Author's personal copy
Journal of Economic Psychology 30 (2009) 651–666
Contents lists available at ScienceDirect
Journal of Economic Psychology
journal homepage: www.elsevier.com/locate/joep
A parameter-free analysis of the utility of money for the general
population under prospect theory
Adam S. Booij a, Gijs van de Kuilen b,*
a
b
University of Amsterdam, School of Economics, Roetersstraat 11, 1018 WB Amsterdam, The Netherlands
TIBER, CentER, Tilburg University, P.O. Box 90153, 5000 LE Tilburg, The Netherlands
a r t i c l e
i n f o
Article history:
Received 24 July 2008
Received in revised form 20 March 2009
Accepted 28 May 2009
Available online 3 June 2009
JEL classification:
C91
C93
D81
PsycINFO classification:
3000
3920
Keywords:
Prospect theory
Utility measurement
Loss aversion
Gender differences
a b s t r a c t
Extensive data has convincingly demonstrated that expected utility, the reigning economic
theory of rational decision making, fails descriptively. This descriptive failure casts doubt
on the validity of classical utility measurements. Prospect theory can better explain choice
behaviour because it makes the plausible assumption that risk attitudes are not only driven
by sensitivity towards outcomes (utility curvature), but also by sensitivity towards probabilities (probability weighting), and by sensitivity towards whether outcomes are above or
below a reference point (loss aversion). This paper presents an experiment that completely
measures the utility- and loss aversion component of risk attitudes, using a representative
sample of N = 1935 respondents from the general public, in a parameter-free way. This
study thereby provides the first measurement of the utility of money for the general population that is valid under (cumulative) prospect theory, does not depend on prior assumptions about the underlying functional form of utility, is externally valid, and does not rule
out heterogeneity of individual preferences. The results confirm the concave–convex pattern of utility as predicted by prospect theory, suggest that utility curvature is less pronounced than suggested by classical utility measurements, and show that women are
significantly more loss averse than men.
Ó 2009 Elsevier B.V. All rights reserved.
1. Introduction
Expected utility is the reigning economic theory of rational decision making under risk. In the expected utility framework,
outcomes are transformed by a utility function and prospects are evaluated by their probability-weighted average utility.
Then risk attitudes are solely explained by utility curvature. For example, risk aversion (preferring the expected value of
a prospect to the prospect itself) holds if and only if the utility function is concave, implying diminishing marginal utility.
However, extensive experimentation has convincingly shown that ‘‘risk aversion is more than the psychophysics of money”
(Lopes, 1987, p. 283); numerous studies have systematically falsified expected utility as a descriptive theory of individual
decision making (Allais, 1953; Kahneman & Tversky, 1979). This descriptive inadequacy has been the main inspiration for
the development of alternative theories of individual decision making under risk (Starmer, 2000). The most prominent of
these non-expected utility models is prospect theory (Kahneman & Tversky, 1979; Tversky & Kahneman, 1992).
Prospect theory entails that besides the transformation of outcomes, probabilities are transformed by a subjective probability weighting function, reflecting the tendency of preferences to be overly sensitive to probabilities close to zero, and to
* Corresponding author. Tel.: +31 (0) 13 4668056; fax: +31 (0) 13 4663042.
E-mail addresses: [email protected] (A.S. Booij), [email protected] (G. van de Kuilen).
0167-4870/$ - see front matter Ó 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.joep.2009.05.004
Author's personal copy
652
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
certainty. In addition, prospect theory entails that outcomes are evaluated relative to a reference point, reflecting sensitivity
towards whether outcomes are better or worse than the status quo. Risk attitudes are thus determined by a combination of
utility curvature, subjective probability weighting, and the steepness of the utility function for negative outcomes (losses)
relative to the steepness of the utility function for positive outcomes (gains), i.e. loss aversion. As a result, the one-to-one
relationship between utility curvature and risk attitudes no longer holds under prospect theory, and the validity of measurements of risk attitudes from the general population based on the descriptively invalid expected utility model (e.g. Andersen,
Harrison, Lau, & Rutström, 2008; Barsky, Juster, Kimball, & Shapiro, 1997; Cohen & Einav, 2007) can be questioned. Indeed,
classical utility measurements have often led to preference inconsistencies (Hershey & Schoemaker, 1985) or theoretical
implausibility’s (Rabin, 2000a).
A detailed separation of risk attitudes into a subjective probability weighting-, a utility curvature-, and a loss aversion
component requires the collection of considerable data. The few studies that have used prospect theory to analyze risk attitudes for the general population were forced to adopt stringent parametric assumptions because not enough variation in the
data was available for such a detailed separation of risk attitudes (e.g. Donkers, Melenberg, & van Soest, 2001; Tu, 2005).
Moreover, these studies either assume homogeneity in preferences or model heterogeneity in an inflexible parametric
way, such as through normally distributed random coefficients. The experimental approaches that use non-parametric
methods to elicit utilities at the individual level (e.g. Abdellaoui, 2000; Abdellaoui, Barrios, & Wakker, 2007a; Bleichrodt
& Pinto, 2000) circumvent these problems, but the external validity of the results of these studies can be questioned if risk
attitudes are related to socio-demographic characteristics that differ between the general population and the student subject
pools used in these experiments.
This paper presents the results of an experiment that measures the utility component of risk attitudes for positive and
negative monetary outcomes, using a large and representative sample of N = 1935 respondents from the Dutch population,
in a completely parameter-free way. The utility measurement technique we use is the tradeoff method introduced by Wakker and Deneffe (1996). This method is robust against subjective probability distortion and is parameter-free in the sense
that no prior assumptions about the true underlying functional form of the utility function have to be made. In addition,
we obtain parameter-free measurements of ‘‘the core idea of prospect theory” (Kahneman, 2003, p. 1457); loss aversion.
Empirical research has shown that loss aversion is a major component of risk attitudes (Kahneman, Knetsch, & Thaler,
1991), and it can explain a wide variety of anomalous behaviour in the field such as the equity premium puzzle (Benartzi
& Thaler, 1995; Mehra & Prescott, 1985), the status quo bias (e.g. Johnson, Hershey, Meszaros, & Kunreuther, 1993), and a
downward sloping labour supply curve (Camerer, Babcock, Loewenstein, & Thaler, 1997). Unlike previous measurements
of risk attitudes, this study thus provides the first measurement of risk attitudes of the general population that is valid under
(cumulative) prospect theory, does not depend on a priori assumptions about the underlying functional form of the utility
function, is externally valid, and does not rule out heterogeneity of individual preferences. The dataset also allows us to test
whether utility curvature is more or less pronounced for losses than for gains, whether scaling up monetary outcomes leads
to a higher or a lower degree of utility curvature or loss aversion, and to relate the obtained measurements of utility curvature and loss aversion to socio-demographic characteristics.
First, the results show that utility is concave for gains and convex for losses, reflecting diminishing sensitivity as predicted
by prospect theory but contradicting the classical prediction of universal concavity. In addition, our result support Rabin’s
(2000b, p. 202, Chap. 11) claim that diminishing marginal utility is an ‘‘implausible explanation for appreciable risk aversion,
except when the stakes are very large”; we confirm that utility curvature is less pronounced than suggested by studies that,
erroneously, assume the classical expected utility model. Second, our results confirm the common finding that females are
more risk averse than males (e.g. Cohen & Einav, 2007; Fellner & Maciejovsky, 2007; Hartog, Ferrer-i-Carbonell, & Jonker,
2002). Unlike previous studies that were not able to unambiguously decompose the different components of risk attitudes
and often ascribe this gender difference solely to differences in the degree of utility curvature, our results show that this phenomenon is primarily driven by the fact that females are more loss averse than males. Our results also show that highly educated respondents are less loss averse, suggesting that measurements of loss aversion based on highly educated student
samples as often used in experiments lead to an underestimation of loss aversion. Finally, we did not find significant evidence for the hypothesis that both the degree of utility curvature and the degree of loss aversion increases with the size
of the outcomes involved.
The remainder of this paper is organized as follows. Section 2 briefly presents Tversky and Kahneman’s (1992) prospect
theory. Section 3 provides an explanation of the measurement techniques we used to obtain parameter-free measurements
of utility curvature and loss aversion at the individual level. A description of the experiment method is in Section 4, followed
by the presentation of the experimental results in Section 5. A discussion of the experimental results can be found in Section
6. Section 7 concludes. Details of the experimental instructions are in Appendix A, and the details of how we corrected for
sample selection can be found in Appendix B.
