Direct and indirect effects of pathological gambling on risk attitudes

Judgment and Decision Making, Vol. 2, No. 2, April 2007, pp. 126–136
Direct and indirect effects of pathological gambling on risk attitudes
Pablo Brañas-Garza?
Nikolaos Georgantzís
Pablo Guillen
Universidad de Granada
Universitat Jaume I
University of Sydney
Abstract
We study individual decision making in a lottery-choice task performed by three different populations: gamblers under
psychological treatment ("addicts"), gamblers’ spouses ("victims"), and people who are neither gamblers or gamblers’
spouses ("normals"). We ?nd that addicts are willing to take less risk than normals, but the difference is smaller as a
gambler’s time under treatment increases. The large majority of victims report themselves unwilling to take any risk at
all. However, addicts in the ?rst year of treatment react more than other addicts to the different values of the risk-return
parameter.
Keywords: risky decision making, pathological gambling, attraction and repulsion to chance.
1 Introduction
in the Diagnostic and Statistical Manual of Mental Dis-
orders published by the American Psychiatric Associ-
Since the late 1940s, individual decision making under
ation (1980). Patients with spectrum-related disorders
risk has been one of the most popular issues studied by
show an intense desire to perform a speci?c behavior
economists and psychologists. On one hand, theoreti-
preceded by unpleasant feelings and physiological acti-
cal analysis, initially undertaken mostly by economists,
vation, all of which are relieved when the behavior is per-
has framed the basic problem as a generic situation in
formed (Cartwright et al., 1998). Thus, several authors
which individuals choose from a number of probability-
consider pathological gambling (PG) as an obsessive-
outcome pairs. On the other hand, empirical contribu-
compulsive spectrum disorder (Frost, 2001 ). Contrary
tions from both disciplines have adopted a variety of
to this view, other authors argue that gambling is essen-
methodologies. These include questionnaires, economic
tially egosyntonic for the patients in all phases of the
experiments, and real-world data. The most salient and
disorder, in contrast to what happens in the obsessive-
intriguing result across all these different methodologies
compulsive spectrum disorders, where the behavior is
is that decisions in a risky environment are very sensitive
consistently egodystonic (American Psychiatric Associ-
to the framing of the choice task and to some individ-
ation, 1994). Moreover, compulsive behaviors include
ual characteristics. In this paper, we focus on the effects
increased evasive behavior, anticipatory anxiety and risk
of problem gambling on individual choice under uncer-
aversion, which are not usually observed in the behavior
tainty, as a natural ?eld for studying interaction between
of pathological gamblers (PGs).
subjects’ characteristics and their observed decision mak-
ing behavior.
This lack of agreement among experts on whether
Since 1980, pathological gambling has been included
gambling is an egosyntonic or egodystonic disorder could
even imply that gamblers may be heterogeneous with re-
?We are grateful to Jon Baron for his support and advice, and
spect to their attitudes towards their addiction. Therefore,
two anonymous referees. We would also like to thank Pilar Sánchez-
at a ?rst stage, whether gambling is an egosyntonic or
Olías, Jordi Brandts, Pedro Rey Biel, Al Roth, and John Galvin. N.
Georgantzís acknowledges ?nancial support by the Spanish Ministry
egodystonic disorder would in?uence the way PGs feel
of Education and Science (SEJ2005–07544/ECON) and Bancaixa.
about their condition. At a second stage, this could in-
Pablo Brañas-Garza acknowledges ?nancial support from DGICYT
terfere with the degree to which they feel more attracted
(SEJ2004–07554/ECON).
than normal subjects by bets involving riskier options.
Addresses: Pablo Brañas-Garza, Dpto. de Teoría e Historia Económica,
Universidad de Granada, Spain (pbg@ugr.es);
Therefore, studying whether PGs behave differently from
Nikolaos Georgantzís,
LINEEX/Laboratori d’Economia Experi-
normal subjects in risky decision-making tasks would re-
mental (LEE) and Dpto. de Economia, Universitat Jaume I, Spain
quire isolating the ?rst level of pleasure or discomfort due
(georgant@eco.uji.es);
to being a gambler from the second level of pleasure due
Pablo
Guillen,
Discipline
of
Economics,
Faculty
Eco-
nomics
and
Business,
The
University
of
Sydney
to betting on riskier options. A natural way of obtain-
(p.guillen@econ.usyd.edu.au)
ing a more homogeneous population of gamblers with re-
126

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
127
spect to their attitude towards their addiction is isolating
jects which are our control population, labeled NOR-
and studying a population of egodystonic gamblers as are
MALS. Sabater & Georgantzís (2002) and Georgantzís
those who have voluntarily decided to quit and participate
et al. (2004) provide us with a much larger data set ob-
in a Gambler Anonymous (GA) therapy group.
