OBJECTIVE RISK LEVELS AND SUBJECTIVE RISK PERCEPTION

Siegfried Streufert
1"40 Purdue University

and

Eugene A. Taylor
Captain, United States Air Force

TECHNICAL REPORT NO. 40

August, 1971

Dr) 0

;tp d- by

NATIONAL TECHNICAL
INFORMATION SERVICE
Springfield, Va 22151

Prepared in connection with research under the Office of Naval Research,

Group Psychology Programs, Contract No. N00014-67-A-0226-0007, NR 177-911.

Reproduction, translation, publication and disposal in whole or in part

by or for the United States Government is permitted.

This document has been approved for public release and sale; its distri-
bution is unlimited.

161
ISULAIMLI NOTICE
00

THIS DOCUMENT IS BEST

QUALITY AVAILABLE. THE COPY
FURNISHED TO DTIC CONTAINED
A SIGNIFICANT. NUMBER OF
PAGES WHICH DO NOT
REPRODUCE LEGIBLY.
 ( 4-cunt3y 6a s siftc o1 '00oftitle. bod y.ofDOCUMENT
ab stra ct* mod
CONTROL.DATA.-R & 0
Indexin g annotation miusebe entered when the overall rep ortI. cla ssified )

I. ORIGINATING ACTIVITY fCorportifO euthwl) 20. REPORT SECURITY CLASSIFICATION

PudeUniversity 2b. GROUP

IREPOnT TITLE

OBJECTIVE RT1K LEVELS AN" SUBJECTIVE RISK PERCEPTION

1. ESCRI;Ta&Vý NOTE9 (Tyrpo .eof itt anlncluttro date.s)

Technical Report NO. 4UO, August 1971
SAU lNOR(SI (First na.,n.. middle intrlial, lost nae).

Siegfried Streufert and Eugene A. Taylor

70. TOTAL NC. Or PAGES lb. NC,. EF

'S
ROfTO
August 1971

E
26 1
Ja. CONTRACT on GRANT NO. SO- ORIGINATOR'S REPORT NUMSEnIS1

00014-67-4-0226-0007. 14R 177-911

5. P"(3J4CT No. No. 40

c. 9b. 0CIHER REPORT NOIS) (Any other n,,mlbee that my he iR.alg~ned

i.QaSTA,1UTION bATCMOLET besppodfr laeadsle-

This Jouei a eaapo~ o ublic aseansl;

lg.,LEUZ4TAR 3"LSI
".'PNSORNG MILITAPY 0fTVT

Group Psychology Programs,

IOffice of Naval Research
ACT

-relationishIp beL%.OZI objective rick levels calculated on the basis of decisiors

mad- by stibjects in a simulated vat 3nd their subjtective risk perceptions xias in-
.- o'. ted. rere informed that greater or lesser amounts of
t%hadtoobe
±'4fa--tiosi -;oulV reach them, and that gr~ater or lesser qLiantities of their
dzccicions v:ould rcach the destination iL z.iich the de~cisions were addressed. In
r:-r-lity subjects played an experimenter-deter-mined program. No direct relationship
heqtt.een objectivs risk lovels and subjective risk perception was found. Perceived F
(cubjective) rirk levelsi decreased over time, while objective riJck levels did not
d-irecrne. Persons assigned to a "rebel" team took greater risks (and perceived
-r-:-r risk tr~in-) than persons assignad to a "large foreign power" team. Risk
trkn t'as not related in any way to experimenter statements about information flow.

PA 2 It~
DLD I NOWCA.1473 (PG~Unclassified
S/N 0101-807-6801 semirity Classification
3ND PPS* 1315.
 m 0 U

-.- .- ý

Lm ~ ~ ~ Sno m ~ - r LM U-o
aC $. . Vo
 OBJECTIVE RISK LEVELS AND SUIJECTIVE RISK PERCEPTION

Siegfried Streufert and Captain Eugene A. Taylor

INTRODUCTION

People make decisions every day of their lives. Many of these decisions
contain certain levels of risk. Drivinp a car, walking across a busy street,
investing money, or spending time and energy in hopes of achieving beneficial

goals all may contain elements of risk: the outcomes are uncertain, and there
is the possibility of injury or loss.
Inherent in decisions of this type are probabilities of success or failure.
If all information about a type of decision (e.g., crossing busy streets) were
known, a probability of success or failure could be assigned. Unless the ob-
jective probabilities of any decision are given and understood, and in most
complex decisions they are not, the individual may perceive and utilize some
subjective estimate of risk. In most cases this process probably does not occur
through careful weighing of probabilities and utilities. However, an individual
should be able to indicate (with some degree of accuracy) how dangerous an act
might be, i.e., how much relative risk may be involved. When his subjective
perception of risk agrees with objective probabilities, a decision maker should
be able to make optimal decisions. For instance, the closer a person's estimate
of risk involved in crossing a busy street reflects the objective measure of
accident statistics, the better his judgment about whether and when to cross

the street should be.

Definition of Risk
The concept of risk as it is utilized in this paper is based on the propo-
sition that risk occurs in decisions that involve the possibility of loss
(Kogan and Wallach, 1967; Pruitt, 1961). Game theory formulations which differ-
entiate between decisions involving risk (probabilities of all alternatives are
known and given) and decisions under uncertainty (probabilities are not known)
are not considered in this context (cf. Luce and Raiffa, 1957). Following the
work of Edwards (1955), Coombs and Pruitt (1960), Pruitt (1962), Lichtenstein
(1965), Kogan and Wallach (1967), and Van der Meer (1963), decision making
involving risk is here seen as a process in which an individual or a group

maximizes a combination of subjective probability and utility where the

 -2-

alternative exhibiting more variance (higher potential gain or loss) is seen

as more risky (assuming that an unfavorable outcome of any decision is possible).