2. Prospect theory
We consider decision under risk, with R the set of possible monetary outcomes. A prospect is a finite probability distribution over (monetary) outcomes. Thus, a prospect yielding outcome xi with probability pi (i = 1, . . ., n) is denoted by
(p1:x1, . . . , pn:xn). A two-outcome prospect (p:x, 1 p:y) is denoted by (p:x, y) and the unit of payment for outcomes is
Author's personal copy
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
653
one euro. In this paper, prospect theory refers to the modern (cumulative) version of prospect theory introduced by Tversky
and Kahneman (1992), that corrected the original ’79 version for violations of stochastic dominance and, more importantly,
can also deal with uncertainty, i.e. the case of unknown probabilities. Prospect theory entails that the value of a prospect
with outcomes x1 6 6 xk 6 0 6 xk+1 6 6 xn is given by:
k
X
i¼1
pi Uðxi Þ þ
n
X
pþj Uðxj Þ
ð2:1Þ
j¼kþ1
Here U: R ! R is a continuous and strictly increasing utility function satisfying U(0) = 0, and p+ and p are the decision
weights, for gains and losses respectively, defined by:
pi ¼ w ðp1 þ þ pi Þ w ðp1 þ þ pi1 Þ for i 6 k; and
pþj ¼ wþ ðpj þ þ pn Þ wþ ðpjþ1 þ þ pn Þ for j > k:
ð2:2Þ
Here w+ is the probability weighting function for gains and w is the probability weighting function for losses, satisfying
w (0) = w(0) = 0 and w+(1) = w(1) = 1, and both strictly increasing and continuous. Thus, the decision weight of a positive
outcome xi is the marginal w+ contribution of pi to the probability of receiving better outcomes, and the decision weight of a
negative outcome xi is the marginal w contribution of pi to the probability of receiving worse outcomes. Finally, note that the
decision weights do not necessary sum to 1 and that prospect theory coincides with expected utility if people do not distort
probabilities, i.e. prospect theory coincides with expected utility if w+ and w are the identity.1
+
3. Measuring utility curvature and loss aversion
3.1. Measuring utility curvature: The tradeoff method
The (gamble-) tradeoff method, introduced by Wakker and Deneffe (1996), draws inferences from a series of indifferences
between two-outcome prospects in order to obtain a so-called standard sequence of outcomes, i.e. a series of outcomes that
is equally spaced in utility units. Unlike other elicitation techniques often used to measure individual utility functions such
as the certainty equivalent method, the probability equivalent method, and the lottery equivalent method (McCord & de
Neufville, 1986), utilities obtained through the tradeoff method are robust to subjective probability distortion. Hence, besides being valid under expected utility, the tradeoff method retains validity under prospect theory, rank-dependent utility
and cumulative prospect theory (Wakker & Deneffe, 1996).
Consider an individual who is indifferent between the prospects (p:x1, g) and (p:x0, G) with 0 6 g 6 G 6 x0 6 x1. In most
existing laboratory experiments employing the tradeoff method (as well as in our experiment) individual indifference is obtained by eliciting the value of outcome x1 that makes a person indifferent between these two prospects while fixing outcomes x0, G, g, and probability p. Under prospect theory, indifference between these prospects implies that:
wþ ðpÞðUðx1 Þ Uðx0 ÞÞ ¼ ð1 wþ ðpÞÞðUðGÞ UðgÞÞ
ð3:1Þ
Thus, under prospect theory, the weighted improvement in utility by obtaining outcome G instead of outcome g is equivalent to the weighted improvement in utility by obtaining outcome x1 instead of outcome x0. Now suppose that the same
person is also indifferent between the prospects (p:x2, g) and (p:x1, G). If we apply the prospect theory formula to this indifference we find that:
wþ ðpÞðUðx2 Þ Uðx1 ÞÞ ¼ ð1 wþ ðpÞÞðUðGÞ UðgÞÞ
ð3:2Þ
Combining Eqs. (3.1) and (3.2) yields:
Uðx2 Þ Uðx1 Þ ¼ Uðx1 Þ Uðx0 Þ
ð3:3Þ
Thus, the tradeoff in utilities between receiving outcome x2 instead of outcome x1 is equivalent to the tradeoff in utilities
between receiving outcome x1 instead of outcome x0. Or, put differently, x1 is the utility-midpoint between outcome x0 and
outcome x2 and the sequence of outcomes x0, x1, x2 is equally spaced in terms of utility units. One can continue eliciting individual indifference between prospects (p:xi, g) and (p:xi1, G) in order to obtain an increasing sequence x0, . . ., xn of gains that
are equally spaced in utility units. A similar process can be used to construct a decreasing sequence of equally spaced losses.
More specifically, individual indifference between the prospects (p:yi, l) and (p:yi1, L) with 0 P l P L P y0 P y1 P . . . P yn
implies that the resulting decreasing sequence of losses y0, . . . , yn is equally spaced in utility units. In what follows, we will
use the term utility increment (utility decrement) to denote the equal utility difference between the elements of an
increasing (decreasing) standard sequence of gains (losses).
1
Strictly speaking this only holds if the expected utility model allows for reference dependence. There is a debate in the literature on whether the expected
utility model presupposes that outcomes are defined strictly in terms of final wealth (which precludes reference-dependence) or not. Both Cox and Sadiraj
(2006) and Rubinstein (2006) argue that it is not defined in terms of final wealth, and subsequently conclude that Rabin’s (2000a) critique of the expected
utility model does not necessarily hold. Wakker (2005) however argues that, even though the fundamental axioms of expected utility do not say anything about
the nature of outcomes, the model only has normative content if it is defined over final wealth.
Author's personal copy
654
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
U
1
y6
c
y1 y0
0
3 x0
x1 d
b
x6
2
-1
Fig. 3.1. Linking the utilities for gains and losses.
3.2. Measuring loss aversion
The tradeoff method allows measuring utilities for either gains or losses. Without further information these measurements cannot be combined, because they are not on the same scale. This requires the elicitation of additional indifferences
that involve mixed prospects, i.e. prospects that involve both gains and losses, which amounts to measuring loss aversion.
With the proper use of the obtained standard sequences, this can be done in a parameter-free way.
Our method to measure loss aversion consists of three steps. We first determine the utility of x0 in terms of utility increments. To do so, we elicit the value of outcome b that makes an agent indifferent between the prospects (r:b, 0) and (r:x1, x0),
where r is some fixed probability. Under prospect theory, indifference between these prospects implies:
Uðx0 Þ ¼
wþ ðrÞ
ðUðbÞ Uðx1 ÞÞ
1 wþ ðrÞ
ð3:4Þ
For given probability weights this equation pins down the utility of outcome x0 in terms of increments of the standard
sequence for gains. Suppose for example that wþ ðrÞ ¼ 1 wþ , and that b falls within the standard sequence for gains, say
half-way between x2 and x3. Then, using linear interpolation to obtain the utility of b, Eq. (3.4) implies that the utility of
x0 is equal to a factor 1.5 times the utility increment for gains, i.e. Uðx0 Þ ¼ 1:5ðUðx1 Þ Uðx0 ÞÞ. The linear approximation of
the utility of outcome b can be justified on the grounds that utility is often found to be linear over small monetary intervals
(Wakker & Deneffe, 1996).2 Graphically, Eq. (3.4) identifies the utility of outcome x0 in terms of the amount of steps of the standard sequence, as illustrated by brace 1 in Fig. 3.1.
In the second step, an analogous question for losses allows for the identification of the utility of outcome y0 in terms of
utility decrements. We obtain the outcome c that makes an agent indifferent between the prospects (r:0, c) and (r:y0, y1).
Under prospect theory, indifference between these prospects implies:
Uðy0 Þ ¼
w ð1 rÞ
ðUðcÞ Uðy1 ÞÞ
1 w ð1 rÞ
ð3:5Þ
Eq. (3.5) identifies the utility of outcome y0 in terms of utility decrements of the standard sequence for losses, which is
illustrated by brace 2 in Fig. 3.1. The outcome c must fall within the standard sequence for losses and its utility can be obtained by using (linear) interpolation again.