tained with normal student-subjects faced with the same
Several aspects of PGs’ behavior have been studied so
task under different payment methods. However, given
far. Such studies are either aimed at shedding light on
the age difference between students and our two focus
speci?c methodological issues that should be accounted
groups, we have created this new sample of normal sub-
for when studying decision making by PGs1 or are di-
jects for the sake of comparability.
rectly addressing the question whether PGs suffer form
Our results show that addicts exhibit a higher degree
some kind of cognitive bias. Among different kinds of
of risk aversion than normal individuals, although their
cognitive bias, the most obvious suspect is probability
behavior tends to convergence towards normals’ decision
distortion due to attraction to risky bets, which could
making behavior as the time under treatment increases.
yield irrational behavior re?ected on higher degrees of
Interestingly, victims are even more risk-averse. In fact,
risk taking as compared to normal subjects. Along this
a large percentage of them (around 70%) refused to take
line are the studies by Toneatto (1999a,b), Gaboury and
any risk at all.
Ladouceur (1989) and, especially, Leopard (1978), while
A second salient result is that addicts in the ?rst year
Goodie (2005) adopts a slightly different approach to
of treatment appear to be more sensitive to risk-rewarding
higher levels of risk taking showing that they are the re-
increases in expected rewards than are all other subjects.
sult of overcon?dence.
In Section 2, we further discuss our objectives and hy-
In this paper, we study risky decisions made by sub-
pothesis. In Section 3, we explain the experimental de-
jects whose lives have been directly affected by patho-
sign. Section 4 summarizes the results and Section 5 con-
logical gambling and have decided to quit by participat-
tains the conclusions. The appendix presents an English
ing in a therapy group of GA. Furthermore, we study the
translation of the instructions.
risk taking behavior of people who are indirectly affected
by pathological gambling because they are married to a
pathological gambler. We want to know whether the deci-
2 Hypotheses
sions of the aforementioned groups in an abstract lottery-
choice task signi?cantly differ from those taken by “nor-
There are few precedents for experimental economics re-
mal” subjects and, if so, in what way. In order to ad-
search on “special subject pools.” For instance, Battalio
dress this question, 82 subjects played a hypothetical ver-
et al. (1973) report the results of a token economy ex-
sion of the lottery-choice task introduced by Sabater &
periment run with 38 patients of the Central Islip State
Georgantzís (2002) and further developed and discussed
Hospital. More recently, Bosch-Doménech et al. (2005)
in Georgantzís et al. (2004). The task is designed to cap-
conducted research with Alzheimer patients, and Ernst
ture two dimensions of decision making under risk. First,
Fehr has reported currently ongoing experiments with
it can be used to distinguish between risk-averse and risk-
schizophrenics. Contrary to economists, psychologists
neutral/risk-loving subjects. It also measures an individ-
have extensively studied cognitive distortions related to
ual’s degree of risk aversion. Second, the task captures a
pathological gambling (for example, Toneatto (1999a,b),
subject’s reaction to different risk premia.
or Gaboury and Ladouceur (1989). These ?ndings moti-
Our sample consists of three different subsamples. The
vate our ?rst hypothesis:
?rst, labeled ADDICTS, consists of 32 PGs attending a
Hypothesis 1: Addicts’ attitudes towards risk are sig-
Gambler Anonymous (GA) session at the Annual Meet-
ni?cantly different from those of normal subjects.
ing of the Cordobesian Association for Patholical Gam-
We formulate the ?rst hypothesis in this generic form,
blers (ACOJER)2. The second subsample, labeled VIC-
because the difference could go in either direction. One
TIMS3, consists of 30 spouses of subjects from the ?rst
possibility is that gambling tasks are sensitive to the un-
subsample. The third subsample consists of 20 sub-
derlying attraction that addicts have toward gambling.
This is possible because the task itself does not involve
1For example, Ladouceur et al. (2007) show that gamblers exhibit
real money and is thus different from the compulsive be-
an increased willingness to participate in studies on gambling.
havior that the addicts are trying to overcome. The other
2At the moment of the experiment, they were heterogeneous with
respect to their times under psychological treatment: 15 of them were
possibility is that the addicts’ new aversion toward gam-
in their “?rst year” under treatment; 4 were in the second year; 2 in the
bling will extend to the laboratory task. Thus, the way we
third year; 5 of them in the 5th; 2 in the 6th year; 2 in the 7th year and
address this question concerns whether laboratory gam-
2 had been under treatment for over 10 years.
bling tasks are sensitive to basic impulses which are pre-
3We call gamblers’ spouses “victims” because they are the ones who
have suffered the negative consequences of pathological gambling with-
sumably still present, or to PGs’ re?ective commitment
out having a gambling problem themselves.
to give up gambling.

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
128
It is not clear how victims should be expected to be-
have towards risk. On one hand they are people who have
Table 1: Demographic Data
not been diagnosed as PGs. So, ex-ante, their behavior
could be expected to be indistinguishable from normals.
