Judgment and Decision
Making, vol. 1, no. 2, November 2006, pp. 153-158.
The rich get richer and the poor get poorer:
On risk aversion in behavioral decision-making
Ingmar H. A. Franken1, Irina Georgieva, Peter Muris
Institute of Psychology, Erasmus University Rotterdam, The
Department of Psychology
University of Amsterdam
Some studies have found that choices become more risk averse
after gains and more risk seeking after losses, although other
studies have found the opposite. The latter tend to use
hypothetical cases that encourage deliberation. In the current
study, we examined the effects of prior gains and losses on a
task designed to encourage less reflective decision making, the
Iowa Gambling Task (IGT). Fifty participants conducted a
manipulated decision-making task in which one group gained
money, whereas the other group lost money, followed by the IGT.
Participants who experienced a prior monetary loss displayed
more risky choice behavior on the IGT than subjects who
experienced a prior gain. These effects were not mediated by a
positive or negative affect, although the sample size may have
been too small to detect a small effect.
Keywords: implicit decision-making, reward, punishment,
Iowa Gambling Task, monetary choices, risk behavior.
Kahneman and Tversky (1979) noted that people are often risk
averse for gains and risk seeking for losses. Whether people
consider a consequence of their choice as a loss or as a gain is
dependent on their point of reference. This reference point,
which is often equivalent to the current wealth position, plays a
key role in the theory of choice.
It should be possible to manipulate perceptions of the domain
(gain or loss) with actual prior gains or losses. People may see
their starting point, before the gain or loss, as the reference
point. If they had lost money, for example, they may see new
gambles as in the domain of losses, and they therefore might be
risk seeking. Earlier studies of the effect of gains and losses
show conflicting results. Thaler and Johnson (1990) found the
opposite results - which they called a "house money effect"
- although their participants would take risks to gain back all
of their loss. Weber and Zuchel (2005) review this literature
and find some conditions that support the Prospect Theory
prediction. Aside from their result, however, most of the
results consistent with Prospect Theory are from studies that use
more realistic situations such as investment, rather than
Some traditional economic studies addressing theories of
decision-making assume that decision-making is based on deliberate
evaluations of varying option-outcome scenarios, that is, people weigh
the pros and cons of various choices against each other and base their
decision on the outcome of this comparison. These kinds of choices can
be characterized as deliberate, and carefully thought-out.
However, some recent psychological studies addressing decision-making
show that decisions can also be driven by less carefully thought-out
choices (Dijksterhuis, Bos, Nordgren, & van Baaren, 2006), are often
implicit and automatic (Hastie, 2001), and are based on
"gut-feelings" (Damasio, 1996) or emotions
(Loewenstein et al., 2001; Sanfey, Loewenstein, McClure, & Cohen,
2006). Recently, Sanfey et al. (2006) made a clear distinction between
these two psychological systems involved in economic decision-making:
an emotional system, which involves the activation of automatic
processes and a deliberative system involving controlled processes,
with each having separate neural substrates. In the present
contribution, we want to apply this recent knowledge to risk aversion.
Is risk aversion after gains the consequence of people's
deliberate, conscious decisions to avoid risk? Or is the case that risk
aversion can largely automatic, whereby people's
current reference point leads them to pursue less risky options without
deliberately weighting all outcome scenarios?
In the present study we examined the role of reference point in a
task designed to encourage automatic, emotional driven
decision-making, the Iowa Gambling Task (IGT). During the IGT
participants have to select cards from four decks that range in
probability and magnitude of rewards and punishments (Bechara,
Damasio, Damasio, & Anderson, 1994; Bechara, Damasio, &
Damasio, 2000). To translate our hypothesis pertaining to risk
aversion to the IGT, it is necessary to explain the IGT in some
detail. In the IGT, participants can repeatedly choose (usually
up to 100 times) between four decks of cards. Two of the
decks (e.g., A and B) are disadvantageous. They produce large
immediate gains, but these gains are followed by large losses,
leading to an overall loss in the long run. The other remaining
decks (e.g., C and D) are advantageous. The gains are modest but
consistent and the losses are small. Consistently choosing these
decks leads to gains in the long run. This means that people who
are risk seeking would be predominantly choose decks A and B,
leading to losses in the long run. Conversely, people who are
risk averse will predominantly choose decks C and D, leading to
overall gain. This means that risk aversion translates into
better performance (overall gains) on the IGT, whereas risk
seeking would translate into poor performance (overall losses).