The first two steps connect both domains in the zero outcome, but do not determine their relative scale. In the third and
final step, the size of the utility increment for gains in terms of the utility decrement for losses is determined by eliciting the
outcome d > x0 that makes the agent indifferent between the mixed prospects (r:d, y1) and (r:x0, y0). Under prospect theory,
indifference between these prospects implies:
Uðy0 Þ Uðy1 Þ ¼
2
wþ ðrÞ
ðUðdÞ Uðx0 ÞÞ
w ð1 rÞ
Interpolation requires that outcome b falls within the standard sequence, which is why we used x0 and x1 as outcomes for the second prospect.
ð3:6Þ
Author's personal copy
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
655
From a measurement perspective this equation amounts to relating the utility decrement of the standard sequence of
losses to the utility increment of the standard sequence of gains, as depicted by brace 3 in Fig. 3.1. The utility of outcome
d can again be approximated by using interpolation.
Eqs. (3.4)–(3.6), the fact that the standard sequences are equally spaced in utility units, and (linear) interpolation of the
utility of indifference outcomes b, c and d, determines the utilities of the outcomes {yn, . . . , y0, 0, x0, . . . , xn} up to scale. Setting, for example, U(yn) = 1 completely fixes the utility function such that it can be depicted graphically.
In the above steps, the probability weights corresponding to the probabilities used in the elicitation procedure were assumed to be known, while in fact they are unknown a priori. Several parameter-free techniques to obtain these probability
weights have been proposed in the literature (Abdellaoui, 2000; Bleichrodt & Pinto, 2000). Hence, if combined with these
measurement methods, the three indifferences stated above can in principle be used to measure the utilities of the standard
sequences of gains and losses on the same scale. In the present study, we do not use additional questions to obtain the probability weights, and will assume either linear probability weighting as in classical economic analyses or use the empirical
estimates of the probability weights found by Tversky and Kahneman (1992), and Abdellaoui (2000) in the analysis. A different method to measure loss aversion in a parameter-free way is in Abdellaoui, Bleichrodt, and Paraschiv (2007b).
4. The experiment: method
4.1. Participants
N = 1935 Dutch participated in the experiment which was held in February 2006. We used the DNB Household survey
which is a household panel that completes a questionnaire every week on the Internet or, if Internet is not available in
the household, by a special box connected to the television. The household panel is a representative sample of the Dutch
population; it consist of people from all income groups and age groups above 16 years (see Table B.1 for further details).
4.2. Procedure
Respondents first read experimental instructions (see Appendix A) and were then asked to answer a practice question to
familiarize them with the experimental setting. In the instructions it was emphasized that there were no right or wrong answers. In order to obtain indifference between prospects we used direct matching, that is, respondents were asked to report
an outcome of a prospect for which they would be indifferent between two particular prospects. There is evidence that
obtaining indifferences by using a bisection choice method (Abdellaoui, 2000) or by using a choice list (Holt & Laury,
2002; Tversky & Kahneman, 1992) yields more consistent results (Bostic, Herrnstein, & Luce, 1990; Luce, 2000). However,
these elicitation methods are time consuming and, hence, proved not to be tractable in this large-scale field experiment with
the general public.
Prospects were framed as depicted in Fig. 4.1. Respondents were directly asked to report the upper prize of the left prospect that would make them indifferent between both prospects. The wheel in the middle served to explain probabilities to
respondents. Both the probabilities reported in the wheel and the colours of the wheel corresponded to the probabilities of
the prospects.
The outcomes of the prospects used in our experiment were hypothetical because a large part of the experiment concerned substantial losses and, hence, real incentives could not be used for ethical reasons. In addition, implementing real
incentives was not feasible because substantial payoffs were used in the experiment, which we chose to do because utility
measurements are primarily of interest only for significant amounts of money since utility is often assumed to be linear for
moderate amounts of money (Abdellaoui, Bleichrodt, & l’Haridon, 2008; Rabin, 2000b; Savage, 1954, p. 60).
There have been extensive debates about whether offering real or hypothetical incentives yields better or more reliable
data. Camerer and Hogarth (1999) and Hertwig and Ortmann (2001) provide excellent summaries of the numerous studies
that have addressed the role of incentives in measurements of risk attitudes, often finding mixed results. In general, real
incentives do seem to reduce data variability (Camerer & Hogarth, 1999; Smith & Walker, 1993), and increase risk aversion
(e.g. Holt & Laury, 2002; Weber, Shafir, & Blais, 2004) compared to hypothetical incentives, although several studies have
suggested that the latter difference is small if tasks are cognitively easy, such as making direct choices between prospects
(Beattie & Loomes, 1997; Camerer, 1989; Camerer & Hogarth, 1999). Outcome matching as used in our experiment is possibly cognitively more demanding than direct choice, and Kachelmeier and Shehata (1992) provided evidence that real
incentives indeed do increase risk aversion in matching tasks. It is, however, unclear whether eliciting risk attitudes with
Fig. 4.1. The framing of the prospect pairs.
Author's personal copy
656
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
Table 4.1
The obtained indifferences.
Matching question
Prospect L
1
2
..
.
7
8
..
.
13
14*
15*
16*
(0.5:
(0.5:
..
.
(0.5:
(0.5:
..
.
(0.5:
(0.5:
(0.5:
(0.5:
a, 10)
x1, g)
x6, g)
y1, l)
y6, l)
b, 0)
0, c)
d, y1)
Prospect R
..
.
..
.
(0.5:
(0.5:
..
.
(0.5:
(0.5:
..
.
(0.5:
(0.5:
(0.5:
(0.5:
50, 20)
x0, G)
x5, G)
y0, L)
y5, L)
x1, x0)
y0, y1)
x0, y0)
Notes: underlined outcomes are the matching outcomes and questions marked with an asterisk were presented in randomized order.
real incentives using direct matching yields more accurate (‘‘true”) responses, i.e. responses that are better able to predict
behaviour in everyday life, because real incentives are often implemented by using artificial rewarding schemes at the cost
of realism and an increase in the degree of cognitive effort required from subjects (Read, 2005). For example, in Kachelmeier
and Shehata’s (1992) experiment, real incentives were implemented using the Becker–DeGroot–Marschak procedure (Becker, DeGroot, & Marschak, 1964), which has been shown to be prone to strategic behaviour by subjects (Braga & Starmer,
2005), and is fairly complex and not easy to understand (Halevy, 2007; Plott & Zeiler, 2005). Hence, the observed difference
in risk attitudes between real and hypothetical payoffs could also have been caused by a general misunderstanding of the
procedure or strategic behaviour by subjects. Also note that utilities obtained using the tradeoff method are invariant to
any bias affecting probability weighting.
4.3. Stimuli
For each respondent we obtained a total of 16 indifferences; see Table 4.1. Following the first practice question, matching
questions 2–7 served to obtain an increasing sequence of gains x0, . . . , x6 that are equally spaced in utility units, followed by
six matching questions to obtain a decreasing sequence of losses y0,. . ., y6 that are equally spaced in terms of utility (see Section 3.1). A potential drawback of the tradeoff method is that elicited answers are used as inputs in subsequent questions,
and, hence, response error might propagate over the questions posed. However, based on simulations, both Abdellaoui, Vossmann, and Weber (2005) and Bleichrodt and Pinto (2000) conclude that the effect of error propagation on their measurements of utility obtained using the tradeoff method was negligible. A theoretical argument for using the tradeoff method
to measure utilities in order to reduce the effect of error on the inferred utility function is in Blavatskyy (2006).
Matching questions 14–16 served to obtain a parameter-free measurement of the degree of loss aversion at the individual
level (see Section 3.2). As can be seen in Table 4.1, the parameter values of p and r used throughout Section 3 were set at 0.5,
as in Bleichrodt and Pinto’s (2000) experiment.
4.4. Treatments
In order to be able to test whether the degree of utility curvature and the degree of loss aversion are more pronounced for
larger monetary outcomes, respondents were randomly assigned to two different treatments. These treatments solely differed in the parameters values used for G, g, x0, L, l, and y0. In the low-stimuli treatment, these parameters values were
set at G = 64, g = 12, x0 = 100, L = 32, l = 6, and y0 = 50. In the high-stimuli treatment, all parameter values were scaled
up by a factor 10, i.e. the parameters values were set at G = 640, g = 120, x0 = 1000, L = 320, l = 60, and y0 = 500.