Measurement of Risk
k number of different measures that were designed to tap individual differ-
ences in risk taking propensity have been proposed. Slovic (1962, 1964) and
Kogan and Wallach (1964) have shown that the relationships between the various
measures are low or nonexistent. Attempts at relating other personality mea-
sures as need for success and fear of failure (Atkinson and Feather, 1966) have
had little more success. Apparently people who are .isky in one situation or
on one measure may not be risky if tested in a different setting or on a differ-
ent test.
One possible solution to this problem is to ask people how risky they
think they are. The obtained scores may then be compared to objective risk
measures. The present study uses concepts derived from Pruitt's (1962) PLR
model (cf. also Kogan and Wallach, 1967). A complex decision-making environment
is selected as the research environment to permit utilizing the concept of the
disproportionate negative utility of large losses in this concept of risk.
Subjective estimates of risk are compared to increasing variance patterns in
objective risk.

Objective Hleasures of Risk

Slovic (1962) gave subjects several tests and questionnaires covering dif-
ferent categories of risk taking in an extensive validation study. Subjects
were given tests measuring category width, speed versus accuracy, and gambling
response-set, the Williams Job Preference Inventory, Torrence and Ziller's
Risk Scale, a vocabulary test in which subjects adjusted Lhe number of points
they were willing to risk winning or losing on each item (thus controlling
variance), and a test designed to assess probability and variance preferences.
In addition, subjects were asked to peer-rate fellow fraternity brothers on a
bipolar trait of general willingness to take risks.
Twenty-eight correlations were obtained among measures presumed to be posi-
tively related. Of these correlations, only five were significant. Slovic
(1962) concluded that either very few of the tests measured risk taking, or
that a generalized risk-taking propensity does not exist.
 In a more exhaustive review and analysis of validity of risk taking measures,
Slovic (1964) examined numerous risk measures under categories of gambling set,
inclusiveness set, speed versus accuracy, judgmental measures, questionnaire
measures, and probability and variance preference measures. Slovic (1964) con-
eluded (p. 227):
"There are enough positive results to encourage the investi-
gators to further pursue the notion of risk taking propensity as
a general disposition.... However, a large amount of evidence
bearing on the convergent validity of these methods is negative."

In discussing the negative evidence, Slovic stated (p. 228):

"Observation of choices made among acts entailing various

degrees of objective risk cannot distinguish the extent to which
risk taking is determined by individual differences in perception
of risk or by differences in reaction to that perceived risk.
Perceptual differences might take the form of discrepancies among
persons' subjective probabilities and subjective values (utilities)
and differences in the manner that perceived values of these com-
ponent dimensions are weighted and combined to produce a resultant
evaluation of risk.... Researchers may thus be forced to scale
subjective probabilities and utilities in order to allow any
orderliness inherent in risk taking behavior maximum opportunity
to emerge."

In another survey Kogan and Uallach (1964) sampled a diverse array of task
and situational contexts within which risk taking might be studied, to see whether
an individual who is a risk taker as measured by one procedure would turn out to
be a risk taker when measured by all the procedures. They found relatively
little generality across the diverse range of risk-taking tasks represented.
In studying achievement activation, Atkinson (1960) has utilized level of
aspiration or expectation measurements as indicators of risk-taking propensity.
In a typical task, subjects participated in a ring-toss game in which the dis-
tance from which subjects threw a ring supposedly indicated their level of
desirability for difficulty, or risk. The results are discussed by Atkinson
and Feather (1966, p. 90):
"An issue of fundamental importance that is treated rather
inadequately here is the measurement of subjective probability
of success. Strong assumptions are made concerning how the
chances of success must appear to Ss .... The actual probabili-
ties of success obtained from...Ss in this experiment, while not
altogether relevant, since it is probability of success as it

I
F . . . . . . . .. . . .. I i i
I
-4-

appears to Ss at the beginning of the game that is required to

test hypotheses frem the risk-taking theory, do nevertheless
provide some support for the assumptions made."

14, " ptions arc that individualr -i,•h ,'c-I for sucrpss dominating fear
of failure would pick probabilities of success at medium levels of .4 - .5.
However, the average observed probability of success was only .23, indicating
the possibility of subjective factors operating which were not accounted for
in the measure.

Summary of Objective Risk Measures

The apparent lack of validity between risk measures seems to indicate serious
inadequacies inherent in many of the risk measures. In other words, classifying
an individual as risky because he scored toward the "risky" end of an objective
measure may be invalid unless subjective perceptions of risk are also known and
relate positively to the objective measure. If the individual is using informa-
tion related to the decision that alters his subjective probability and percep-
tion of utility from that assumed in the objective risk measure, then the extent
to which a score reflects the individual's riskiness is questionable.
A measure of subjective perception of risk, then, could be a valuable
46,eL in validation of any arbitrary objective measure. An objective measure
of risk would be valid only to the extent that it incorporated those dimensions
which are perceived and utilized by the decision maker.

Comparing Objective and Subjective Risk Iteasures

In an extensive review of risk taking and decision making, no comparisons
were found between objective measures of risk in decision making and subjective
estimates of risk as perceived by the decision maker.
To establish support for the feasibility of comparing objective and sub-
jective risk, it will be beneficial to examine studies comparing objective and
subjective probability. While it is acknowledged that these probability mea-
sures are not exactly equivalent to their respective measures of risk, a close
relationship exists. An objective risk measure utilizes a set of objective
probability vaiues and an individual, in deciding whether to take a certain
course of action, often may subjectively consider the probability of success
and failure of that action.
 -5-

Edwards (1962) proposed that a set of functions can be arrived at relating

subjective probability (SP) to objective probability (OP). Edwards develops
his idea in this manner (1962, p. 115):
"...each event to which a decision theory might be applied
has certain as yet unspecified characteristics which determine
its SP. One of these characteristics is,of course, OP, when-
ever OP is defined. The others are unknown. this paper will
call them identifying characteristics."