The psychological process that determines people's
behavior in the IGT is crucial to our hypothesis that risk aversion is
not only based on deliberately weighting all outcome scenarios. A
general consensus is that people performing the IGT at some point steer
towards certain (profitable) decks, in rather automatic way. Whether
this automatic behavior is entirely unconscious is still subject to
debate; see Maia and McClelland (2004), and Dunn et al. (2006).
Behavior on the IGT can be seen as a form of implicit learning (Reber,
1993), whereby behavior changes before people can verbalize why they do
what they are doing. Therefore, the IGT can be regarded as an
instrument capable of assessing intuitive and emotion-based
In addition to our central aim - to test the relative automaticity of
risk aversion - we have another goal. Economic studies addressing
theories of decision-making often rely on hypothetical situations and
choices in which participants are confronted with monetary gambles
without any real consequences. Although the use of real incentives is
often not crucial for the outcome of experiments, using real incentives
has an important role to play in establishing the quality, credibility,
and generalizability of experimental data (Beattie and Loomes, 1997).
In the present study, we addressed this point by using real monetary
remunerations in order to mimic real-life decision-making more closely.
For the purpose of the present study, we experimentally manipulated the
reference point. That is, participants first performed a manipulated
gamble-task in which they either gained or lost money as a result of
their performance (in actuality, they had no influence on these gains
or losses). Note that this experimental set-up comes close to real-life
situations in which a person's reference point (real
or perceived) is often the result of their prior choices.
It is known that individual differences can influence behavioral
decision-making. These individual difference variables include reward
sensitivity (Franken & Muris, 2005), gender (Overman, 2004), and age
(Wood, Cox, Davis, Busemeyer, & Koling, 2005). In line with previous
research (Peters & Slovic, 2000), we expected that our experimental
manipulation would have an effect on participants'
affect. More precisely, a prior gain would yield an increase of
positive affect, whereas an earlier loss would result in an increase of
negative affect. It has been suggested that affect might influence
decision-making (Ashby, Isen, & Turken, 1999; Loewenstein et al.,
2001). Positive affect can promote increased sensitivity to losses
(Isen, Nygren, & Ashby, 1988). In the present study, we investigated
whether the above-mentioned individual differences and affect may have
an additional effect on the participants'
The main hypothesis was that people who experienced a prior gain on a
gambling task performed better (i.e., made more advantageous choices as
a consequence of risk aversion) on the IGT as compared to persons who
experienced a prior loss. Furthermore, we asked whether this
effect was influenced by subjective affect, and various other
Fifty undergraduate psychology students (11 males) were recruited
to participate in the present study. Their mean age was 20.6
years (SD = 3.2). All participants received course credits for
participating and could gain additional money depending on their
performance on the IGT, ranging between 1 and 6 .
Participants were randomized into two groups: a Prior Loss (PL)
group (n = 25; 5 males) or a Prior Gain (PG) group (n = 25; 6
males). All subjects signed informed consent prior to the
beginning of the experiment.
For the present study we used the computerized version of the IGT
to measure decision-making (Bechara, Tranel, & Damasio, 2000; we
used the same monetary outcomes but substituted Euros for
dollars). This task consists of 100 successive trials, which were
split into five 20-trial blocks for analysis, in which subjects
are instructed to try to gain as much money as possible by
drawing cards from one of four decks. The decisions to choose
from the decks are motivated by reward and punishment schedules
inherent in the task. Two of the decks (i.e., A and B) are
disadvantageous, producing immediate gains (large rewards) but
these are accompanied by larger losses in the long run (larger
punishments). The C and D decks are advantageous: gains are
modest but more consistent and losses are smaller. See Bechara,
Tranel, & Damasio, 2000, for the payoff and probability scheme
of the IGT. The net-score (the number of advantageous decks
choices minus the number of disadvantageous decks choices) was
used as dependent variable. A higher score indicates that a
subject is more often choosing advantageous decks. There is
general consensus that the "IGT has proved to be a sensitive,
ecologically valid measure of decision-making" (Dunn et al.,
The BIS/BAS Scales (Carver & White, 1994) were presented as a
self-report questionnaire that has been constructed to assess
individual differences in personality dimensions that reflect the
sensitivity of two motivational systems, the aversive and appetitive
system (BIS and BAS; Gray, 1987). The BIS/BAS Scales consist of 20
items that can be allocated to two primary scales: the Behavioral
Inhibition System scale (BIS; 7 items) and the Behavioral Approach
System scale (BAS; 13 items). The BAS can be divided into 3 subscales:
Fun Seeking (4 items), Reward Responsiveness (5 items), and Drive (4
items). The Dutch version of the BIS/BAS Scales has been described in
previous studies (Franken, 2002; Franken, Muris, & Rassin, 2005).