5. The experiment: results
This section presents both parametric and non-parametric results regarding the shape of the utility function for gains and
losses (Sections 5.1 and 5.2), and loss aversion (Section 5.3). Since the panel of respondents is a representative sample of the
Dutch population, the raw sample means would provide unbiased estimates of the population means if we would observe
answers from all individuals in the panel. Not surprisingly, there is non-response in the data which, if non-random, will bias
the statistical inferences. We deal with this sample selection problem by constructing sampling weights in Appendix B. In
what follows, reported coefficients are estimates of the population means, obtained by weighting each observation by the
inverse of the (predicted) probability of being included in the sample.3 These coefficients are unbiased estimates of the population means under the assumption that non-response is random after conditioning on the selection variables that we observe.
The weighting procedure only affects the estimates of loss aversion. Therefore, we will report both the estimated sample means
and the estimated population means for the loss aversion coefficients.
3
Table B.2 in Appendix B provides summary statistics on the unweighted data.
Author's personal copy
657
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
Table 5.2
Estimated population means utility curvature.
i
Gains
Losses
High (N = 383)
1
2
3
4
5
6
Low (N = 431)
High (N = 330)
Low (N = 360)
xi
xi – xi1
xi
xi – xi1
yi
|yi – yi1|
yi
|yi – yi1|
1984
(602)
2975
(1122)
4017
(1665)
5093
(2260)
6179
(2897)
7310
(3597)
984
(602)
991
(583)
1042**
(655)
1076**
(690)
1086
(740)
1131**
(820)
204
(93)
318
(192)
440
(339)
577
(624)
729
(953)
892
(1346)
104
(93)
114*
(136)
122***
(165)
137**
(305)
151**
(357)
164**
(434)
849
(231)
1242
(432)
1663
(641)
2073
(864)
2488
(1078)
2910
(1308)
349
(231)
393***
(244)
420***
(254)
410
(253)
415
(255)
422
(270)
87
(37)
126
(59)
168
(83)
211
(105)
255
(129)
299
(156)
37
(37)
39*
(31)
42**
(31)
43*
(29)
43
(30)
45
(32)
Notes: Estimated population averages in euros. Standard deviations in parenthesis.
Indicates that the predicted mean is significantly higher than its predecessor at the 10% level.
**
Indicates that the predicted mean is significantly higher than its predecessor at the 5% level.
***
Indicates that the predicted mean is significantly higher than its predecessor at the 1% level.
*
5.1. Utility curvature: Non-parametric analysis
Table 5.2 summarizes the results regarding the obtained utility function for monetary gains and losses under the different
treatments. As can be seen in the table, the difference between the successive elements of the average standard sequences is
mostly increasing for both gains and losses in both treatments.
We performed t-tests to test whether the differences between the successive elements of the standard sequence
for gains and losses are indeed significantly increasing.4 As can be seen in the table, a total of eight differences between the obtained successive elements of the standard sequence appear to be significantly increasing for gains. In the
loss domain there are five significant differences. Only one mean difference is lower than its predecessor, but this difference is not significant. Overall, these results thus imply concave utility for gains and convex utility for losses, reflecting diminishing sensitivity: people are more sensitive to changes in outcomes near the status quo than to changes in
outcomes remote from the status quo, as predicted by prospect theory but contrary to the classical prediction of universal concavity.
We also classified the utility function of each respondent by calculating the area under the normalized utility function for
both gains and losses. A respondent’s utility function was classified as concave (convex; linear) if this area was larger than
(smaller than; equal to) 0.5 for gains. For losses, a utility function was classified as concave (convex; linear) if the area was
smaller than (larger than; equal to) 0.5. Table 5.3 gives the resulting estimated population fractions. There appears to be considerable variability in the elicited shapes of utility, with a significant amount of observations in all cells.
The predominant pattern is again a concave–convex shape of the utility function, implying diminishing sensitivity in both
domains. If we consider unconditional fractions, the case for diminishing sensitivity is stronger, with 42% of respondents
being classified as concave for gains and 47% as convex for losses.
5.2. Utility curvature: Parametric analysis
Although we obtained utilities in a parameter-free way using the tradeoff method, we also used parametric methods to
analyze the data. For each respondent, we estimated the power utility function with parameter q for both treatments. Thus,
for each respondent and for each separate domain (gains and losses) we estimated:
UðxÞ ¼ xq
for q > 0
ð5:1Þ
UðxÞ ¼ lnðxÞ
for q ¼ 0
ð5:2Þ
UðxÞ ¼ xq
for q < 0
ð5:3Þ
by minimizing the sum of squared residuals. The power utility function with parameter q is the most popular parametric
family for fitting utility (Wakker, 2008) and is also known as the family of constant relative risk aversion (CRRA) because
the ratio xU00 (x)/U0 (x), i.e. the index of relative risk aversion, is constant and equal to 1 q.5 The average individual
parametric estimate of the power coefficient q in the population for gains was 0.95 for the high-, and 0.94 for the low-stimulus
4
Non-parametric tests on the raw data yields similar results.
We also estimated the exponential (CARA) utility function and the expo-power utility function (Abdellaoui et al., 2007a) for both gains and losses, yielding
similar results.
5
Author's personal copy
658
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
Table 5.3
Classification of respondents.
Losses
Gains
Convex
Linear
Concave
Total
Concave (%)
Linear (%)
Convex (%)
Total (%)
8.95
24.90
8.61
22.47
5.22
17.54*
7.68
30.44
14.88
6.78
25.42*
47.09
29.06
29.23
41.72
100
*
A test of equality of probability is rejected in favour of the hypothesis that the concave–convex shape is more prevalent than the linear–linear shape (onesided test, z = 2.84, p-value = 0.002).
Table 5.4
Regression results.
Variable
q+
q
k
Low stimuli
0.032
(0.042)
0.052
(0.040)
0.003*
(0.001)
0.044
(0.038)
0.004
(0.018)
0.880***
(0.068)
811
0.841
(0.593)
0.036
(0.603)
0.004
(0.011)
0.292
(0.597)
0.305
(0.258)
0.263
(0.851)
686
0.027
(0.200)
0.415**
(0.213)
0.007
(0.007)
0.489***
(0.172)
0.015
(0.085)
1.551***
(0.438)
437
Female
Age
High education
Income/1000
Constant
N
Notes: weighted linear regression. Standard errors allow for clustering within households.
*
Significant at the 10% level.
**
Significant at the 5% level.
***
Significant at the 1% level.
treatment. This result is consistent with a mean estimate for gains of 0.91 found by Abdellaoui et al. (2007a), based on a parameter-free analysis. A two-sided t-test, adjusted for the sampling weights (as are all t-tests reported in this and the following sections), does not reject the null-hypothesis that the estimated mean q-parameters for gains are equal across the high- and the
low-stimuli treatment (t = 0.539, p-value = 0.593).
Analysis of the individual q-parameters on the basis of one-sided t-tests does indicate that the q-coefficients for gains are
significantly lower than 1 in both the high- and the low-stimulus treatment (low: t = 3.04, p-value = 0.001; high:
t = 2.105, p-value = 0.018), which implies a significant overall degree of diminishing marginal utility for gains. For losses,
the obtained parameter estimates are 0.92 for the high-stimulus treatment and 0.93 for the low-stimulus treatment respectively, and we cannot reject the hypothesis that they are equal (t = 0.100, p-value = 0.918). In addition, the estimated q-coefficients for losses proved to be significantly lower than 1 in both the high and the low-stimulus treatment (low: t = 2.414,
p-value = 0.008; high: t = 2.256, p-value = 0.012).
An overall two-sided Wilcoxon rank sum test on the raw data does reject the null-hypothesis that the ranks of the obtained q-coefficients are equal between the gain and the loss domain in favour of the hypothesis that the q-coefficients
are higher in the gain domain compared to the loss domain (z = 2.159, p-value = 0.031). Thus, for our estimation sample,
we have obtained evidence suggesting that utility is concave for gains but is even more convex for losses. However, the standard errors are higher for the population and, therefore, we cannot reject the hypothesis that the degree of utility curvature
for gains is similar to that of losses (t = 1.315, p-value = 0.189).