Similar events have similar idcutifying characteristics which are influential.

Mathematically, Edwards (1962) conceives of SP as a function of OP and of identi-
fying characteristics. Events with similar OPs and similar identifying charac-
teristics should have similar SPs. Edwards (1962) uses a formula with a non-
additive SEU model in which SPs do not need to reach unity.
Shuford (1959) performed an experiment in which Ss viewed square matrices
of 400 red and green thumb tacks. Ss were very accurate in estimating relative
frequency of each color. SP approached OP.
Edwards (1959) utilized similar matrices in a probability preference experi-
ment. Subjects estimated relative frequency of each color and then chose one
of two bets, which had probabilities of winning based on frequencies of occur-
rence of the component colors. Ss did not choose bets with the highest proba-
bility of winning, though estimates of relative frequency were excellent.
Edwards concluded that people perceive OPs correctly but often misuse them by
not optimizing this informdtion while betting.
Davis, Hoppe, and Hcvrnseth (1969) conducted an experiment in which subjects
made a decision to accept a known olLtcomP (wiiy vr 1op one unit) or to take a
risk and accept the consequences of an alternative event (win or lose 2, 3,
4, ... , 11 units) unknown at the time of decision. Risk-taking patterns were
detected in which strategies of subjects were influenced by expected value,
variance, and level of risk. A clear preference was shown by subjects for middle-
value strategies combining high EV and high variance. Extreme strategies were
avoided, including the one containing the optimum EV.
Comparative studies on horse race betting have been reported where fi&,al
odds were treated as subjective probabilities (Giffith, 1949; NcGlothlin, 1956),
as a relation between money bet on each horse versus total amount bet. Objec-
tive probabilities were the actual outcomes. Although SP - OP relationships
were obvious, bettors did not take advantage of the lowest and highest probabiAi-
ties.
 -- 6-

Munson (1962) found persons betting in a carnival situation perceived and

utilized objective probabilities in betting and chose optimal bets in relation
to an Expected Value Model.
Cohen (1960) had English football -layers estimate probability of success
in kicking a football through the goal posts at different distances in a skill-
type task. The players then performed the task to obtain measures of actual
probability. The estimateq were very accurate, coinciding well with the actual
probabilities.
Results of the studies discussed above seem to provide support for the
contention that individuals can accurately estimate varying levels of risk
among alternatives involving simple tasks where the parameters are few and
known, even when making more complex decisionq if enough information is krown
and correctly utilized.

Effect of Information on Risk

When making complex decisions under conditions of uncertainty, where objec-
tive probabilities are not known, a decision maker is forced to raly on whatever
information is available, lie must decide whether and when he has an optimal
amount of knowledge to act upon. To the extent that he does noL have an adequate
amount of information he may not be able to a-curately comprehend the situation
and will be taking a certain amount of risk when arriving at a final choice
(assuming the possibility of a negative outcome).
Studies relating information to risk in decision making are typically ones
in which a subject is required to make a decision about which there are a given
number of bits of information available. The more information the subject has,
the higher the probability of making a correct decision. Uhen rewards are given
for correct decisions and each bit of information costs money, a measure of risk
is involved to the extent that the subject makes a choice with less than total
information in order to maximize gain.
In an experiment by Irwin and Smith (1957), Ss were asked to estimate
whether the mean of numbers in a deck of cards was above or below zero, when
each card seen cost the subject money and a correct answer received a prize.
Number of cards requested bv an S was directly related to the value of the prize
and inversely related to the monetary cost per card.
V

-7-

Pruitt (1961) used this same procedure in relating the actual amount of

information desired by a S and an optimal amount specified by an expected value

theory. lie used an apparatus programmed to yield a sequence of red and green

lights. The subject made a decision whether the machine was biased toward red

or green. Pruitt found large individual differences, though the sample mean (of

requests) was close to the optimal value.

Edwards and Slovic (1965) -Nnducted a similar experiment with monetary in-

centives based on an optimum information-probability strategy. Their Ss picked

the optimal strategy only 34% of the time and also showed large individual dif-

ferences.

Lanzetta and Kanareff (1962) utilized decision problems much more real-life

oriented than the previous studies. Using 25 information-seeking problems with

a high degree of social content, this study provided a monetary incentive for

correct answers and a cost for information. Five possible information bits were

inversely related to probability of payoff with decisions based on no information

having a zero probability of winning and a decision using all five bits of infor-
mation having a probability of winning of one. They found information search
was directly related to value of the prize and inversely related to cost.
Hoge and Lanzetta (1968) examined the effects of subjective uncertainty
and amount of information. Subjects made decisions concerning problems in which
there were differing amounts of available information. Confidence scales were
marked by the subjects after each decision, and decision times were recorded.
Objective values of response uncertainty were calculated from the distribution
of choices using an information theory formula for average uncertainty.
The data showed that subjective uncertainty was closely related to the
objective measure of response uncertainty; increases in respo-.se uncertainty
produced decreases in confidence in decision, and increases in decision time.
The authors stated (p. 1087):
 "The prediction that subjective uncertainty would bc affected
by the degree of knowledge factor was confirmed, though it is
T
apparent that the relationship is a complex one. lhile both the
size of the set of known da a items and the size of the set of un-
known data items had a significant effect on confidence, there
were indications that the |roportion of known to total items
forms the true basis for tie effect. Such a finding is reasonable
if it can be assumed that the source of the effoct relates, as with
response uncertainty, to the individual's perception of the effect
of this factor on the probability of making a correct choice. With-
in the situation employed here, the ratio of known to unknown items
appears to offer the best standard against which to compare the
relative values of the various matrices in leading to the correct
choice."
These studies point to saveral conclusions. First, subjects seem to take
more risk when information is costly in terms of money or time required to make
the decision. Also large individual differences were found, implying that maxi-
mization models do not seem to apply, and pointing to differences in subjective
estimates of successfully arriving at the correct decision, or differential per-
ception of amount of risk i.nvolved. Secondly, the amount of information avail-
able affects the degree of subjective uncertainty and confidence in a decision.