On the other hand, the evidence reported by Darbyshire
VICTIMS ADDICTS NORMALS
(2001) concerning children’s behavior living in a family
AGE (YEARS)
41.2
42.06
33.35
where parental gambling is a problem suggest that indi-
MALE (%)
13.3%
90.6%
60%
rect effects may also affect the behavior of spouses. This
n
32
30
20
motivates the second hypothesis:
Hypothesis 2: Victims behave in a signi?cantly differ-
ent way towards risk as compared to addicts and to nor-
In our experiment, no subject received any monetary
mals.
or other real reward. Subjects made decisions about
probabilities of earning hypothetical money. This pro-
cedure was followed for ethical reasons: medical proto-
3 Experimental design
cols advise against offering real rewards in gambling sit-
uations to individuals recovering from pathological gam-
Our main objective is to explore the direct and indirect
bling (see, for instance, Stinch?eld, 2003) because absti-
effects of pathological gambling on risk attitudes. We
nence from gambling is the ultimate goal of the treatment.
compare three different subsamples: addicts, victims, and
For the sake of comparability, the hypothetical framing
normals.
was also used in the case of the other two subsamples.
Our data on the two main groups were collected from a
Our instructions stressed that we were not asking for
single experimental session at Hotel El Pilar in La Carlota
names and therefore that the experimental results were
(Córdoba, Spain) in November 2003. The subject pool
going to be analyzed in a completely anonymous way.
in this session consisted of members of the “Asociación
Moreover, in order to avoid any Experimenter effect, we
Cordobesa de Jugadores en Rehabilitación” (ACOJER)
were introduced as scientists performing an anonymous
during their annual meeting. This is an association dedi-
socio-economic academic research for scienti?c purposes
cated to the psychological treatment of PGs. We ran two
rather than a medical one.
treatments in this session:
Note that there is a higher proportion of males in the
addicts sample than in the other two subsamples. Some
i. In the ?rst (addicts) treatment, all the subjects were
studies indicate that males are less risk averse than fe-
compulsive gamblers belonging to the aforemen-
males (see Harris et al. (2006) for ?nancial risk; García-
tioned GA group. Thirty-three people participated
Gallego et al. (2006) for a task similar to the one used
in the addicts treatment. Nevertheless, we gathered
here; also, Olsen & Cox (2001), and Byrnes et al. (1999)
only 32 independent observations because one sub-
and the recent review by Eckel & Grossman, in press).
ject refused to play the game at all.
So, this might introduce a bias in the comparison be-
tween addicts and normals, making addicts less risk-
ii. In the second (victims) treatment, subjects were play-
averse. Victims are mostly women. In this case, the pos-
ers’ spouses and, thus, victims of their compulsive
sible bias would favor a less risky behavior by the victims.
behavior. We gathered 30 independent observations
Our experimental design is based on the following
under the victims treatment.
slightly revised version of the ternary lotteries approach
(see Roth & Malouf, 1979, or Murningham et al. 1988,
iii. We compare the results obtained from these two sub-
for example).
ject populations to those obtained from another ex-
Let a lottery (p, X) imply a probability p of earning
perimental session run with normal subjects at the
X (else nothing). Consider a continuum of such lotteries
Instituto de Estudios Sociales Avanzados (CSIC).
constructed to compensate riskier options with increases
This is a research center which is also located in
in the expected payoff. Formally, each continuum of lot-
Córdoba. We made a public announcement for a
teries will be de?ned by the pair (c, r) corresponding, re-
hypothetical experiment and we recruited 20 volun-
spectively, to the certain payoff c above which the ex-
teers among the administrative staff. This subsam-
pected payoff is increases by r times the probability of
ple was preferred over college students because of
earning nothing. Therefore,
demographic similarities (age, geographic origins,
etc.) to the other two subsamples. Table 1 presents
descriptives on the composition of the three subsam-
c + (1 ? p)r
ples in terms of gender and age.
pX(p) = c + (1 ? p)r =? X(p) =
.
p

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
129
Panel 1
P
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Xpuntos
1.00 1.12 1.27 1.47 1.73 2.10
2.65
3.56
5.40
10.90
Preferencia
Panel 2
P
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Xpuntos
1.00 1.20 1.50 1.90 2.30 3.00
4.00
5.70
9.00
19.00
Preferencia
Panel 3
P
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Xpuntos
1.00 1.66 2.50 3.57 5.00 7.00 10.00 15.00 25.00 55.00
Preferencia
Panel 4
P
1.0
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
Xpuntos
1.00 2.20 3.80 5.70 8.30 12.00 17.50 26.70 45.00 100.00
Preferencia
Figure 1: Lottery Panels
In order to simplify the decision problem faced by our
subjects, we used lottery panels. Each panel corresponds
Table 2: Between-subject analysis.
to a discrete version of a continuum of lotteries for a dif-
Panel
ferent r. Figure 1 presents the four panels used in this
study. In the second row of each panel we present the
A.