The dataset allows us to relate the degree of utility curvature to socio-demographic variables. Therefore, we performed
a weighted linear regression analysis with the individual estimates as dependent variables and a treatment dummy and
several socio-demographic variables as explanatory variables.6 By including these variables simultaneously, we can interpret the coefficients as being partial correlations conditional on the other covariates. The results of this regression are reported in the first three columns of Table 5.4 (the fourth column will be explained in Section 6). As can be seen in the
table, the estimated coefficient for gains (q+) is (marginally) significantly lower for older respondents, but other socio-demographic variables such as income, age, and education do not appear to have a significant effect on the degree of utility curvature for both gains and losses. The same holds for variables concerning profession, affinity with financial matters, being a
house owner, and religion.
6
We used several alternative non-linear specifications, but this did not change the results.
Author's personal copy
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
659
Table 5.5
Mean results loss aversion.
Sample
b
c
d
kEU
kPT
kAB
Population
High
N = 210
Low
N = 229
High
Low
4016
(111)
1569
(42)
1842
(57)
1.69
(0.11)
1.79
(0.12)
1.88
(0.12)
387
(10.00)
157
(3.90)
180
(5.80)
1.64
(0.09)
1.74
(0.10)
1.83
(0.10)
3970
(51)
1572
(20)
1833
(27)
1.73
(0.06)
1.84
(0.06)
1.93
(0.06)
391
(4.72)
163
(1.98)
184
(2.70)
1.79
(0.05)
1.90
(0.05)
2.00
(0.05)
Note: standard errors in parenthesis.
5.3. Loss aversion
Because a commonly accepted definition of loss aversion does not exist in the literature (Abdellaoui, Bleichrodt, & Paraschiv, 2007b), we define the loss aversion coefficient k as the ratio between the steepness of the utility function for gains on
the interval [0, x0], and the steepness of the utility function for losses on the interval [y0, 0], that is:
k¼
Uðy0 Þ x0
Uðx0 Þ y0
ð5:4Þ
This definition best approximates the definition of loss aversion suggested by Benartzi and Thaler (1995) and formalized
by Köbberling and Wakker (2005), as well as the definition of loss aversion used by Tversky and Kahneman (1992).7 The main
advantage of these ‘‘local” definitions of loss aversion compared to ‘‘global” definitions of loss aversion, such as the definitions
proposed by Bowman, Minehart, and Rabin (1999), Kahneman and Tversky (1979) and Wakker and Tversky (1993), is that they
capture attitudes toward losses in a single number describing the relative degree of steepness of the utility function for losses
compared to the steepness of the utility function for gains, irrespective of the degree of utility curvature for gains and losses.8
Hence, measurements of loss aversion using a local definition of loss aversion do not depend on the utility of the outcomes used
in the measurement process, and attitudes towards losses can be unambiguously separated from utility curvature (Abdellaoui,
Bleichrodt, & Paraschiv, 2007b).
Table 5.5 presents the summary statistics and estimated population means for the different indifference values of outcomes b, c, and d, and the resulting loss aversion coefficient based on various assumptions concerning subjective probability
weighting.9 More specifically, loss aversion is estimated assuming the classical expected utility model (w(0.5) = w+(0.5) = 0.5),
prospect theory (w(0.5) = 0.4540 and w+(0.5) = 0.4206), and using the values of w(0.5) and w+(0.5) found by Abdellaoui
(2000), being w(0.5) = 0.457, and w+(0.5) = 0.394. The latter values are used because they yield the largest difference between
w(0.5), and w+(0.5) found in the literature. As a result, our estimate of loss aversion based on these values yields an upper
bound, while our estimate of loss aversion under the classical expected utility model yields a lower bound of the estimated loss
aversion coefficient.
As can be seen in Table 5.5, the estimated population mean value of k under expected utility, denoted by kEU, is 1.79 for
the low- and 1.73 for the high-stimuli treatment, while these mean estimates based on the findings of Abdellaoui (2000),
denoted by kAB, are 2.00 and 1.93, respectively. Table 5.5 also shows that there are significant differences in the loss aversion
coefficient between the weighted and non-weighted estimates. This is caused by the fact that lower educated individuals are
underrepresented in our sample, while they are significantly more loss averse on average.
Under prospect theory, the estimated population mean value of k, denoted by kPT, is equal to 1.84 for the high-stimuli
treatment and 1.90 for the low-stimuli treatment. The 95% confidence intervals of the population loss aversion parameters
are [1.72, 1.95] for the high-stimuli, and [1.80, 1.99] for the low-stimuli treatment respectively. This implies that there is
significant evidence of loss averse behaviour in the population, i.e. the loss aversion coefficient is significantly higher than
7
Köbberling and Wakker (2005) defined loss aversion as the ratio between the left and the right derivate of the utility function at the reference point, i.e.
U0 "(0)/U0 ;(0), while Tversky and Kahneman (1992) implicitly used the negative of the ratio between the utility of a loss of one dollar and a gain of one dollar, i.e.
U($1)/U($1), as definition of loss aversion.
8
Kahneman and Tversky (1979) implicitly defined loss aversion as the mean or median of U(x)/U(x) for positive values of x. Wakker and Tversky (1993)
implicitly defined the loss aversion coefficient as k = U0 (x)/U0 (x) for positive values of x, while Bowman et al. (1999) defined loss aversion as the requirement
that U0 (x) P U0 (z), for positive values of x and z, which could be translated into a loss aversion coefficient of k = inf U0 (x)/sup U0 (z).
9
^ bÞ,
^ where the hat variables are approximated
Given the values b, c and d, the loss aversion coefficient can be written as: k ¼ ðð1 wþ ðrÞ=ð1 w ðrÞÞð^cd=
utilities. Hence, our estimate of loss aversion depends proportionally on (1 w+(½))/(1 w(½)).
Author's personal copy
660
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
1. Generally, these results suggests that people are loss averse and weight a particular loss about 1.87 times as heavily as a
commensurable gain, with an average upper bound of 1.97, and an average lower bound of 1.76. In what follows, we will
mainly focus on the loss aversion coefficient implied by Tversky and Kahneman’s (1992) original estimates of probability
weighting, i.e. on kPT.
The mean results also suggest that the degree of loss aversion is decreasing over the treatments. The standard errors are
too high to draw any strong conclusions, however, i.e. we cannot reject the hypothesis that the loss aversion parameter is the
same across treatments (t = 0.321, p-value = 0.747). More interestingly, if we regress the obtained measurement of loss aversion on socio-demographic characteristics, we find that females are significantly more loss averse than males as the final
column of Table 5.4 shows. Conditional on other covariates (e.g. income), females weight losses about 0.415 more heavily
than males.10 The difference in the unconditional population means is about the same, with an estimated loss aversion coefficient of 2.10 for women, and 1.66 for men. In addition, higher educated individuals, defined as having a higher vocational education or better, appear to have a significant lower degree of loss aversion.
6. Discussion of results
This section provides a discussion of the experimental results and relates our findings to previous findings in the literature. There are four possible confounding factors in previous studies that are not present in the current study, and that may
explain differences in the findings. First and foremost, classical studies assume expected utility and, thus, ignore the important role of probability weighting and loss aversion in risk attitudes (e.g. Andersen et al., 2008; Barsky et al., 1997). Second, in
the few studies that have used prospect theory to analyze risk attitudes for the general population (e.g. Donkers et al., 2001;
Tu, 2005) the functional form of the utility (and probability weighting-) function are assumed beforehand and, therefore, the
estimations depend critically on the appropriateness of the assumed functional form: conclusions drawn on the basis of the
parameter estimates need no longer be valid if the true functional form differs from the assumed functional form. Third,
most studies use aggregate data to estimate the different assumed functional forms, ruling out heterogeneity of individual
preferences. Finally, student populations are commonly used as subjects in experiments measuring risk attitudes, making
the external validity of the results questionable.