Risk Taking in Complex Environments

Studies involving more complex decision-making environments have related
information load (total amount of information) to information search and utili-
zation (Streufert, Suedfeld, and Driver, 1965; Suedfeld and Streufert, 1966),
to decision-makinp complexity (Streufert and Schroder, 1965; Streufert, Driver,
and Haun, 1967), to rate of unintegrated and retaliatory responses (Streufert
et al., 1967), and to risk taking (Streufert and Streufert, 1968).
These last studies have used a complex simulated game environment involving
an internation conflict situation, where subjects must make decisions concerning
events and activity related to the game. Subjects use information provided them
by experimenters during the course of the simulation to produce decisions related
to various aspects (military and economic) of the game. Player decisions affect
outcomes, and rewards-losses are experienced as direct results of actions taken.
Tactical and strategic decisions are made based on differing amounts of informa-
tion and under various degrees of uncertainty.
The studies utilizing this simulation indicate that there is a certain
"optimal" level of information load and that decision makers operating under

this optimal load condition make "better" decisions. It was found that these
 -9-

optimal load conditions produced more complex, integrated, strategic, and well-
planned decisions, and fewer unintegrated, retaliatory, and inappropriate
responses.
These "better" decisions, made under optimal load conditions, also involved
higher levels of risk as determined by an objective measure in which risk in-
creased with number of men committed in battle, or with amount of money spent or
invested in an economic endeavor. There were costs and rewards related to infor-
mation quantity. If too much time was consumed in obtaining information before
making a decision, the opportunity could be lost for winning a military engage-
ment or receiving returns on economic investments. On the other hand, it could
also be costly if a decision were made without enough information.
The concept of risk used in these studies is related to the amount of infor-
mation utilized in making decisions. In the more simple tasks there was a certain
desirable amount of information which afforded an optimal strategy for maximizing
gains. Decisions based on less information increased the risk level and decisions

based on more information limited the rewards. In the complex simulation task
higher levels of measured risk were attached to optimal information loads.

Present Research
The Pruitt model has been fairly well substantiated in research dealing with
simple gambling and decision tasks, in which variance preferunces have been noted
and where large variance patterns have been avoided because of an undesirable
level of risk attached to losing. The subjective perception of risk was observed
to increase as variance became greater, reaching a point where cost of losing
became prohibitive.
A lcgical progression would be to use the Pruitt model in determining the
extent to which risk is perceived to increase with variance when making more
complex decisions, i.e., those containing the element of uncertainty found in
real-life situations where total information about the alternatives is not known.
The present study uses Pruitt's PLR model in this kind of environment, comparing
an objective measure of risk to subjective perception of risk.
The task for this study provides a controlled laboratory environment in
which complex decision making is required of the subjects. An objective measure
of risk related to the Pruitt model has been employed in previous research

utilizing this environment.

i
 -10-

MIETHOD

Subjects
One-hundred and twenty-four male undergraduate students from courses in
psychology at Purdue University participated as paid volunteers in an experi-
mental pre-programmed simulated internation conflict situation. Sixty dyad
decision-making teams were formed in which the members acted as equal rank
commanders with responsibility for making decisions regarding the military,
economic, and intelligence environment.

The Task
The Tactical and Negotiations Game (TNG) experimental simulation is an
internation conflict environment (with characteristics similar to Vietnam) and
is described in Streufert, Kliger, Castore, and Driver (1967). Teams were told
that they would play a tactical and economic game against equivalent decision
makers representing the opposing forces, and that the experimenters would arbi-
trate as judges. In reality, however, all functions of the "enemy team" were
prograwned and performed by the experimenters, insuring exposure to standardized
inputs. Subjects were free to make decisions and utilize available resources
within the limitations outlined by their respective instruction manuals.
One-half of the dyads participated in the game from the position of the
Union of Northern Hemispherical States 'UNHS), a largo outside power coopera-
ting with the military government controlling the simulated country of "Shamba."
The remaining dyads took the position of the Free Republic of Shamba (FRS),
an indigenous army supported by a neighboring neutral country, attempting to
establish their own control through force.
The Ss spent the initial two hours of the experiment studying a plavers
manual of relevant information about the history and nature of the conflict
and the current political, military, and economic conditions. The manual con-
tains equivalent information and instructions for both sides, although the
development is slightly different for the UNHS and FRS, being oriented toward
attitudinal support of the respective positions. In addition to familiarizing
the teams with the situation and available resources, the reading period pro-
vided a similar two-hour pre-experimental environment for the participants.
Identification with, and orientation toward, the Ss' particular position
in the game was enhanced by a detailed relief map of the country, displaying
 -l1-

initial location of both friendly and enemy troops, and by locating appropriate
flags, national symbols, and progress charts in various parts of the room.
All dyads participated in the simulation for five consecutive 30-minute
periods. To avoid an end effect, teams were not told which period would be
their last. During each period, the subjects received ten written messages
spaced in equal time intervals: 60% of the messages were neutral in content,
and 40% of the messages communicated success. The initial experimental period
was a warm-up period. Messages for the experimental periods contained informa-
tion on the military, economic, intelligence, and negotiations situation and
were oriented toward the developments occurring as the game progressed.
The teams of Ss made written decisions on appropriate decision forms and
included statements of purpose and rationale for the decision.
Short interim periods occurred after each playing period. During these
periodL. the subjects filled out reports and responded to various scales.
Subjects were not aware or suspicious of the actual role of the experimen-
ters on communicating Drogrammed inputs. In previous research, Streufert and
Streufert (1969) established that subject attribution of causality to experi-
menter influence has consistently been well below 20%.