1
2
3
4
payoffs (Xpuntos, expressed in Euros) corresponding to
KRUSKAL-W ?2.
15.48 27.62 30.74 29.87
the favorable outcome of each lottery which occurs with
p ? value 0.00
0.00
0.00
0.00
probability p. Such probabilities are given in the ?rst row.
The third row (P ref erencia) consists of empty cells,
MEDIAN ?2
16.82 25.31 34.00 35.20
one of which should be used by each subject to mark
p ? value 0.00
0.00
0.00
0.00
his or her preference (see a translation of the instructions
B.
in the Appendix). These panels were constructed using
c = 1 and r = 0.1, 1, 5, 10.
Addicts vs. Victims -3.34 -3.58 -3.82 -3.61
By inspection, the farther right the lottery chosen by
p ? value 0.00
0.00
0.00
0.00
a subject, the less risk-averse the subject is. Risk-neutral
Victims vs. Normals -3.52 -5.15 -5.36 -5.29
(or risk-loving) subjects would choose p = 0.1 in all pan-
p ? value 0.00
0.00
0.00
0.00
els. In fact, as shown in Georgantzís et al. (2004), an ex-
pected utility maximizing subject with utility U (X) =
Addicts vs. Normals -0.38 -1.98 -2.08 -2.23
X1/t would choose the lottery with a winning probabil-
p ? value 0.70
0.04
0.03
0.02
ity p = (1 ? 1 ) · (1 + c ), while a Constant Relative Risk
t
r
Aversion utility maximizer with U (X) = X1?t would
1?t
choose p = ct + t. Apart from guaranteeing that the
Risk Aversion.4
r
probabilities chosen in the task relate monotonically to a
subject’s risk aversion parameter, these predictions imply
4However, there are many alternative approaches which could ex-
that a subject should choose riskier lotteries as we move
plain our subjects’ choices as attraction to some prominent payof (Al-
from panel 1 to panel 4. These predictions also hold for
bers & Albers, 1983), a subject’s need to take some optimal degree
of risk (Pope, 1998; 2000) or the result of some heuristic (Tversky &
other well-known utility functions like those exhibiting
Kahneman, 1982) whose exhaustive review is beyond the scope of this
Constant Relative Risk Aversion and Constant Absolute
article.

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
130
a: Panel 1
b: Panel 2
c: Panel 3
d: Panel 4
Figure 2: Cumulative Frequency of Choices per Lottery Panel; A: Addicts, V: Victims, N: Normals.
4 Results
same analysis in each panel. Table 2 summarizes these
tests.
4.1 Differences in risk attitudes
Both series of tests yield identical results: samples are
not drawn from the same population. Neither the average
First, we compare behavior across subject subsamples.
nor the median can be considered invariant across sub-
Figures 2a-d present cumulative frequencies of choices
ject populations in any lottery panel. Each group’s be-
by subject subsample. Each ?gure presents choices per
havior statistically differs from the other two subsamples’
lottery panel. The horizontal axis represents p, which is
choices. A series of Mann Whitney non-parametric tests
the winning probability of a subject’s preferred lottery.
for k = 2 unpaired samples report a similar message:
Notice that lotteries are then ordered in the ?gures start-
with the exception of the comparison between addicts and
ing by the riskiest and ?nishing with the certain outcome.
normals in panel 1 (where the risk-premium trade off is
The vertical axis represents the cumulative frequency of
very low) the remaining cases show differences among
choices. We can see how a very high percentage of vic-
populations.5 Moreover in all the comparisons where vic-
tims (dashed line with black dots) prefer the safe option
tims are involved we see that test are always signi?cant
(p = 1) regardless of the panel.
differences for any value of ?. Table 2b shows this series
We can see that the baseline population (normal sub-
of tests (the p ? value is shown between brackets):
jects, continuous line) is the riskiest (see, for example,
the high percentage of people choosing p = 0.1). Finally,
Looking at each population’s average choice across
in all panels, the behavior of addicts (dashed line with
panels (victims = 0.867, addicts = 0.560 and normals
square markers) lies between the behavior of the other
= 0.385), we get that:
two samples.
Result 1a: Addicts choose safer options than normal
We can check, now, whether results are statistically dif-
individuals, and:
ferent across subsamples in each panel, based on a series
Result 1b: The large majority of victims report them-
of Kruskal-Wallis and Median non-parametric tests for
selves unwilling to take any risk at all.
k = 3 unrelated samples. The null hypothesis is that the
average (or the median) is the same in all the three sub-
5A series of Kolmogorov-Smirnov tests (non reported here) indicate
samples (victims, addicts and normals). We perform the
identical results.

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
131
a: Victims
differently across lottery panels.