6.1. Utility curvature
Parametric studies that provide measurements and decompositions of risk attitudes into a utility component and a probability weighting component are numerous and, consequently, there is an ongoing debate about the shape of the utility function. For gains, most studies have corroborated that the utility function is concave-shaped, reflecting the natural intuition
that each new euro brings less utility than the euro before and implying diminishing marginal utility (Abdellaoui, 2000;
Wakker & Deneffe, 1996). The results of our study confirm this finding; the mean power of the utility function for gains
is 0.94. If we compare our findings with other measurements of risk attitudes using large representative datasets, we find
that our estimated relative risk coefficient for gains of 0.06 (=1 0.94) is relatively small. For example, Andersen et al.
(2008) found a mean risk aversion coefficient of 0.74, and Barsky et al. (1997) found a mean risk tolerance (the reciprocal
of the constant relative risk coefficient) of 0.24 which translates into a mean relative risk coefficient of 4.16. The smallest
degree of relative risk aversion coefficient found by Hartog et al. (2002) was 20. For the domain of losses, we found that sensitivity to outcomes diminishes (i.e. the utility function is convex), which is in line with the predictions of prospect theory,
but contradicts the classical prediction of universal concavity. Although the majority of studies have found similar results for
losses (Abdellaoui, 2000; Abdellaoui et al., 2007b; Etchart-Vincent, 2004; Fennema & van Assen, 1998; Tversky & Kahneman,
1992), some studies have found the opposite result (Abdellaoui et al., 2008; Davidson, Suppes, & Siegel, 1957).
Our finding that there is no significant gender effect concerning the degree of utility curvature confirms the findings of
Fehr-Duda et al. (2006), who also do not find gender differences in the utility functions for both gains and losses based on
parametric fittings using a student subject pool. However, this result contradicts the conclusion of many classical studies
that ascribe gender differences in risky attitudes solely to differences in the degree of utility curvature (Barsky et al.,
1997; Cohen & Einav, 2007; Hartog et al., 2002).
All in all, our results show that utility for money is concave for gains and convex for losses, supporting the presence of a
reflection effect (i.e. risk attitudes for gains are the mirror image of risk attitudes for losses) as predicted by prospect theory,
but contradicting the classical prediction of universal concavity. In addition, our results give empirical support to Rabin’s
(2000b, p. 202, Chap. 11) conjecture that diminishing marginal utility is an ‘‘implausible explanation for appreciable risk
aversion, except when the stakes are very large:” utility curvature is less pronounced than suggested by classical utility measurements. This suggests that psychological phenomena such as probability weighting and loss aversion are valid outside the
laboratory, that is, our results support the external validity of subjective probability weighting and loss aversion and cautions against assuming expected utility when measuring risk attitudes outside the laboratory.
10
It could be argued that this is the case because females and males weight probabilities differently as the study by Fehr-Duda, de Gennaro, and Schubert
(2006) suggests. Fehr-Duda et al. (2006) report that women underweight (overweight) w+(0.5) (w–(0.5)) more than men, which implies that our obtained
gender difference in loss aversion would become even stronger if we allow probabilities to be weighted in a nonlinear way.
Author's personal copy
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
661
These findings are crucial from a policy perspective because choice behaviour based on diminishing marginal utility of
money is considered to be rational by most researchers in the sense that these choices satisfy the fundamental axioms of
expected utility (in particular the independence axiom) whereas choice behaviour based on subjective probability weighting
for example, does not agree with these normatively compelling axioms (Bleichrodt, Pinto, & Wakker, 2001; Wakker, 2005).
Hence, it could be argued that policy recommendations should be based solely on measurements of the amount of utility
that people derive from money obtained in isolation, i.e. in a choice environment where confounding effects such as probability weighting did not play a role. This study provides such a measurement using a large and representative dataset.
6.2. Loss aversion
Our estimated mean loss aversion coefficients of 1.84 for the high-stimuli, and 1.90 for the low-stimuli treatment under
prospect theory are lower than the parametric estimate of k = 2.25 obtained by Tversky and Kahneman (1992), and the nonparametric estimate of k = 2.04 based on the definition of loss aversion proposed by Kahneman and Tversky (1979), found by
Abdellaoui et al. (2007b). On the other hand, our mean estimates of k under prospect theory are higher than the average estimated coefficient of 1.43 obtained by Schmidt and Traub (2002) and 1.38 obtained by Harrison and Rutström (2009), while
they are quite consistent with data collected in the field by Gächter, Johnson, and Herrmann (2007). In the latter study, loss
aversion is measured in a riskless setting using WTA and WTP disparities of a large sample of car buyers, resulting in an average (within-subject) loss aversion coefficient of 1.95.
The finding that females and lower educated respondents are significantly more loss averse than males is in agreement
with findings in the laboratory (Brooks & Zank, 2005; Schmidt & Traub, 2002). This suggests that, contrary to classical studies
that ascribe gender differences in risky behaviour solely to differences in the degree of utility curvature (Hartog et al., 2002;
Pälsson, 1996), this phenomenon is primarily driven by loss aversion. Interestingly, our finding that lower educated respondents are more loss averse suggests that measurements of loss aversion based on (highly educated) student samples lead to
an underestimation of the loss aversion coefficient.
When conditioning on other covariates Gächter et al. (2007) did not find a significant gender difference in the degree of
loss aversion, but did find strong effects of age, income and level of education. We do not find a strong effect of either age or
income, which may be caused by differences in the sample compositions, the correlation between the covariates, or the elicitation context (riskless versus a risky setting). The fact that the direction of all the effects is the same in both studies suggests that loss aversion is a personal trait, but further research into these associations is needed, especially if we want to
draw conclusions on the causal effect of education and income on loss aversion and vice versa.
6.3. Gains versus losses
A further point of debate in the literature is whether utility curvature for losses is either more or less pronounced compared to utility curvature for gains. More pronounced convexity for losses than for gains was found by Fishburn and Kochenberger (1979), virtually no difference in the degree of utility curvature was found by Abdellaoui et al. (2007b), and Schunk
and Betsch (2006), whereas less pronounced convexity for losses than concavity for gains has been reported by Abdellaoui
et al. (2005), Fennema and van Assen (1998) and Wakker, Köbberling, and Schwieren (2007). Although respondents in the
sample were significantly less risk averse in the gain domain, we could not draw such a strong conclusion for the population,
supporting the findings of Abdellaoui et al. (2007b), and Schunk and Betsch (2006).
6.4. Treatment effects
Finally, there is also no consensus on whether the degree of utility curvature and the degree of loss aversion are more (or
less) pronounced for larger outcomes. For direct choice between prospects, Holt and Laury (2002) found that decision makers
become more risk averse the larger the amounts at stake, but only when payoffs are real. However, subsequent studies have
shown that this effect, called the magnitude effect, was partially driven by order effects (Harrison, Johnson, McInnes, & Rutström, 2005; Holt & Laury, 2005). The effect can also be explained by differences in the amount of cognitive effort exerted by
subjects between the treatments, either because the hypothetical nature of the hypothetical task was highlighted in the
instructions (Read, 2005), or because the task itself was surrounded by real tasks, making the contrast between the real
and hypothetical treatments particularly salient. Indeed, the magnitude effect has been replicated using hypothetical payoffs
(Kühberger, Schulte-Mecklenbeck, & Perner, 2002), while Friend and Blume (1975) and Blake (1996) have found the opposite
effect, i.e. a decrease in the relative risk aversion coefficient when the outcomes are increased, when analyzing portfolio
holdings of investors. We did not find significant evidence for the existence of a magnitude effect in our measurements of
utility curvature. Besides the aforementioned possible explanations for this difference in findings (i.e. differences in subject
populations, measurements methods, and the assumed model), it is worth mentioning that the high-payoff conditions of the
studies by Holt and Laury (2002) and Kühberger et al. (2002) involved payoffs between $50 and $350, comparable with our
low-stimuli treatment.
For loss aversion, Abdellaoui et al. (2007b) found that the degree of loss aversion decreased with the size of the gains and
losses involved, i.e. they found a magnitude effect for loss aversion. However, this magnitude effect was found under the
Author's personal copy
662
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
global definition of loss aversion proposed by Kahneman and Tversky (1979) only, which can explain why we did not find a
similar effect; we used a local (‘‘utility-free”) definition of loss aversion in our analysis.