Objective Measure of Risk

The objective measure of risk used in this study was the same as that
defined and used in previous research involving the TNG (e.g., Streufert and
Streufert, 1968). Decisions were classified as risky based on the dccision
context described by the subject on the decision form. For the purposes of
this research, only those decisions involving military activity were considered.
Decisions considered risky were aggressive rather than defensive in nature,
i.e., initiating possible contact with enemy troops, or placing troops or equip-
ment in positions where they could be in danger of attack by enemy forces.
Decisions not considered risky were those involving (1) preparatory moves, in
anticipation of later moves involving contact with the enemy; (2) retaliatory
moves, e.g., defending or meeting offensives initiated by the enemy; and (3)
withdrawal moves, in which retreats from precarious situations were ordered,
or moving troops to locations not involved in conflict.
This concept of risk was used as the objective measure because decisions
based on this same criterion were classified as risky in previous simulation
 -12-

research. It is thought that decisions meeting these criteria contain clearly

understood conditions of possible riskiness with a minimum of ambiguity in
interpretation.
The number of men specified on a decision form and ordered into a combat
position or engagement defined as risky was used as raw data for the analysis
of milizary risk taking. In the context of risk as developed in this paper, the
level of risk is assumed to increase with the number of men committed to a posi-
tion in which the possibility of loss of troops and equipment exists.

The Subjective Measure of Risk

The subjective perception of risk used was taken from the military decision
forms classified as risky by standards described above, and from which the ob-
jective measures were obtained. On each decision form the subject responded Lo
the question: "How risky is this decision?" The S marked a 1-7 point scale
on which the perceived inherent risk in the decision ranged from low to high.

Manipulation of Perceived Information Quantity

This manipulation was induced by an experimenter operating as an independent
information source, e.g., simulating the functions of intelligence agents, spies,
organization agents, etc., with alliance toward either the UNHS or the FRS.
The teams were called by this experimenter at the start of each playing
period and this statement read to them.
"This is UNHS (or FRS) intelligence. Intelligence sources
estimate that due to the current situation in Shamba, only %
of your communications will reach their destiiLation, and like-
.ise only __% of the messages sent to you will reach you."

At the start of the first playing period all teams received the messages
with 90% inserted in the blank spaces. In the four remaining periods the teams
were informed that estimates were either 20%, 40%, 60%, or 80%. The teams re-
ceived all four of the percentage estimates; their order varied randomly for
each team. The first period, considered a warm-up period, was not used for
data analysis involving the "percent of information" variable.
In reality, the percentage estimates did not affect the information flow,
and the subjects received all ten messages programmed for each period.
 -13-

RESULTS

A one-way ANOVA was performed to compare subjective risk with objective risk.
The PLR model of risk suggests that in the TNG environment risk level should in-
crease with number of men committed in an offensivz engagement or as variance
in amount of possible gain/loss increases. If subjective risk is closely related
to the objective measure used, then perceived risk should increase with number of
men committed in decisions by the Ss.
During the simulation Ss committed men by battalion (approximately 3000 men)
and decisions committing troops were placed in categories of either 1, 2, 3, or
4+ battalions. A total of 336 decisions made by 62 teams during four periods
of play were classified as risky under the criteria stated previously and were
used in this analysis. Figure 1 shows the results of the comparison of number
of battalions committed with the corresponding subjective level of risk. The
relationship between the two risk measures did not reach significance. Table
1 summarizes the data analysis.

TABLE 1

Summary of Analysis of Variance for Comparison

of Objective Risk With Subjective Risk

Source SS df MS F
Between subjects
12-5983 3 4.1994 1.9606
Within subjects
Error 711.1041 332 2.1419

A three-way ANOVA (2 x 2 x 4) was performed comparing objective risk and

subjective risk by (UNIS vs. FRS) and by period (1, 2, 3, 4). Original risk
scores for objeL.ive risk were obtained by placing decisions in categories of
1, 2, 3, 4+ battalions. Subjective risk was scored on a scale from 1 to 7. To
:ompare risk by team and by period (and later by percent communication effective-
ness), scores were standardized (Z-scores) to obtain equivalent scale units.
The total number of risky decisions made during the game was not equal
across periods. Decisions were discarded randomly to arrive at an equal number
 7.0j

6.0 -

5.5

to, 5. 0

A.5

4.0

o I I I ,
0.0

1 2 3 4+
B'ATTALIONS

Figure 1. Mean Level of Subjective Risk on a 7-Point Scale as a

Function of Number of Battalions Comitted in Combat.
 -15-

per cell. Eighty decisions made by UNHS teams and 80 decisions made by FRS teams
were used to assign 20 decisions per cell.

The results indicate a significant main effect due to Team (F = 40.4,

p < .001; df = 1,152) There is also a main effect due to Period with levels
of risk dropping significantly in the last Period (Period 5) (F = 3.02,
p < .05, df = 3,152). Table 2 summarizes the data analysis and Figure 2
illustrates these effects.