Cronbach's alphas for various scales were found to
range from .61 to .79.
The Positive and Negative Affect Scales (PANAS; Watson, Clark, &
Tellegen, 1988) were administered as a measure of positive and
negative affect. The PANAS is a 20-item bidimensional mood
inventory with a 5-point Likert-scale response format. Positive
affect reflects the extent to which a person feels enthusiastic,
active, and alert, whereas negative affect is a general dimension
of subjective distress and unpleasurable engagement that subsumes
a variety of aversive mood states, including anger, contempt,
disgust, guilt, fear, and nervousness (Watson et al., 1988).
Psychometric properties of the PANAS scales are good (Boon &
Peeters, 1999; Watson et al., 1988).
2.3 Procedure and manipulation
Participants were told that they participated in a gambling study
and that we aimed to investigate decision-making qualities.
First, participants completed all questionnaires. Subsequently,
half of the subjects carried out the "loss" version of the
manipulated IGT, while the other half conducted the "gain"
version of the manipulated IGT. For both groups, we used a
fixed, pseudo-random, gain/loss schedule irrespective of the
choices that participants made. This manipulated IGT was
programmed to yield a gain of four in the PG group and a
loss of 10 in the PL group. Irrespective of the card
choice, there was always a pre-determined pattern of
gains/losses. The proportion of cards with losses were in all
tasks and all decks 50%. There were no differences among the A,
B, C, and D decks, they were all equal. The difference between
the PL and PG condition was were the amount of losses, which were
of course larger in the PL condition. In order to make the
reference point (i.e., gain or loss) more salient (Heath,
Larrick, & Wu, 1999), participants in the PG group were told
that they gained money above average on this task, whereas
participants in the PL group were told that they lost more than
average on this task. In addition, participants were instructed
that complete new rules applied to the second game, that they
needed to employ other decision-strategies in order to gain
money, and that other decks would be advantage and disadvantage.
Again, they were told that some decks would be more advantageous
than others. Furthermore, all participants were told that their
prior loss or gain would be the starting point for the second
task. In other words, the PG group started with an initial credit
of four , and the PL group started with an initial debt of 10
. After the manipulated IGT, subjects completed the PANAS for
a second time in order to measure whether the experimental
manipulation resulted in a change of affect. Finally,
participants carried out the "real" IGT, which measured their
actual behavioral decision-making.
In order to test the main hypothesis, an hierarchical regression
analysis was carried out with the IGT net-score as dependent
variable and age, gender, group, affect (pre minus post affect
scores2), and BIS, and BAS as covariates. We entered gender
and age in the first block of the regression, group in the
second, positive and negative affect in the third, and BIS and
BAS in the fourth block. Additionally, differences on affect (pre
versus post) were tested using a 2 (time) x 2 (group) ANOVA.
Further, in order to investigate the performance of the two
groups per block (i.e., 20 cards), a multivariate ANOVA (MANOVA)
was performed with the scores on the five subsequent blocks as
Figure 1: IGT score over the five blocks per group (with standard
Figure 1 displays the mean IGT net scores of both groups over the
five blocks. As can be seen in Table 1, the group variable made a
unique and significant contribution to IGT-scores. Age, gender,
affect, BIS, and BAS did not predict IGT-scores, indicating that
these variables, including affect, had no influence on decision
Table 1: Results of hierarchical regression analyses predicting
performance on the net score of the Iowa Gambling Task.
There was a significant group x time effect for positive affect,
F(1,48) = 10.32, p = .002, and negative affect F(1,48) = 31.54, p
= .000001. More specifically, the manipulated IGT resulted in an
increase in positive affect in the PG group and an increase in
negative affect in the PL group. However, in the regression
analysis, this change in affect did not influence the relation
between the prior loss or gain and "real" IGT performance.
The MANOVA showed a significant multivariate effect, Wilks'
lambda = .75, F(5,44) = 2.89, p = .024. Follow-up MANOVAs
performed on the participants performance on the separate blocks
revealed a significant difference in the IGT scores for block 2,
F(1,42) = 8.41, p = .006, and block 3, F(1,42) = 11.05, p = .002.