Formal tests can be used to support these ?ndings in-
forming us on the extent to which subjects within each
subsample are sensitive to increases of the risk-return pa-
rameter as we move from panel 1 to panel 4.
VICTIMS: Clearly, we do not observe any variation
across panels; on average their choices are 0.85 (panel
1, hereafter p1), 0.89 (p2), 0.87 (p3) and 0.86 (p4).
Both the Friedman (?23 = 2.65; p = 0.44) and Kendall
(?23 = 2.65; p = 0.44) tests for k = 4 related samples do
not reject the null hypothesis of equality of distributions.
Thus, we cannot reject the hypothesis that all samples are
drawn from the same population. Hence, victims do not
b: Addicts
react to the 4 different values of the risk-return parameter
used to construct the four panels.
ADDICTS: The invariant average behavior observed in
the previous group is also observed among addicts. The
average behavior does not vary across panels: 0.59 (p1),
0.59 (p2), 0.53 (p3) and 0.53 (p4). Both the Friedman
(?23 = 2.62; p = 0.45) and Kendall (?23 = 2.62; p =
0.45) tests do not reject the null hypothesis. Hence, ad-
dicts did not vary their behavior across panels.
NORMALS: In contrast to the other samples, our base-
line population reacted to the risk-return trade-off in the
expected way, choosing riskier lotteries as we move from
panel 1 to panel 4. In the ?rst panel (mean choice = 0.52)
c: Normals
they behaved similarly to addicts. However, they varied
their choices when they were faced with higher values of
the risk-return parameter. Therefore, in panels 2, 3 and
4 choice averages clearly fall: 0, 38 (p2), 0, 33 (p3) and
0, 30 (p4). In contrast to what we reported above on vic-
tims and addicts, both the Friedman (?23 = 8.84; p =
0.03) and Kendall (?23 = 8.84; p = 0.03) tests reject the
null hypothesis. Thus, normal subjects do vary their be-
havior across panels.6
In a separate analysis, we de?ned premium sensitiv-
ity as the slope of the best ?tting line when choice was
plotted against panel (counting panels 1–4 as equally
spaced). A higher slope indicates a willingness to take
more risk when the premium for risk taking was higher.
Figure 3: Choice differences across panels for Victims,
Premium sensitivity did not depend on age, sex, or on
Addicts & Normals (Cum. Freq.)
overall risk attitude. It did, however, differ signi?cantly
among the three groups by a simple analysis of variance
4.2 Behavior across panels
(F2,79 = 4.60, p = .013). Mean slopes (change in re-
sponse for each step from one panel to the next) were
Now we explore within-subject behavior across lotteries.
0.000 for victims, 0.023 for addicts, 0.070 for normals.
Figures, 3a-c show cumulative distributions across pan-
6
els for each subsample. Again, the horizontal axis rep-
A more detailed examination of this ?nding concerning normal
subjects can clarify the origin of the difference across lottery panels.
resents the winning probabilities of the lotteries chosen
As we move from panel 1 to the following panels, signi?cant differ-
(p), starting by the riskiest option and ?nishing with the
ences appear [Z-Wilcoxon tests for p1 vs. p2: ?1.97 (p = 0.04); p1
sure outcome. The vertical axis represents the cumulative
vs. p3: ?2.24 (p = 0.02); p1 vs. p4: ?2.52 (p = 0.01)]. However,
frequency. Here we can observe how behavior does not
the same test fails to ?nd any difference for the remaining comparisons
[Z-Wilcoxon tests for p2 vs. p3: ?1.26 (p = 0.20); p3 vs. p4: ?0.59
seem to signi?cantly vary across panels for victims (3-
(p = 0.55); p2 vs. p4: ?1.64 (p = 0.10)]. That is, subjects are
a) and addicts (3-b) while normals (3-c) seem to behave
sensitive only to large increases in the risk-return parameter.

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
132
Post-hoc examination of the three pairs of groups showed
a signi?cant difference only between victims and normals
(F1,48 = 10.45, p = .007, with Bonferroni correction).
Addicts were in between, with somewhat greater pre-
mium sensitivity (but not quite signi?cantly in this anal-
ysis) for those in the ?rst year of treatment than those in
later years. (We discuss time in treatment further, below.)
We summarize the preceding remarks as follows:
Result 2a: Both addicts and victims tend to maintain
their choices invariant across different scenarios of the
risk-return parameter.
Result 2b: Normal subjects’ choices are sensitive to
large risk-return variations, and the normal subjects differ
signi?cantly from the victims.
4.3 Effect of time in treatment
Figure 4 presents cumulative frequencies by panel of
choices by PGs, distinguishing between those who are
in their ?rst year of treatment and those who have un-
dergone treatment for longer periods. A more detailed
analysis of the time under treatment variable would be
desirable, but attempting this in our study would lead to
excessively small subsamples for each year. Therefore,
both here and in the statistical model below we adopt the
dichotomous treatment of the variable.