TABLE 2

Summary of Analysis of Variance for

Comparison of Risk by Team and Period

Source SS df MS F

Between subjects
Team 35.0681 1 35.0681 40.419***
Team X Period .6369 3 .2123 .244
ieam X Risk .8101 1 .8101 .934
Team X Period X Risk 1.4406 3 .4802 .5534
Error 131.8798 152 .8676

Within subjects
Period 6.2268 3 2.0756 3.u2*
Risk .8637 1 .8637 1.255
Period X Risk 4.2125 3 1.4042 2.0409
Error 104.5757 152 .6880

* p < .05
*** p < .001

To study the effect of manipulation of the communication factor, a three-

way ANOVA (2 x 2 x 4) was performed comparing standardized objective and sub-
jective risk scores by team (UNHS vs. FRS) and by the percent information sup-
posedly communicated (20%, 40%, 60%, 80%) (see Figure 3). One-hundred and
sixty decisions were analyzed with 20 decisions per cell.
A main effect due to Team was found in this analysis with FRS teams indi-
cating a significantly higher degree of both objective and subjective risk per
decision than the UNHS teams (F - 46.6, p < .001, df = 1,152). The effect for

level of information was not significant (see Table 3).

 -16-

.6 objective
subjective ....................

.A FRS
e.......... -.0........6, "

.2

.0

-. 2 S

UNH_ \. 0,
UN .* .,"

-.6

123 4
PERIOD

Figure 2. Standardized Objective and Subjective Risk Scores

Compared as a Function of Team and Period.
t -17-

.6 RSobjective
Ssubjective.......

.0 -

UNHS %
-
M.4
%.
44

-. 4

%6 401
-. 6

20% 40% 60% 8096

PERCENT INFORMATION

Figure 3. Standardized-Objective and Subjective RIA~ Scores

Compared as a Function of Team and Percent Information
Communicated.
 -18-

TABLE 3

Summary of Analysis of Variance for Comparison of

Risk by Team and by Percent Information Communicated

Source SS df US F

Between subjects
Team 35.5238 1 35.5238 46.61***
Team X Percent 2.1387 3 .7129 .875
Team X Risk .4019 1 .4019 .493
Team X Percent X Risk 1.8201 3 .6067 .745
Error 123.8041 152 .8145

Within subjects
Percent 3.7352 3 1.2451 1.72
Risk 2.0697 1 2.0697 2.86
Percent X Risk 4.0584 3 1.3528 1.87
Frror 109.8374 152 .7226
*** p < .001

The 31 FRS teams averaged 7.8 decisions per game that met the risk criteria
and 31 UNHS teams averaged only 3.03 decisions per game. The difference between
these means obtained significance (F = 27.85, p < .001, df = 1,60). Table 4
summarizes the data analysis.
TABLE 4

Summary of Analysis of Variance of

Test for Mean Number of Decisions Made per Team

Source SS df MS F

Between subjects
425 1 425 27.85***

Within subjects
Error 916 60 15.28

*** < .001

 A
-19-

DISCUSSION

The results of the one-way ANOVA comparing numbers of men committed (by
battalions),in a decision classified as risky, with level of perceived or sub-
jective risk seem to indicate that in the context of this simulation subjects
did not feel riskier as they engaged more men in combat.
The results obtained in this experiment are not in agreement with Pruitt's
PLR model. It should be noted, however, that the present research shows consid-
erable deviations from Pruitt's experimental design.
Pruitt's model incorporates probabilities and payoff ratios including the
definite possibility of a negative outcome associated with all levels of risk.
The more severe consequences relatcd to higher variance patterns are considered
as being more risky than low variance patterns.

In the present rescarch involving the TNG, however, there are no established
objective probabilities nssociated with a specific situation or numbers of men
involved. Subjects are not given win/lose probabilities to consider when making
decisions. At the sime_ time ncgati-ie outcom2s or loss factors, related in mag-
nitude to number of men, rire not specified.
Because objective probabilities are not given, the subjects are forced to
create their own subjective p:obabilities (expectations) based on available
information or known conditions. The data suggest that peak perception of risk
may occur more likely when groups of decision makers are committing two battalions.
It should Le noced, however, that tnese suggestions are not based on significant
shifts in risk perceptions and should be viewed very cautiously. The results
imply a possible interaction between probability of losing and cost of losing
(cf. Streufert and Streufert, 1968), wT-_h the maximum comibination potentially
occurring where two battalions are engaged in an unknown situation, and the
lowest combination occurring with four or more battalions.
It s ems reasonable to assuire that in many situations the probability of
losing an engagement would become less as the number of men increased. If proba-
bility functions were given and there was the definite possibility of losing a
relatively large army, then Pruitt's variance assumptions would have been tested
more precisely. Committing a large army may insure victory more often, but
losing a sizable force might present a large negative utility factor.
 -20-

Subjects were able to increase or decrease the size of bcth offensive and
defensive forces during the course of the game. After considering the content
of messages stating that engagements were not going well, subjects would typi-
cally either commit additional units or would withdraw and disperse to safety.
Thus, the mobility of forces usually could exempt the teams from suffering very
severe consequences. In other words, subjects may well have responded to the
simulated events with decisions which they viewed as less risky than they actu-
ally were since a relatively safe escape seemed inherent in the environmental

conditions.
The three-way analysis comparing risk by period indicated a significant
decrease in risk level in the fifth or last period. Planned post hoc comparison
tests indicate that this drop is due to perceptions of risk, not to actual risk
taking (p < .05). Possible reasons for this are not immediately evident. The
same effect has been found previously (Streufert and Streufert, 1969) where
significant decreases in level of risk occurred in this same period (five).
In the research cited the simulation was conducted for a total of seven playing
periods. Since teams were not aware of when the game would end, it eliminated
the possibility of the effect being due to the last period. The possibility
that this finding is related to number of decisions made by the teams can also
be eliminated (number of decisions per period remained approximately the same
during the four periods).
The explanation for this effect might be found in the simulation design
rather than in any manipulation. It may thus be a function of time and cumula-
tive experience. As the game progresses, economic and diplomatic considerations
by Ss increase. This shift in concentration may be a function of time and the

developing game environment. A shift toward greater use of alternate (non-

military) strategies may produce a feeling of greater "security" when military

decisions are made, resulting in lowered perceptions of subjective risk levels.