The fact that only in blocks 2 and 3 participants in the PG group
made more advantageous choices than participants in the PL group
is consistent with our theorizing. In block 1, people are
generally oblivious to the nature of the decks, leading to rather
random choice behavior. In blocks 2 and 3, people are developing
preferences for certain decks, leading to more systematic
choices. Later during the task (blocks 4 and up), more and more
people start to understand the nature of the decks, leading to
consistent favorable (risk averse) choices irrespective of
|| S.E. B
|| || ||.01
|| || ||.15*
|| || ||.00
|| || ||.02
|* p .001.|
Our results show that a reference point manipulation using prior
gains or losses affected decisions with monetary consequences.
The study adds further experimental evidence that people who
"have" make more risk-averse decisions, while the "have-nots"
make more risk-seeking decisions. This phenomenon has frequently
been observed from studies using hypothetical decision-making
situations (Thaler & Johnson, 1990) and agrees with the
increased risk aversion principle of Prospect Theory. This
theory predicts exactly what we found, that is, prior losses put
the subject in the domain of losses and prior gains have the
Insofar as the IGT is, as hypothesized, sensitive to
non-deliberative mechanisms of decision making, our results show
that risk seeking and risk aversion as a function of prior gains
and losses does not need to be the result of a deliberate,
well-considered choice strategy: risk seeking and risk aversion
can be automatic and non-deliberately, it can be seen as a
spontaneous process, steering people towards or away from risk.
A secondary goal was to investigate the role of emotions
(affect). We successfully induced positive affect in the PG group
and negative affect in the PL group. However, affect variables
did not influence the relation between prior loss/gain and
decision-making. Additional correlation analysis between
positive/negative affect scores (i.e., pre, post, and pre-post
difference scores) and IGT score showed that there were no
significant links between affect and decision-making (all p's
.05). Accordingly, from the present findings, it
can be concluded that the effect of a reference point on
behavioral decision was not mediated by positive or negative
affect. This is in contrast with earlier findings of Peters and
Slovic (2000), who found that high negative affect was associated
with more avoidance of high-loss options and high positive affect
was associated with more choices from high-gain options. An
explanation for these different results might be that Peters and
Slovic used a different version of the Iowa gambling task.
Whereas we used the original task, Peters and Slovic used a
gambling task that was on several points different from the
original task. In addition, the present sample size may have
insufficient power to detect a significant result concerning the
influence of affect.
Although it is conceivable that, by the fifth block, PL
participants might have thought that the risky decks could undo
their prior loss, this could not occur in the second block, and
the difference between PL and PG conditions was already present.
Thus, we conclude that the PL does increase risk seeking in the
IGT, as predicted by Prospect Theory.
Ashby, F. G., Isen, A. M., & Turken, U. (1999). A neuropsychological
theory of positive affect and its influence on cognition.
Psychological Review, 106, 529-550.
Beattie, J., & Loomes, G. (1997). The impact of incentives upon risky
choice experiments. Journal of Risk and Uncertainty, 14(2),
Bechara, A., Damasio, A. R., Damasio, H., & Anderson, S. W. (1994).
Insensitivity to future consequences following damage to human
prefrontal cortex. Cognition, 50, 7-15.
Bechara, A., Damasio, H., & Damasio, A. R. (2000). Emotion, decision
making and the orbitofrontal cortex. Cerebral Cortex, 10,
Bechara, A., Damasio, H., Tranel, D., & Damasio, A. R. (1997). Deciding
advantageously before knowing the advantageous strategy.
Science, 275, 1293-1295.
Bechara, A., Tranel, D., & Damasio, H. (2000). Characterization of the
decision-making deficit of patients with ventromedial prefrontal cortex
lesions. Brain, 123, 2189-2202.
Boon, M. T., & Peeters, F. P. (1999). Affectieve dimensies bij
depressie en angst (Dimensions of affectivity in depression and
anxiety). Tijdschrift voor Psychiatrie, 41, 109-113.
Carver, C. S., & White, T. L. (1994). Behavioral inhibition, behavioral
activation, and affective responses to impending reward and punishment:
The BIS/BAS scales. Journal of Personality and Social
Psychology, 67, 319-333.
Crone, E. A., Bunge, S. A., Latenstein, H., & Van Der Molen, M. W.
(2005). Characterization of children's decision
making: Sensitivity to punishment frequency, not task complexity.
Child Neuropsychology, 11, 245-263.