However, it is also true that the ?rst year of treatment
is certainly special and, as we will see, a signi?cant effect
of the ?rst year dummy is observed.7 Figure 5 presents
the same data in a way which allows us to observe the
reaction of each type of PG to the different values of the
risk return parameter which were used to construct the
four panels. The risky decision making behavior of PGs
in the ?rst year of treatment exhibits two major differ-
ences with respect to the behavior of PGs under longer
treatment periods: First, the former make safer options,
especially avoiding lotteries involving the riskiest bets.
Second, while the behavior of all other subjects remains
largely invariant in the presence of higher risk-return pa-
rameters, PGs in the ?rst year of treatment are strongly
attracted by higher values of the risk-return parameter.
4.4 Overall analysis of individual differ-
ences
Finally, we study in a quantitative way the determinants
of individual decisions across the four panels. The es-
timation results reported in Table 3 refer to a model in
which p
Figure 4: Panels 1–4. Comparison between PGs in the
ij is subject i’s choice in panel j ? {1, 2, 3, 4}.
?rst year of treatment and PGs under longer treatment
7However a set of non-parametric Mann–Whitney test do not report
periods (Cum. Freq.).
clear differences: panel 1 (Z = ?1.26; p = 0.23), panel 2 (Z =
?1.23; p = 0.23), panel 3 (Z = ?0.90; p = 0.39), panel 4 (Z =
?0.23; p = 0.82). The largest differences are observed in panels 1 and
2 however these differences disappear for panels 3 and 4.

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
133
Table 3: Individual Behavior Model. Dependent Variable: pij (i’s choice in panel j)
Variable
Coef?cient
Std. Error
t-Statistic
Prob.
C
0.443966
0.034378
12.91430
0.0000
PREMIUM
-0.004405
0.003738
-1.178490
0.2395
FIRST
0.107583
0.050234
2.141624
0.0330
MALE
-0.073739
0.013846
-5.325516
0.0000
GA
0.149497
0.046938
3.184985
0.0016
VICTIM
0.470210
0.040997
11.46942
0.0000
2
R2 = 0.365123 R = 0.355265 S.E. of Regression = 0.2835 F-statistic = 37.03698 Prob(F-statistic) = 0.00000
The independent variables used are: a constant, C; the
We summarize the results obtained from the aforemen-
risk premium rj used to construct panel j; a FIRST
tioned model estimation together with some of the ?nd-
dummy taking the value 1 for gamblers under the ?rst
ings reported above on Figures 4 and 5 in the following
year of treatment and 0 otherwise; a MALE dummy tak-
results.
ing the value 1 for male subjects and 0 for female ones;
Result 3a: Gamblers and victims exhibit signi?cantly
GA is a dummy taking the value 1 for gamblers un-
higher degrees of risk aversion.
der treatment and 0 otherwise; and ?nally, a VICTIM
Result 3b: Gamblers in the ?rst year of treatment are
dummy.
more risk averse than those in posterior years of treat-
It is interesting to note that FIRST separates gamblers
ment, but they are attracted more than other subjects by
under the ?rst year of treatment from both normal sub-
higher degrees of return to risk.
jects and gamblers under longer treatment periods, as
Result 3a is a synthesis of Results 1a and 1b, while
well as from victims. This is inspired by the preceding
Result 3b can be interpreted as the consequence of some
discussion of ?gures 4 and 5, according to which PGs
consciously egosyntonic behavior by gamblers at an early
under the ?rst year of treatment are those whose behavior
stage of a psychological treatment. In fact, although if
differs most from that of normal subjects. The regression
there were more observations on each treatment year it
results con?rm that gamblers under the ?rst year of treat-
would be interesting to ?t a nonlinear model, this ?nd-
ment make safer options than other subjects. Therefore,
ing indicates that the ?rst year of treatment is special, be-
PGs under longer periods of participation in GA sessions
cause probably subjects in early stages of the treatment
are not so different from normal subjects.8
are more concerned with their self-image as people who
The remaining parameter estimates suggest that sub-
are free from their pathological attraction to risky bets.
jects who are indirectly affected by gambling (victims)
Their behavior in the lottery choice task implemented in
are willing to take fewer risks than all other subjects.
this study looks as if they were committed to avoid taking
Also, as reported in most previous studies on gender dif-
risky bets, but they could not hide a secondary element of
ferences in risky choice, males are willing to take more
their attraction to riskier bets when the returns to risk are
risk than females.
high.