For example, a population conceived as friendly (because of aid on the economic
side) may make troops operating in their areas less endangered. Additional
research will be necessary to isolate specific causes.
The percent information communicated factor did not have a significant
effect on risk. A possible explanation is that the manipulation may not have
been effective enough to cause a significant number of Ss to seriously consider
 -21-

the communication. Although level of information factors of 90%, 8O%, 60%,

40%, and 20% were communicated to the subjects during the five periods of play,
they still received ten messages per period. However, this explanation is
probably incorrect. In most cases Ss did respond to the communication and fre-
quently (particularly during the 20% and 40% manipulations) sent additional
messages with exactly the same content to increase the probability of a message
reaching its destination. Another possibility is thaL the information manipu-
lation modified decision frequency but not risk taking.
Significant differences were found between UNHS teams and FRS teams in both
level of risk and in number of decisions made. There were sizable differences
between teams in both objective and subjective risk. FRS teams committed more
men per decision and at the same time felt or perceived a greater degree of risk
in doing so. An examination of the peneral differences in style of conducting
military engagements should explain this effect of Team in the simulation.
The UNHS military position in Shamba, because of control of all major
cities, towns, and military bases, was more defensive in nature than the FRS
position. The UNHS teams had forces in the important cities throughout the
country and their major objective was to retain control of the portions of the
country they occupied.
This, combined with the fact that the UNHS did not know exact locations of
FRS units, resulted in UNHS teams playinf, mostl, a "waiting game." UNHS teams
would wait for FRS activity or offensives and then order troops already in the
area to defend the attacked city or installation. The teams would then commit
additional units if necessary. Because UNIIS teams commanded superior forces,
the perception of risk was probably much lower than for FRS teams. Increased
mobility and vastly superior air strike capability may have contributed to a low
subjective risk factor.
Many of the types of the UNHS decisions described above did not meet the
criteria for risky decisions because they were not offensive or did not initiate
combat. This mainly defensive style of warfare is evidenced by the relatively
small number of military decisions used in UNUS team analysis. In the four
periods analyzed, UNHS teams made a total of 20, 25, 23, and 22 decisions.
During these same periods, FRS teams made 48, 60, 59, and 54 decisions. The
difference was significant.
 -22-

The FRS military position was the opposite of that of the UMIS forces.
The FRS did not control major cities and towms in the country, although Lhey
did control various segments of the rural areas throughout Shamba, FRS teams
were able to plan offensives at will. The primary tactic of FRS teams seemed
to be to commit large numbers of men against an objective and then wait for
initial results of the engagement. If results were not favorable, teams would
withdraw their units and attack at another location. FRS teams were constantly
moving troops throughout the country, attacking various enemy positions and
attempting to take control of numerous towns and cities.
This aggressive, offensive strategy contributed to the high number of FRS
decisions. An attack on a town or military unit, however, would reveal location
and troop numbers to the UNIIS forces. Risk of retaliation by larger numbers of
men with superior fire-power was high, probably contributing greatly to the per-
ception of risk when initiating attacks against UNHS-held positions.
One should note that the greater number of decisions and the greater number
of troops committed by the FRS (objective risk) covaries with greater estimates
of (subjective) risk. This finding could not be demonstrated by combining FRS
and UNHS data (see Figure 1). In this case, however, the data may be due to a
subjective vs. objective risk relationship or due to team (UNHS vs. FRS) condi-
tion. Further research appears needed.

SUMARY AND CONCLUSIONS

The failure of the present research to find a clear positive relationship

between objective and subjective risk scores illustrates the difficulties en-
countered when attempting to obtain valid measures with an objective scale of
risk. The problems in prediction and correlation demonstrated in studies cited
previously (Slovic, 1962, 1964; Kogan and Wallach, 1964) resulted in part from
not knowing enough about subjective considerations made by the persons being
scaled. Risk scales used in these studies did not include impo.Lant subjective
parameters involved. In the present study some of the same problems may have
occurred, and subjacts' perception of amount of risk involved in a decision may
not have been adequately reflected in the objective measure.
The slight and insignificant decrease in subjective perception of risk
when committing more than two battalions suggests a need to isolate the subjective
 -23-

parameters affecting the decision making, and incornorate these parameters in

the objective scale used. Tt may be found, however, that the subjective para-
meters are too numerous or complex to make this feasible when constructing a
risk scale.
The significant decrease in risk during the fifth period presents some
interesting questions regarding the simulation environment and the accrued
experiences of the Ss. A detailed examination of the game process and Ss' re-
actions and learning experiences will be required to answer these questions.
The failure of manipulation of information level to affect perceived risk
also suggests the need for further research. It appears that the percent com-
munication effectiveness factor may not have affected the Ss' perception of
risk involved. The variables may not be related or the relationship may be
more complex than anticipated.
The Team effects found in every analysis appear to he very similar to the
differences in military tactics found in South Vietnam between the United States/
South Vietnamese (ARVN) forces and the Viet Cong/North Vietnamese forces. This
is especially true when considering the pattern of the war in the years before
1968 when U.S./ARVN strategy changed to a more aggressive ground war. It would
be interesting to determine whether the relative differeuces in subjective per-
ception of risk found in the present research also occur among decision makers
for the respective sides in South Vietnam.
The Ss' actions and decisions in the TNG closely parallel those of their
counterparts in South Vietnam. The findings of significant differences due to
Team in the present research may have practical applications in stulying deci-
sion-making processes in that country.

