Decision Analysis
Vol. 3, No. 3, September 2006, pp. 163–176
issn 1545-8490 �eissn 1545-8504 �06 �0303 �0163
informs ®
doi 10.1287/deca.1060.0075
©2006 INFORMS
A Survey Study of Factors Influencing
Risk-Taking Behavior in Real-World
Decisions Under Uncertainty
Manel Baucells
IESE Business School, Avenida Pearson, 21, 08034 Barcelona, Spain, mbaucells@iese.edu
Cristina Rata
Universitat Pompeu Fabra, Ramon Trias Fargas 25-27, 08005 Barcelona, Spain, cristina.rata@upf.edu
With the goal of investigating decision making under uncertainty in real-world decisions, we conduct asurvey requiring 261 subjects to describe a recent real-life decision and to answer questions about several
dimensions of such decision, including reference-dependence, domain, the default alternative, and the type of
consequences. We confirm a key prediction of prospect theory, namely, that perceiving the sure outcome as a loss
increases risk-taking behavior. Such perception of losses also increases the attractiveness (perceived probabilities
and estimated consequences) of the risky option. The results also confirm that the domain (professional versus
private) of a decision is a factor influencing risk-taking behavior. Risk-taking behavior does not vary across the
three groups considered (undergraduates, MBA students, and executives) and does not depend on the type of
consequences (monetary or not). We confirm that reference-dependence, and not the default alternative, is the
driver of risk-taking behavior.
Key words : real-world decisions; risk-taking behavior; reference-dependence; domain; default alternative; type
of consequence
History : Received on October 13, 2005. Accepted by Don Kleinmuntz and George Wu on August 5, 2006, after
2 revisions.
1. Introduction
Most of the descriptive research on decision making
under uncertainty is done by means of laboratory
experiments (Wu et al. 2004). Laboratory research has
the advantage of performing controlled experiments,
while allowing for the use of real monetary incen-
tives. The extent to which the results of these exper-
iments can be generalized to real-world decisions is
still under debate (e.g., Kühberger et al. 2002). Our
study aims to contribute to this debate by investigat-
ing risk-taking behavior in decisions that are likely to
be more representative of real-world decisions.
The methodology we chose was to design a sur-
vey in which subjects were asked to describe certain
aspects of a recent real-life decision. We approached
three groups of subjects—undergraduates, MBA stu-
dents (hereafter “MBAs”), and executives. The sub-
jects were required to fit the description in a simple
decision analytic framework consisting of two alter-
natives and one uncertainty. Subjects then answered
questions regarding several dimensions of their
decisions including reference-dependence, domain,
default alternative, and type of consequences. Based
on this survey, our goal is to examine the general-
izability of various phenomena that have been doc-
umented in the laboratory, as well as to investigate
some new factors. We see our methodology as a com-
plement to standard laboratory studies.
Before designing the survey, we collected a list of
potential factors that may influence risk-taking behav-
ior. Traditional decision theory argues that the mag-
nitude of outcomes and their probabilities, together
with stable risk preferences over outcomes, are the
major factors that influence risk-taking behavior
(Keeney and Raiffa 1976, Clemen and Reilly 2001).
Behavioral decision theorists expanded the list of rel-
evant drivers of risk attitudes and risk perceptions.
163
Baucells and Rata: Survey Study of Factors Influencing Risk-Taking Behavior
164 Decision Analysis 3(3), pp. 163–176, © 2006 INFORMS
Specifically, Kahneman and Tversky (1979) proposed
reference-dependence as an important factor influenc-
ing the risk-taking propensity. Subjects rely on com-
parisons to form judgments, so the comparison with
respect to a reference point induces outcomes to be
perceived as a gain or a loss. In studies of deci-
sion under uncertainty, reference-dependence is often
referred to as framing, a term that admits multiple
connotations (Levin et al. 1998). In this paper, the term
framing will refer specifically to the perception of the
sure outcome as a gain, neutral, or a loss. The ten-
dency to choose the default alternative (Samuelson
and Zeckhauser 1988, Schweitzer 1994, Johnson and
Goldstein 2003) has also been proposed as a factor
influencing the risk-taking behavior. The default alter-
native is often taken as the reference point to com-
pare outcomes, and we expect reference-dependence
to be related to the default alternative. Most studies
of these factors use experimental settings in which the
reference point and the default alternative are manip-
ulated or induced. Our goal is to separately measure
and examine the effects of reference-dependence and
the default alternative.