Finally, when all observations are pooled together, sub-
Finally, we ?nd that:
jects exhibit limited attraction (although on the expected
Result 4: Males are less risk-averse than females.
direction) by higher returns to risk. This ?nding contrasts
This result is compatible with numerous previous ?nd-
with what was observed on Figure 5 above concerning
ings on the relation between gender and risky decision-
the behavior of PGs in the ?rst year of treatment, exhibit-
making.10
ing a strong reaction to higher risk-return parameters. It
also contrasts with our ?nding reported in Result 2b con-
cerning normal subjects’ attraction by large risk-return
5 Conclusions
parameters.9
This paper explores attitudes toward risk among two fo-
8For instance, in panel 1, a Mann-Whitney test for differences be-
cus populations: pathological gamblers under psycholog-
tween normal subjects and PG’s under more than one year of treatment
yields z = ?0.26 with p ? value = 0.79.
not to ?t the data suf?ciently well.
9We have tried to deal with these effects in the framework of the
10See for example, Byrnes et al. (1999), Harris et al. (2006) for the
model reported in table 3. The introduction of a premium-subsample
case of ?nancial risk, as well as the literature reviewed and results re-
interaction variable deals with these effects in a linear way which seems
ported by García-Gallego et al. (2005).

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
134
There is hardly any doubt that behavior in a risky task
can be explained as the result of a strategy aiming at
what the subject sees as the best option, after uncertainty
and reward attraction-repulsion have been accounted for.
This issue has been extensively studied so far under dif-
ferent theoretical frameworks. However, our results in-
dicate that the effects of a given strategy or a decision
making task as a whole on the perception of oneself and
others (Cross et al., 2002) also matter. Gamblers who
are voluntarily under treatment exhibit a higher risk aver-
sion than normal subjects, because probably they feel that
risky bets have already cost them a lot. In fact, patho-
logical gamblers in the ?rst year of treatment appear to
be more risk averse than normal subjects, whereas as the
number of years under treatment increase, their degrees
of risk taking approach that of normal subjects. As we
said before, this may be the result of PGs’ willingness to
present themselves as totally cured from their attraction
to risky bets. However, our results reveal a secondary
element in a PG’s behavior which should be taken into
account because it cannot be easily controlled by con-
sciously egosyntonic intentions. This element is attrac-
tion to riskier bets in the presence of higher returns to
risk. In that aspect, PGs in the ?rst year of treatment have
exhibited the strongest attraction to more pro?table risky
bets among all other subjects studied here. Furthermore,
the partners of PGs under treatment, are even more un-
willing to make risky bets, as the majority of them take
no risk at all.
Our results tend to con?rm our main hypothesis. That
Figure 5: Comparison between PGs in the ?rst year of
is, our three different subsamples behave differently in
treatment (top) and PGs under longer treatment (bottom)
an abstract lottery-choice task. The result concerning
with respect to their reactions to different risk-return pa-
the victims is consistent with the psychological literature
rameters.
focused on children’s behavior living in a family where
parental gambling is a problem. However, it is not clear
how addicts would behave if real rewards were offered.
ical treatment (“addicts”) and gamblers’ relatives (“vic-
We cannot give monetary prizes to gamblers under treat-
tims”). We compare these subsamples to a control pop-
ment, but we can do it with people who go to casinos
ulation subsample (“normals”). Our results can be sum-
and are not under medical supervision. This might be an
marized as follows:
interesting step for further research.
• Addicts are willing to take fewer risks than normal
individuals.
References
• Victims are even more risk-averse than addicts and
the majority of them are unwilling to take any risk
Albers, W., & Albers, G. (1983). On the prominence
at all.
structure of the decimal system. In Scholz, R. W. (Ed.),
Decision Making under Uncertainty, pp. 271–287. El-
• Both addicts and victims maintain their choices in-
sevier North Holland.
variant across different scenarios of the risk-return
American Psychiatric Association. Diagnostic and Sta-
trade-off.
tistical Manual of Mental Disorders. Third Ed. Wash-
ington DC: American Psychiatric Press, 1980.
• In contrast, normals’ behavior presents the expected
American Psychiatric Association. Diagnostic and Sta-
pattern of choosing weakly riskier lotteries in the
tistical Manual of Mental Disorders. Fourth Ed. Wash-
presence of a higher return to risk.
ington DC: American Psychiatric Press, 1994.

---

Judgment and Decision Making, Vol. 2, No. 2, April 2007
Pathological gambling and risk attitude
135
Battalio, R. C., Fisher, E., Kagel, J. H., Basmann, R. L.,
Murningham, J. K., Roth, A. E., & Schoumaker, F.
Winkler, R. C. & Krasner, R. (1973). A test of con-
(1988). Risk aversion in bargaining: An experimen-
sumer demand theory using observations of individual
tal study. Journal of Risk and Uncertainty 1, 101–124.
consumer purchases. Western Economic Journal 11,
Olsen, R. A. & Cox, C. M. (2001). The in?uence of gen-
411–428.
der on the perception and response to investment risk:
Bosch-Domènech, A., Nagel, R. & Sánchez-Andrés, J.