7. Conclusion
We have obtained parameter-free measurements of the utility component as well as the loss aversion component of risk
attitudes using a representative sample from the Dutch population. The results suggest that utility is concave for gains and
convex for losses, implying diminishing sensitivity towards outcomes, as predicted by prospect theory. In addition, our results suggest that classical utility measurements overestimate concavity, which can be explained by the ignorance of probability weighting and loss aversion in these measurements. In addition, we have obtained parameter-free measurements of
loss aversion. The results show that respondents were significantly loss averse and weighted a loss about 1.87 times as much
as a commensurable gain. Interestingly, we have found evidence that males and higher educated persons are significantly
less loss averse. The former result suggests that gender differences in risk attitudes are primarily driven by loss aversion
and not by utility curvature as suggested by previous studies that assume the classical expected utility model. The latter result suggests that measurements of loss aversion based on relatively highly educated student samples lead to an underestimation of the loss aversion coefficient. Finally, we found that both the degree of utility curvature and the degree of loss
aversion are not altered by scaling up the monetary outcomes involved.
Acknowledgments
We thank Peter P. Wakker, Stefan T. Trautmann, B.M.S. van Praag, and an anonymous referee for useful comments, and
CentER data, in particular Vera Toepoel, for programming the experiment and providing the data.
Appendix A. Experimental instructions
(Instructions have been translated from Dutch to English and will appear online after publication)
Welcome at this experiment on individual decision making. The experiment is about your risk attitude. Some people like
to take risks while other people like to avoid risks. The goal of this experiment is gain additional insight into the risk attitude
of people living in the Netherlands. This is very important for both scientists and policymakers. If we get a better understanding of how people react to situations involving risk, policy can be adjusted to take this into account (for example with information provision on insurance and pensions, and advice for saving and investment decisions). Your cooperation at this
experiment is thus very important and is highly appreciated.
The questions that will be posed to you during this experiment will not be easy. We therefore ask you to read the following explanation attentively. In this experiment, there are no right or wrong answers. It is exclusively about your own preferences. In those we are interested.
Probabilities (expressed in percentages) play an important role in this experiment. Probabilities indicate the likelihood of
certain events. For example, you probably have once heard Erwin Krol [well known Dutch weatherman] say that the probability that it will rain tomorrow is equal to 20 percent (20%). He then means, that rain will fall on 20 out of 100 days with
the same weather condition. During this experiment, probabilities will be illustrated using a wheel, as depicted below.
Suppose that the wheel depicted in the picture above is a wheel consisting of 100 equal parts. Possibly you have seen such a
wheel before in television shows such as The Wheel of Fortune. Now imagine that 25 out of 100 parts of the wheel are orange
and that 75 out of 100 parts are blue. The probability that the black indicator on the top of the wheel points at an orange part
after spinning the wheel is equal to 25% in that case. Similarly, the probability that the black indicator points at a blue part
after spinning the wheel is equal to 75%, because 75 out of 100 parts of the wheel are blue. The size of the area of a colour on
the wheel thus determines the probability that the black indicator will end on a part with that colour.
Besides probabilities, lotteries play an important role in this experiment. Perhaps you have participated in a lottery such
as the National Postal Code Lottery yourself before. In this experiment, lotteries yield monetary prizes with certain probabilities, similar to the National Postal Code Lottery. However, the prizes of the lotteries in this experiment can also be negative. If a lottery yields a negative prize, you should imagine yourself that you will have to pay the amount of money. In the
Author's personal copy
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
663
following explanation we will call a negative prize a loss and a positive prize a gain. During this experiment, lotteries will be
presented like the example presented below:
In this case, the lottery yields a gain of 1000 € with probability 50%. However, with probability 50%, this lottery yields a loss
of 200 €. You should image that if you participated in this lottery, you would get 1000 € with probability 50%, and with probability 50% you would have to pay 200 €.
During this experiment you will see two lotteries, named Lottery L (Left) and Lottery R (right), on the top of each page.
Between these lotteries you will see a wheel that serves as an aid to illustrate the probabilities used. You will see an example
of the layout of the screen on the next page.
In this example, Lottery R yields a gain of 500 € with probability 50% and with probability 50% it yields a loss of 200 €. You
should imagine that, if we would spin the wheel once and the black indicator would point at the orange part of the wheel,
Lottery R would yield a gain of 500 €. However, if the black marker would point at the blue part of the wheel, Lottery R would
yield a loss of 200 €.
Similarly, Lottery L yields a loss of 300 € with probability 50%. However, as you can see, the upper prize of Lottery L is
missing. During this experiment, we will repeatedly ask you for the upper prize of Lottery L (in €) that makes Lottery L
and Lottery R equally good or bad for you. Thus, we will ask you for the upper prize of Lottery L for which you value both
lotteries equally.
You could imagine that most people prefer Lottery L if the upper prize of Lottery L is very high, say 3000 €. However, if this
prize is not so high, say 500 €, most people would prefer Lottery R. Somewhere between these two prizes there is a ‘‘turnover
point” for which you value both lotteries equally. For high prizes you will prefer Lottery L and for low prizes you will prefer
Lottery R. The turnover point is different for everybody and is determined by your own feeling. To help you a little bit in the
choice process, we will report the range of prizes in which the answer of most people lies approximately at each question.
How this works precisely will become clear in the practice question that will start if you click on the CONTINUE button
below. If something it not clear to you, you can read the explanation of this experiment again by pressing the BACK button
below.
A.1. Practice question
The practice question is now over. The questions you will encounter during this experiment are very similar to the practice question. If you click on the BEGIN button below, the experiment will start. If you want to go through the explanation of
this experiment again, click on the EXPLANATION button. Good luck.
Appendix B. Sample selection
Because there were no financial incentives to participate in the experiment, we expected that some members of the panel
would not respond, or would not provide answers to all questions. About a fifth of the panel members did not start the
Author's personal copy
664
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
experiment or stopped answering the questions before reaching the end of the experiment. This non-response is the first of
four types of sample selection that we observed in the dataset. The second type of sample selection is the result of an imposed monotonicity condition: the obtained standard sequence of gains (losses) had to be strictly increasing (decreasing),
which precludes violations of stochastic dominance. This is a natural requirement, since individuals that violate it either
have a lack of understanding or make a mistake, indicating a lack of attention. We found that in our sample about 40% of
the subjects displayed at least one inconsistency. Also, we dropped extreme responses that would solely have a significant
effect on the (mean) results. These outliers were defined as observations that belonged to the top or bottom percentile of the
distribution of outcomes for individuals who answered all questions (including the inconsistent observations). This amounts
to about 1% of the observations. We chose this strategy over the use of medians, because correcting the mean estimator for
sample selection is straightforward. Finally, the last type of sample selection involves the loss aversion questions. We imposed the condition that the values of outcomes b, c, and d, had to lie in the interval of the obtained standard sequence
of the individual, and, hence, dropped 12 observations. The absolute frequencies of the classification of the sample selection
mechanism are reported at the bottom of Table B.1. The samples for gains, losses and loss aversion differ because different
consistency conditions have been imposed, as already mentioned.
Table B.1
Sample selection probit and classification.
Variable
Fraction (%)
Gains
Losses
Loss aversion
Low stimuli treatment
50
Female
46
0.144**
(0.058)
0.052
(0.060)
0.101*
(0.059)
0.079
(0.061)
0.088
(0.065)
0.116*
(0.067)
Primary school (Ref. cat.)
5
Lower secondary education
26
Higher secondary education
14
Intermediate vocational training
19
Higher vocational training
25
Academic education
11
0.127
(0.142)
0.335**
(0.151)
0.046
(0.146)
0.258*
(0.143)
0.488***
(0.158)
0.029
(0.143)
0.148
(0.152)
0.148
(0.148)
0.039
(0.144)
0.305*
(0.158)
0.218
(0.177)
0.507***
(0.182)
0.147
(0.180)
0.394**
(0.175)
0.681***
(0.187)
Age 16–34 (Ref. cat.)