The domain of a decision has been proposed as
another factor that influences risk perceptions and
behavior (Slovic 1972; Hershey and Schoemaker 1980,
1994; March and Shapira 1987; Schoemaker 1990;
Weber et al. 2002; Blais and Weber 2006; see also the
survey by Levin et al. 1998 on the interaction between
domain and reference point effects for risky choices).
However, most of the studies considered only a few,
exogenously given, domains (Fagley and Miller 1990,
Rettinger and Hastie 2001). The domain of a decision,
of course, can be defined and measured in multiple
ways. Therefore, the effects of domain are more dif-
ficult to isolate than the influence of the reference
point. This explains why more studies are devoted to
framing effects than to domain effects. Our approach
allows us to analyze a range of domains. Specifi-
cally, we compare the broad categories of professional
versus private domains, together with a finer clas-
sification (investment, career, leisure, etc.) of up to
17 domains.
Incentives can be real or hypothetical, as well as
monetary or nonmonetary (e.g., use candies instead
of money). Most behavioral research in laboratories
is conducted using monetary outcomes (hypothetical
or real), or hypothetical stimuli about nonmonetary
decisions. An implicit assumption often made is that
the conclusions of studies with monetary stimuli also
apply to nonmonetary outcomes. Fagley and Miller
(1997) compared the way people make choices in
decisions having a monetary component as opposed
to being nonmonetary (human life), and concluded
that the effect of the reference point is independent of
whether the outcomes are monetary or nonmonetary.
Since in reality many decisions are either nonmone-
tary or a combination of nonmonetary and monetary
outcomes it is important to measure the influence, or
lack of influence, of the types of consequences (mone-
tary and nonmonetary such as comfort, convenience,
social, time, etc.) on risk-taking behavior.
The contribution of this paper is to develop an orig-
inal decision-making survey to explore risk behav-
ior under uncertainty. Our study complements the
laboratory studies on decision making under uncer-
tainty by sampling from actual decisions made by
subjects, thus adding realism and descriptive rel-
evance. This provides an alternative way to ver-
ify several laboratory findings, for instance, whether
reference-dependence influences risk-taking behavior
in the direction predicted by prospect theory.1 Besides
reference-dependence, we analyze other factors such
as domain, default, and the type of consequences of
a decision. Finally, our varied subject pool gives us
the opportunity to observe similarities and differences
in decision making among different subjects, not only
undergraduates.
Other methodologies exist to study real-world de-
cisions. For instance, in the experience sampling
method (ESM), subjects are alerted by mobile phone
messages and requested to fill out a short question-
naire reporting a recent decision at random times of
the day. For instance, Hogarth (2006) used the ESM
to look at the effect of feedback on confidence in
everyday decision making. One of the methodologi-
cal advantages of the ESM method compared to our
method is that it avoids memory biases in reporting
past decisions.
1 We are not the first to study prospect theory in the real world.
However, previous research examined mainly reference depen-
dence (e.g., Camerer 2000) and risk attitudes (e.g., Binswanger
1980) for given domains.
Baucells and Rata: Survey Study of Factors Influencing Risk-Taking Behavior
Decision Analysis 3(3), pp. 163–176, © 2006 INFORMS 165
Table 1 Characteristics of Subjects
Undergrads MBAs Executives Total
N 77 131 53 261
Median age 24�5 28 36 28
Female (%) 45 27 4 28
Country 86% United States 30 diff. countries 91% Spain
The remainder of the paper is organized as fol-
lows. Section 2 describes the survey design, and dis-
cusses the measurement, coding of the variables, and
methodological issues. Section 3 explains the statisti-
cal results. Section 3.1 presents an overview of sim-
ilarities and differences across the three groups with
respect to the measured dimensions. Section 3.2 stud-
ies the relative attractiveness of the risky outcome (in
terms of probabilities and outcomes) and its relation-
ship with the other factors. In §3.3, we apply a logistic
regression model to examine the effect of reference-
dependence and domain on risk-taking propensity.
Section 3.4 offers an expanded logistic regression
model to examine the effect of the default alterna-
tive, type of consequences, and group. Finally, §3.5
discusses the relationship between default alternative
and reference points. Section 4 concludes. We include
an appendix which briefly reviews the predictions of
prospect theory as applied to our decisions (sure out-
come versus a binary gamble), as one varies the ref-
erence point continuously from the worst to the best
outcome.
2. Research Methods
2.1. Subjects
We distributed a questionnaire to three groups of par-
ticipants.2 The first group consisted of 77 undergradu-
ate students from Duke University. The second group
was made up of 131 MBA students at IESE Business
School in Barcelona, Spain. The third group consisted
of 53 executives who were enrolled in the executive
education program at IESE Business School. Table 1
summarizes the different demographical character-
istics of the undergraduates (subsequently “Under-
grads”), MBA students, and Executives.