REFERENCES

Atkinson, J. U. An introduction to motivation. Princeton, New Jersey. Van

Nostrand, 1964.
Atkinson, J. W., Bastian, J. R., Earl, R. I', and Litwin, G. H. The achievement
motive, goal-setting, and probability preferences. Journal of Abnormal
and Social Psychology, 1960, 60, 27-36.
Atkinson, J. W., and Feather, N. T. (Eds.) A theory of achievement motivation.
New York: Wiley, 1966.
 -24-

Cohen, J. Chance, skill, and luck. Baltimore: Penguin, 1960.

Coombs, C. H., and Pruitt, D. G. Components of risk in decision making:
Probability and variance preferences. Journal of Experimental Psychology,
1960, 60, 265-277.
Davis, J. I1., Hoppe, R. A., and Hornseth, J. P. Risk takingp Task, response
pattern, and grouping. Organizational Behavior and Human Performance,
1968, 3, 124-142.
Edwards, -. The prediction of decisions among bets. Journal of Experimental
Psychology, 1955, 51, 201-214o
Edwards, W. Subjective probability in decision theories. Willow Run Laboratory
Report No. 2144-361-T (University of Michigan), 1959.
Edwards, 'I. Subjective probabilities inferred from decisions. Psychological
Review, 1962, 79, 109-135.
Edwards, U., and Slovic, P. Seeking information to reduce the risk of decisions.
American Journal of Psychology, 1965, 78, 188-197.
Griffith, R. H. Odds adjustment by American horse-race bettors. American
Journal of Psychology, 1949, 62, 290-294.
[loge, R. D., and Lanzetta, J. T. Effects of response uncertainty and degree of
knowledge on subjective uncertainty. Psychological Reports, 1968, 22,
1081-1090.
Irwin, F. W., and Smith, !I. A. S. Value, cost, and information as determiners
of decision. Journal of Experimental Psychology, 1957, 54, 229-232.
Kogan, N., and Wallach, rt. A. Risk taking; A study in cognition and personality.
New York: Holt, Rinehart and Winston, 1964.
Kogan. N., and Wallach, 11. A. Risk taking as a function of the situation, the
person, and the group. New Directions in Psychology, New York: Rinehart
and Winston, 1967, 3, 113-275.
Lanzetta, J. T., and Kanareff, V. T. Information cost, amount of payoff, and
level of aspiration as determinants of information seeking in decision
making. Behavioral Science, 1962, 7, 459-473.
Lichtenstein, S. Bases for preferences among three-outcome bets. Journal of
Experinental Psychology, 1965, 69, 162-169.
Luce, R. D., and Raiffa, H. Games and decisions. New York: Wiley, 1157.
McGlothlin, W. li. Stability of choices among uncertain alternatives. American
Journal of Psychoogy1 , 1956, 69, 604-615.
 -25-

Munson, R. F. Decision-making in an actual gambling situation. American

Journal of Psychology, 1962, 75, 640-643.
Pruitt, D. G,. Information requirements in making decisions. American Journal
of Psychology, 1961, 74, 433-439.
Pruitt, D. G. Pattern and level of risk in gambling decisions. Psychological
Review, 1962, 69, 187-201.
Shuford, E. H. A comparison of subjective probabilities for elementary and
compound events. Chapel Hill, N. C.: University of North Carolina,
Psychometric Laboratory, 1959.
Slovic, P. Convergent validation of risk taking measures. Journal of Abnormal
and Social Psychology, 1962, 65, 68-71.
Slovic, P. Assessment of risk taking behavior. Psychological Bulletin, 1964,
61, 220-233.
Streufert, S., Driver, M. J., and Haun, K. T1. Components of response rate in
complex decision making. Journal of Experimental Social Psychology, 1967,
3, 286-295.
Streufert, S., Kliger, S. C., Castore, C. 11., and Driver, HI.J. A tactical and
negotiations game for the analysis of decision integration across decision
areas. Psychological Reports, 1967, 20, 155-157.
Streufert, S., and Schroder, 11. 11. Conceptual structure, environmental complexity,
and task performance. Journal of Experimental Research in Personality,

1965, 1, 132-137.
Streufert, S., and Streufert, S. C. The effect of information quantity and time
spent in military decision making on the degree of risk taking. Purdue
University: ONR Technical Report No. 15, 1968.
Streufert, S., and Streufert, S. C. Risk taking in military and economic
decision making: An analysis via an experimental simulation. Purdue
University- ONR Technical Report No. 16, 1968.
Streufert, S., and Streufert, S. C. The effect of conceptual structure, failure
and success on attribution of causality and interpersonal attitudes.
Journal of Personality and Social Psychology, 1969, 11, 138-147.
Streufert, S., Suedfeld, P., and Driver, M. J. Conceptual structure, informa-
tion search and information utilization. Journal of Personality and
Social Psychology, 1965, 2, 736-740.
 -26-

Suedfeld, P., and Streufert, S. Information search as a function of conceptual

and environmental complexity. Psychonomic Science, 1966, 4, 351-352.
Van der Meer, R. C. Decision-making: The influence of probability preference,
variance preference, and expected value on strategy in gambling. Acta
Psychologica, 1963, 21, 231-259.