,
and only those with a choice structure of 40-60 per cent are employed

in the study.

 

l06

 

 

 

 

 

 

 

 

Figure A-I. Sample Slide

 

107

 

Trial Number Ability Expectation

J+-J l-fl

1 Right Right
2 Right wrong
3 Right Wrong
4 wrong Right
5 Right wrong
6 Right Right
7 Right wrong
8 Wrong Right
9 Right Right
10 Right wrong
11 Right wrong
12 Right Right
13 Right wrong
14 Right Wrong
15 wrong Right
16 Right Wrong
17 Right Right
18 Right wrong
19 Right Wrong
20 Right wrong

 

Figure A2.--Schedule of Right-wrong for Phase I

 

 

l08
Interview Schedule

Before we discuss the results of the study, I'd like to get your

reactions to it. There are a number of things which affect the results,

and I want to talk with you about some of them, First, your name is

? And what is your major field of

 

study,

 

(first name) ? And your age?

I. Phase I

1.

In general, what are your feelings about the study?

(Just to get him talking and to determine very suspicious
subjects)

a. Have you ever participated in a study like this one befOre?

(If yes, probe for its description and why it was like
this one)

b. Have you ever read or heard about a study like this one?
(If yes, probe as in 2.a)

a. When the task was first described to you at the beginning
of the first test, how well did you expect to do on it?
Why?

b. At that same time, how well did.you think the other person
would do? Why is that?

Do you know the other person at all?

(If YES: (If NO:

Find out as much about prior Find out what impressions
acquaintance as possible and the subject got before the
probe for its effects on test and what effect these
subject's opinions of his impressions had on his

own ability relative to the impressions of the ability
other persons' ability) of the other subjects)

a. How did you go about trying to get the correct choice in
the first test?

b. In general, how confident were you of your choices on the
first test? Why (not)?

Do you think the results of the first test were a good measure
of your Spatial Judgment Ability? 'Why (not)?

“AI“?

 

II.

I09

Now let's talk about the second test.

1.

a.

b.

After the second test was described--before you began
taking it--how well did you as an individual expect to
do on it? Why?

At that same time, how well did you expect the other
person to do? Why?

How well do you think your team did on the second test? '

Why?

How confident were you of your own final decisions on

the second test? Why?

Do you feel that your own ability changed as the second
test went along? How? ‘Why?

Do you think the ability of the other person changed
during the second test? How? Why?

Let's look at your initial choices in the second test.

a.

C.

d.

Do you happen to remember how many times you agreed and
disagreed with your partner on your initial choices?

What did you think and feel when.you fOund your'partner
disagreed with you?

Do you have any ideas why you were disagreeing with him?

Did you begin to feel that someone was usually right or
usually wrong? Who? Why?

Looking back now, is there anything you could have done
differently during the second test that would have improved
your team's score? What? Why?

 

 

 

116

Debriefing
Now, , I would like to briefly explain.what we

 

were trying to study in today's tests. We are studying the relation-
ships between a person's ability, the responsibility he has to a group,
and the decisions he makes. That is, we are studying what effect there
is upon a person's changing his decisions if another person with more,
less, or equal ability disagrees with him on that decision. We are also
studying what effect there is upon changing his decision if that person
has more, less, or equal responsibility to the team.than the other
persons. 80, as you can see, we were not interested in testing spatial
judgment ability as such. Have I made sense so far?

To set up this type of situation, there were two things we needed
to arrange; your ability, and your agreement with the other two persons.
Concerning your ability, all of the slides that we showed you had the
same amount of black and white area. That is, each rectangle in every
pair was fifty per cent black and fifty per cent white. By telling you
in Phase I that you got a high (low) proportion of them correct and that
the other persons had a high (low) number correct, we hoped you would
naturally assume that you had more (less) of this ability than the other
persons. But really your high (low) spatial judgment ability and the
other persons' high (low) spatial judgment ability were fictions. I'm
not even sure there is such a thing as spatial judgment ability. Does
this make sense?

Since in Phase II we are interested in the situation where you
disagree with the other persons on your initial decisions, the informa-

tion you received about their initial decisions was controlled so that

 

Hi
you would disagree most of the time. Actually you probably agreed
with them about half the time. Is this clear?

So that, briefly, is what we were testing, the things that we
arranged, and the reasons why we had to arrange them. Since there are
some fictions involved in these tests, we would appreciate it if you
didn't tell anybody about it. They might be tested later, and such
information might bias their performance and ruin our results. Can I
have your word that you won't disclose this information? If anybody
asks you about the test, it's all right to tell them that it was a
spatial judgment test concerning whether there were more black or white
squares on some pictures; but don't tell them about the rest. Okay?

Thank you very much.

 

112

Reasons for EliminatinggSubjects from.the Sample

Deliberately making wrong initial choices.

Misunderstanding instructions.

Prior acquaintance with other subject which interferes

with process (change of expectation manipulation, friend-

ship determining acceptance of influence, etc.).

Status differences based on physical characteristics which

interfere with process.

Suspicion:

a.

volunteered infOrmation that exchange of infbrmation
was "rigged" (Phase I or II).

Read previously about deception experiments and thought
present study was similar.

Heard from.ethers that there was deception in present
study.

Previous participation in deception study and belief

that present study was similar.

D‘s

 

H3

Recruitment Presentation

I'm from the Sociology Department. We
are currently carrying out a series of studies of how individuals and
groups solve problems. We need individuals to help us out by partici-
pating in one of these studies.

Participation would involve going to Berkey Hall for one time only,
for about anDhour, and we will pay you at least $2.00 for the time you
are there. The studies will be conducted mornings and afternoons
throughout this term, so I'm sure there would be some time when you
could come. The problems you would be asked to solve are E difficult,
and they do M involve mathematics of any form.

These studies are at connected with this course. That is,
whether you do or don't participate in a study will 92$ affect your
grade in this course. So, participation is on a voluntary basis, but
we would appreciate your participation. And, in addition to the $2.00
or more which you will be paid, I think you'll find the emerience
itself to be interesting as well as rewarding.

So, if you feel you might be able to help us, would you please
fill out one of the short forms we're going to hand out. Please fill
out the form unless you are absolutely sure you would not be interested.
It only obligates you to a telephone call from us and, if at that time
you're not interested or you're busy, you can turn us down then. Also,
since we try to match the participants in each study, we may not get
around to everyone who volunteers, but we'll try to call as many of you
as possible. So, if you think you might be interested, will you fill
out one of these forms.

Is everything clear?

 

”lllllliiilm jllllllfllglll '

ll