2 The questionnaire can be downloaded from the research section
at http://webprofesores.iese.edu/mbaucells/.
2.2. Survey Design
Figure 1, not shown in the questionnaire, underlies the
design of the questionnaire. r is the reference point, xs
is the monetary equivalent of the safe alternative (S),
and xs − r is the perceived monetary equivalent gain
or loss associated with such outcome. Likewise, xb− r
and xw − r , with xb > xw, are the perceived gains or
losses associated with the better and worse outcomes,
respectively, of the risky alternative (R). Finally, p is
the probability of the better outcome.
The questionnaire begins with introducing the pur-
pose of the study, and asking the subjects to describe
briefly one recent decision. To conform to the simple
decision analytic scheme of Figure 1, the decision is
constrained to two alternatives, a sure alternative S
and a risky alternative R� Furthermore, the potential
outcomes of the risky alternative had to be summa-
rized in two scenarios, a better outcome scenario and
a worse outcome scenario. Subjects were asked to
briefly describe the two alternatives and the three out-
comes.
Then subjects were required to answer a number of
questions which are meant to measure several dimen-
sions of a decision. Some of those dimensions (e.g.,
p, xb − xw) correspond to the elements of a decision
depicted in Figure 1, while others are meant to help
us to further characterize a decision (the type of con-
sequence, default alternative, reference-dependence,
and final choice). In what follows, we briefly describe
Figure 1 Decision Framework Underlying the Questionnaire
Risky
alternative (R)
p
1–p
xs–r
xb–r
xw–r
Safe alternative (S )
Better outcome
Worse outcome
Sure outcome
Baucells and Rata: Survey Study of Factors Influencing Risk-Taking Behavior
166 Decision Analysis 3(3), pp. 163–176, © 2006 INFORMS
these questions, in the same order in which they were
presented in the questionnaire.
Type of Consequence. Subjects were asked to clas-
sify the outcomes of their decisions according to one
or more of the following seven categories: monetary,
comfort (or discomfort), convenience, time (arriving
on time or late, delays, waiting), social consequences
(fame, embarrassment), career, and other.
Probability p. Subjects were presented with a di-
rect linear scale between 0 and 1 with increments of
10% and were requested to use a cross to indicate the
estimated probability of the better outcome.3
Default Alternative. We explained to the sub-
jects that “a default or nonproactive alternative is
actually chosen if nothing is done.” Subjects were
then requested to decide whether in their decisions
(a) the safe alternative was the default alternative, or
(b) the risky alternative was the default alternative,
or (c) neither alternative was the default due to the
fact that both alternatives required some action to be
taken.
The outcome of the default alternative will be
highly correlated with the continuation of the current
state of the world, also called status quo.
The Attractiveness of the Safe Alternative q. In
the same way that p measures probabilities, q mea-
sures consequences. The fraction
q = xs − xw
xb − xw
indicates the position of the sure outcome relative to
the better and worse outcomes in a 0–1 scale. We call
q the attractiveness of the safe alternative. For non-
trivial decisions, q takes values strictly between 0 to 1.
The difference between p and q measures the relative
attractiveness of the risky option. A risk-neutral deci-
sion maker strictly prefers the risky outcome if and
only if p > q. This observation can be easily seen by
setting r = xw, and observing that the expected values
of R and S are p
xb − xw� and q
xb − xw�, respectively.
3 As a double check, we asked them to write the numerical estimate,
which for most of the subjects was consistent. In case of disagree-
ment, we took the numerical estimate.
Figure 2 Scale to Locate the Sure Outcome with Respect to the Better
and Worse Outcomes
Worse
outcome
Better
outcome
Sure outcome
q = 0.4
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0
We elicited q using a direct scale.4 Specifically, sub-
jects were presented with a scaled line as in Figure 2,
and asked to estimate the location of the sure out-
come with respect to the worse and better outcomes
in terms of their preferences.
Reference-Dependence. The locus of the reference
point r relative to the outcomes yields a gain, neu-
tral, or loss perception for each outcome. To measure
reference-dependence, we directly asked subjects their
perception. For each outcome, we provided the sub-
jects with three checkboxes so that they could indicate
their perception of the outcome: as a gain, a loss, or
neutral (neither a gain nor a loss).
Final Choice. Subjects were asked to write down
whether they chose the safe or the risky alternative.