Damasio, A. R. (1996). The somatic marker hypothesis and the possible
functions of the prefrontal cortex. Philosophical Transactions
Of The Royal Society Of London. Series: B-Biological Sciences, 351,
Dijksterhuis, A., Bos, M. W., Nordgren, L. F., & van Baaren, R. B.
(2006). On making the right choice: the deliberation-without-attention
effect. Science, 311, 1005-1007.
Dunn, B. D., Dalgleish, T., & Lawrence, A. D. (2006). The somatic
marker hypothesis: A critical evaluation. Neuroscience and
Biobehavioral Reviews, 30, 239-271.
Franken, I. H. A. (2002). Behavioral approach system (BAS) sensitivity
predicts alcohol craving. Personality and Individual
Differences, 32, 349-355.
Franken, I. H. A., & Muris, P. (2005). Individual differences in
decision-making. Personality and Individual Differences, 39,
Franken, I. H. A., Muris, P., & Rassin, E. (2005). Psychometric
properties of the Dutch BIS/BAS Scales. Journal of
Psychopathology and Behavioral Assessment, 27, 25-30.
Gray, J. A. (1987). The psychology of fear and stress.
Cambridge: Cambridge University Press.
Hastie, R. (2001). Problems for judgment and decision making.
Annual Review of Psychology, 52, 653-683.
Heath, C., Larrick, R. P., & Wu, G. (1999). Goals as reference points.
Cognitive Psychology, 38, 79-109.
Hertwig, R., & Ortmann, A. (2001). Experimental practices in economics:
a methodological challenge for psychologists? Behavioral and
Brain Sciences, 24, 383-403.
Huber, O., Wider, R., & Huber, O. W. (1997). Active information search
and complete information presentation in naturalistic risky decision
tasks. Acta Psychologica, 95, 15-29.
Isen, A. M., Nygren, T. E., & Ashby, F. G. (1988). Influence of
positive affect on the subjective utility of gains and losses: it is
just not worth the risk. Journal of Personality and Social
Psychology, 55, 710-717.
Kahneman, D., & Tversky, A. (1979). Prospect Theory: An analysis of
decision under risk. Econometrica, 47, 263-292.
Loewenstein, G. F., Weber, E. U., Hsee, C. K., & Welch, N. (2001). Risk
as feelings. Psychological Bulletin, 127, 267-286.
Maia, T. V., & McClelland, J. L. (2004). A reexamination of the
evidence for the somatic marker hypothesis: What participants really
know in the Iowa gambling task. Proceedings of the National
Academy of Sciences of the United States of America, 101, 16075-16080.
Overman, W. H. (2004). Sex differences in early childhood, adolescence,
and adulthood on cognitive tasks that rely on orbital prefrontal
cortex. Brain and Cognition, 55, 134-147.
Peters, E., & Slovic, P. (2000). The springs of action: Affective and
analytical information processing in choice. Personality and
Social Psychology Bulletin, 26, 1465-1475.
Reber, A. S. (1993). Implicit learning and tacit knowledge: An
essay on the cognitive unconscious. New York: Oxford University Press.
Sanfey, A. G., Loewenstein, G., McClure, S. M., & Cohen, J. D. (2006).
Neuroeconomics: cross-currents in research on decision-making.
Trends in Cognitive Science, 10, 108-116.
Thaler, R. H., & Johnson, E. J. (1990). Gambling with the house money
and trying to break even: The effects
of prior outcomes on risky
choice. Management Science, 36(6), 643-660.
Watson, D., Clark, L. A., & Tellegen, A. (1988). Development and
validation of brief measures of positive and negative affect: The PANAS
Scales. Journal of Personality and Social Psychology, 54,
Weber, M. & Zuchel, H. (2005). How do prior outcomes affect
risk attitude? Comparing escalation of commitment and the
house-money effect. Decision Analysis, 2, 30-43.
Wood, S., Cox, C. R., Davis, H., Busemeyer, J., & Koling, A. (2005).
Older adults as adaptive decision makers: Evidence from the Iowa
Gambling Task. Psychology and Aging, 20, 220-225.
1This work was supported by
the Netherlands Organization for Scientific Research (NWO).
Ingmar Franken, Institute of Psychology, Erasmus University
Rotterdam, Woudestein J5-43, P.O. Box 1738, 3000 DR
Rotterdam, The Netherlands, E-mail:
2Using pre-manipulated IGT and post-manipulated
IGT affect scores in the regression model yielded similar
File translated from
On 16 Nov 2006, 10:38.