The case of professional investors. The Journal of Psy-
V. (2005). Social Capabilities Preserved in Alzheimer
chology and Financial Markets, 2, 29–36.
Patients. Working Paper, Universitat Pompeu Fabra.
Pope, R. (1998). Attractions to and repulsions from
Byrnes, J. P., Miller, D.C. & Schafer, W.D. (1999). Gen-
chance. In Leinfellner, W., Köhler E. (Eds.), Game
der differences in risk taking: A meta-analysis. Psy-
Theory, Experience, Rationality. Dordrecht: Kluwer,
chological Bulletin, 125, 367–383.
95–107.
Cartwright, C., Decaria, C., & Hollander, E. (1998).
Pope, R. (2000). Evidence of deliberate violations of
Pathological gambling: A clinical review. Practical
dominance due to secondary satisfactions — Attrac-
Psychiatry and Behavioral Health 4, 277–286.
tions to chance. Homo Economicus, 14, 47–76.
Cross, S. E., Morris.
M. L. & Gore, J. S. (2002).
Roth, A. & Malouf, M. W. K. (1979). Game-theoretic
Thinking about oneself and others: The relational-
models and the role of bargaining. Psychological Re-
interdependent self-construal and social cognition.
view 86, 574–594.
Journal of Personality and Social Psychology 62, 399–
Sabater-Grande, G. & Georgantzis, N. (2002). Account-
418.
ing for risk aversion in repeated prisoners’ dilemma
Darbyshire, P. (2001). The experience of pervasive loss:
games: An experimental test. Journal of Economic Be-
Children and young people living in a family where
havior and Organization, 48, 37–50.
parental gambling is a problem. Journal of Gambling
Stinch?eld, R.,Takushi, R., Hanson, G. & Bogan, S.
Studies 17, 23–45.
(2003). A program for the treatment of pathological
Eckel, C. and Grossman, P. (in press). Men, women and
gambling: Program participation and treatment out-
risk aversion: Experimental evidence. In C. Plott and
comes. Washington State Department of Social and
V. Smith (Eds.), Handbook of Experimental Results.
Health Services, RCW 67.70.350(5) November 1.
New York: Elsevier.
Toneatto, T. (1999). Cognitive distortions in heavy gam-
Frost, R. O. (2001). Obsessive-compulsive features in
bling. Journal of Gambling Studies, 13, 253–266.
pathological lottery and scratch-ticket gamblers. Jour-
Toneatto, T. (1999). Cognitive psychopathology of prob-
nal of Gambling Studies 17, 5–19.
lem gambling. Substance use and misuse, 34, 1593–
Gaboury, A. & Ladouceur, R. (1989). Erroneous percep-
1604.
tions and gambling. Journal of Social Behavior and
Tversky, A. & Kahneman, D. (1982). Availability: A
Personality, 4, 411–420.
heuristic for judging frequency and probability. In D.
García-Gallego, A., Georgantzís, N., Ginés, M. &
Kahneman, P. Slovic, & A. Tversky (Eds.), Judgment
Jaramillo, A. (2006). Gender and risk attitudes in bar-
Under Uncertainty: Heuristics and Biases, pp. 163–
gaining experiments. Universitat Jaume I, mimeo.
178. Cambridge UK: Cambridge University Press.
Georgantzís, N., García-Gallego, A., Sabater-Grande, G.
& Genius, M. (2004). Lottery-speci?c risk attitudes:
Probability and reward attraction vs. risk-return trade-
Appendix: Instructions
offs. Universitat Jaume I, mimeo.
Goodie, A. (2005). The role of perceived control and
Welcome to this decision-making study. This session be-
overcon?dence in pathological gambling. Journal of
longs to a research project directed by Professors Niko-
Gambling Studies 21, 481–502.
laos Georgantzís (Universitat Jaume I) and Pablo Brañas
Harris, C., Jenkins, M. & Glaser, D. (2006). Gender dif-
(Universidad de Jaén and IESA-CSIC). Identical sessions
ferences in risk assessment: Why do women take fewer
have been run in Valencia, Castellón, Crete and Athens.
risks than men? Judgement and Decision Making 1,
This session is going to last 15 minutes. We thank you
48–63.
for your participation.
Ladouceur, R., Arsenault, D., Freeston M.H. & Jacques,
You are going to be asked to take four decisions. In the
C. (1997). Psychological characteristics of volunteers
attached sheet there are four panels [panels are in Figure
in studies on gambling. Journal of Gambling Studies
1]. Take for example the ?rst one. In the ?rst row (P)
13, 69–84.
you can see decimal numbers between 1 and 0.1 (both
Leopard, D. (1978). Risk preference in consecutive gam-
included). These numbers represent probabilities with
bling. Journal of Experimental Psychology: Human
which you can hypothetically earn the amount of money
Perception and Performance, 4, 521–528
shown in the cell below this number (row “Xpuntos”).

---