This is the variable to be explained in our study.
We also asked the subjects—in the case when the
uncertainty was solved—to briefly describe “What
happened?” Many subjects left this question blank
and we do not consider this question in the analysis.
4 We also requested a quantitative measure of the monetary equiv-
alents of the outcomes. Specifically, subjects were asked to imagine
that they had chosen the risky alternative and the worse outcome
had happened. In this case, they have to provide us with their
willingness-to-pay to replace (a) the worse outcome with the better
outcome and (b) the worse outcome with the sure outcome. This
information provides us with the differences
xb−xw� and
xs−xw�.
As a double check, we also requested
xb − xs� in the same way.
The sum of
xb − xs� and
xs − xw� should agree with
xb − xw�.
Although exact agreement between these two measures occurs in
fewer than half of the cases, both estimates have the same order of
magnitude, and the correlation between Ln
xb−xs�+
xs−xw�� and
Ln
xb − xw� is 0.95. The following three ratios—
xs − xw�/
xb − xw�,
xs − xw�/
xb − xs�+
xs − xw��, and 1−
xb − xs�/
xb − xw�—provide
estimates of q. Let qm be the median of these three numbers. The
correlation between qm and q is 0.56. While the correlation is some-
what low, we confirm that our logistic regression coefficients do
not change significantly if we use qm instead of q. We decided to
use the direct scale estimate q because it was more straightforward
and generated more valid answers. Hence, the elicited monetary
equivalents are not used in our analysis.
Baucells and Rata: Survey Study of Factors Influencing Risk-Taking Behavior
Decision Analysis 3(3), pp. 163–176, © 2006 INFORMS 167
2.3. Coding of the Variables
As a prerequisite to analyzing the results of the sur-
vey, we coded the responses as follows.
• Reference-Dependence. We classified the decisions
according to the perception of the sure outcome,
i.e., as gain in the cases that the sure outcome was
reported as a gain, as loss if the sure outcome was
perceived as a loss, and as neutral if the sure outcome
was perceived as neutral. We call this the framing
of the sure outcome. Strictly speaking, only 11 deci-
sions were perceived as either all-gains or all-losses,
in the sense that the worse outcome is perceived as
neutral or gain, or the better outcome as neutral or
loss. Hence, most of the decisions were mixed, having
both gains and losses as consequences. We eliminated
17 cases that showed some inconsistency from the
analysis of reference-dependence. For example, if the
safe alternative was perceived as neutral, then the bet-
ter outcome cannot be perceived as a loss, nor can the
Table 2 Domain of the Decisions for the Different Groups
Undergrads (%) MBAs (%) Executives (%) Total (%) Risky (%)
Professional
Human resources (assign tasks, choose collaborators, — — 19 4 100
organize subordinates)
Start MBA (keep current job or start MBA) — 29 — 15 94
Business (decisions made in the current job) — 2 28 7 73
Job (change job or not) 4 8 30 11 88
Protocol (how to deal with superiors1) 5 2 4 3 83
Studying (continue education or not) 12 2 — 5 73
Subtotal Professional 21 44 81 44
Private
Safety (undertake laser eye surgery, drive after drinking, 12 2 — 4 90
wear helmet)
Location (move to another city/country or not) 6 3 2 4 50
Investment (invest personal wealth) — 7 2 4 70
Relationship (continue/start or not a relationship) 6 3 2 4 37
Buy/sell (whether to buy/sell something and choice 6 10 2 7 79
of supplier)
Flat rental (rent a flat or wait for other opportunities) — 11 2 6 75
Ethics (tell the truth, break the law) 8 — — 2 48
Organization (plan activities, schedule, do now/do later) 16 13 6 12 77
Leisure (entertainment activities and sports2) 6 6 — 5 57
Traveling (traveling/vacation decisions) 3 2 4 3 73
Campout (camp out or do something else) 16 — — 5 80
Subtotal Private 79 56 19 56
1Examples of protocol are “to attend dinner after the interview/not attend,” or “abide with supervisor/confront him.”
2Since our questionnaire asks subjects to recall a risky decision, a few of them reported leisure decisions involving risky sports such as
sky diving or paddling in the open ocean.
worse outcome be perceived as a gain. We then intro-
duce two binary variables, D_GAIN and D_LOSS, to
account for the three levels of reference-dependence.
D_GAIN = 1 for those reporting that the sure out-
come was perceived as gain and 0 otherwise, and
D_LOSS= 1 for those reporting that the sure outcome
was perceived as a loss and 0 otherwise.
• Domain. Subjects were not asked t