23
Age 35–44
18
Age 45–54
22
Age 55–64
18
Age 65+
19
€ 1.150 6 Income < € 1.800
25
€ 1.800 6 Income < € 2.600
31
Income P € 2.600
35
Catholic
30
Protestant
20
0.157*
(0.092)
0.234***
(0.088)
0.313***
(0.094)
0.462***
(0.095)
0.196*
(0.118)
0.200*
(0.115)
0.344***
(0.115)
0.014
(0.068)
0.160**
(0.077)
0.503***
(0.176)
0.202**
(0.093)
0.281***
(0.089)
0.340***
(0.096)
0.428***
(0.096)
0.269**
(0.122)
0.253**
(0.119)
0.411***
(0.119)
0.008
(0.069)
0.126
(0.078)
0.510***
(0.179)
0.284***
(0.099)
0.325***
(0.095)
0.524***
(0.106)
0.632***
(0.108)
0.319**
(0.146)
0.381***
(0.143)
0.526***
(0.142)
0.007
(0.077)
0.215**
(0.084)
1.199***
(0.217)
1935
375
728
1935
422
811
18
814
1276.243
12
690
1226.928
1935
1361 123
12
1
438
978.2592
Constant
N
Non-response
Non-monotone
Outside interpolation int.
Outlier
Sample
Log-likelihood
Notes: Standard errors allow for clustering within households.
Significant at the 10% level. : Combines non-response (2 cases) with not being in the union of the sample for gains and losses (1359 cases).
**
Significant at the 5% level. : Combines non-response (2 cases) with not being in the union of the sample for gains and losses (1359 cases).
***
Significant at the 1% level. : Combines non-response (2 cases) with not being in the union of the sample for gains and losses (1359 cases).
*
Author's personal copy
665
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
Table B.2
Sample Mean Results Utility Curvature.
i
Gains
Losses
High (N = 383)
1
2
3
4
5
6
Low (N = 431)
High (N = 330)
Low (N = 360)
xi
xi – xi1
xi
xi – xi1
yi
|yi yi1|
yi
|yi – yi1|
1993
(602)
3000
(1131)
4060
(1692)
5161
(2311)
6283
(2980)
7447
(3713)
993
205
(94)
319
(184)
441
(313)
576
(561)
727
(865)
893
(1244)
105
851
(231)
1243
(431)
1664
(634)
2075
(856)
2494
(1069)
2920
(1297)
350
86
(36)
126
(59)
168
(83)
211
(106)
254
(130)
298
(156)
36
1007
1060***
1101**
1122**
1164**
114***
122***
135*
151***
166
392***
421***
411
419
426*
40***
42***
43**
43
44
Notes: standard deviations in parenthesis.
Significantly higher than its predecessor at the 10% level (Wilcoxon signed-rank test).
**
Significantly higher than its predecessor at the 5% level (Wilcoxon signed-rank test).
***
Significantly higher than its predecessor at the 1% level (Wilcoxon signed-rank test).
*
Selection with respect to education (and other variables) may introduce a bias in the estimation if the outcome variable of
interest is related to it. Then information on the selection mechanism is needed to conduct proper statistical inference. Since
we have background information on both in- and out-of-sample respondents we can check whether the answers of both
groups of respondents differ systematically, and construct sampling weights by calculating the inverse of the probability
of being observed in the sample. Table B.1 presents the results of probit regressions of appearance in the respective sub samples with respect to gender, age, education, income and religion. All variables have been split up in categories such that the
results cannot be driven by assumptions regarding the functional form.
As can be seen in the table, there appears to be selection on most variables. Since the loss aversion equation compounds
the selection effects of the gain and the loss domains, it shows the most prominent effects. Generally, the probability that a
given respondent is observed in the sample is increasing in income and education level, and decreasing in age. Women are
less likely to answer our loss aversion questions, while individuals with a protestant background have a higher inclination to
respond. These effects are consistent with the findings of Von Gaudecker, van Soest, and Wengstrom (2008) who specifically
investigate sample selection in the same internet panel using risk aversion (choice) tasks. Thus, respondents from the population appear to have more difficulty conducting the risk aversion choice tasks than the typical subject in the lab.
Finally, we should note that since each question poses a new test for monotonicity, a random error model would predict
the exclusion of a higher fraction of observations as the amount of questions rises.
References
Abdellaoui, M. (2000). Parameter-free elicitation of utilities and probability weighting functions. Management Science, 46, 1497–1512.
Abdellaoui, M., Barrios, C., & Wakker, P. P. (2007a). Reconciling introspective utility with revealed preference. experimental arguments based on prospect
theory. Journal of Econometrics, 138, 356–378.
Abdellaoui, M., Bleichrodt, H., & l’Haridon, O. (2008). A tractable method to measure utility and loss aversion under prospect theory. Journal of Risk and
Uncertainty, 36, 245–266.
Abdellaoui, M., Bleichrodt, A., & Paraschiv, C. (2007b). Loss aversion under prospect theory: A parameter-free measurement. Management Science, 53,
1659–1674.
Abdellaoui, M., Vossmann, F., & Weber, M. (2005). Choice-based elicitation and decomposition of decision weights for gains and losses under uncertainty.
Management Science, 51, 1384–1399.
Allais, M. (1953). Le Comportement de l’Homme Rationnel devant le Risque: Critique des Postulats et Axiomes de l’Ecole Américaine. Econometrica, 21,
503–546.
Andersen, S., Harrison, G., Lau, M., & Rutström, E. (2008). Eliciting risk and time preferences. Econometrica, 76, 583–618.
Barsky, R. B., Juster, F. T., Kimball, M. S., & Shapiro, M. (1997). Preference parameters and behavioral heterogeneity: An experimental approach in the health
and retirement study. Quarterly Journal of Economics, 112, 537–579.
Beattie, J., & Loomes, G. (1997). The impact of incentives upon risky choice experiments. Journal of Risk and Uncertainty, 14, 155–168.
Becker, G. M., DeGroot, M., & Marschak, J. (1964). Measuring utility by a single-response sequential method. Behavioral Science, 9, 226–232.
Benartzi, S., & Thaler, R. (1995). Myopic loss aversion and the equity premium puzzle. Quarterly Journal of Economics, 110, 73–92.
Blake, D. (1996). Efficiency, risk aversion and portfolio insurance. An analysis of financial asset Portfolios Held by investors in the United Kingdom. Economic
Journal, 106, 1175–1192.
Blavatskyy, P. R. (2006). Error propagation in the elicitation of utility and probability weighting functions. Theory and Decision, 60, 315–334.
Bleichrodt, H., & Pinto, J. L. (2000). A parameter-free elicitation of the probability weighting function in medical decision analysis. Management Science, 46,
1485–1496.
Bleichrodt, H., Pinto, J. L., & Wakker, P. P. (2001). Making descriptive use of prospect theory to improve the prescriptive use of expected utility. Management
Science, 47, 1498–1514.
Bostic, R., Herrnstein, R. J., & Luce, R. D. (1990). The effect on the preference-reversal phenomenon of using choice indifferences. Journal of Economic
Behaviour and Organization, 13, 193–212.
Bowman, D., Minehart, D., & Rabin, M. (1999). Loss aversion in a consumption-savings model. Journal of Economic Behavior and Organization, 38, 155–178.
Author's personal copy
666
A.S. Booij, G. van de Kuilen / Journal of Economic Psychology 30 (2009) 651–666
Braga, J., & Starmer, C. (2005). Preference anomalies, preference elicitation, and the discovered preference hypothesis. Environmental and Resource
Economics, 32, 55–89.
Brooks, P., & Zank, H. (2005). Loss adverse behavior. Journal of Risk and Uncertainty, 31, 301–325.
Camerer, C. (1989). An experimental test of several generalized utility theories. Journal of Risk and Uncertainty, 2, 61–104.
Camerer, C., Babcock, L., Loewenstein, G., & Thaler, R. (1997). Labor supply of New York City Cabdrivers: One day at a time. Quarterly Journal of Economics,
112, 407–442.
Camerer, C., & Hogarth, R. M. (1999). The effects of financial incentives in experiments: A Review and capital-labour-production framework. Journal of Risk
and Uncertainty, 19, 7–42.
Cohen, A., & Einav, L. (2007). Estimating risk preferences from deductible choice. American Economic Review, 97, 745–788.
Cox, J. C., & Sadiraj, V. (2006). Small- and large-stakes risk aversion: Implications of concavity calibration for decision theory. Games and Economic Behaviour,
56, 45–60.
Davidson, D., Suppes, P., & Siegel, S. (1957). Decision making: An experimental approach. Stanford, CA: Stanford University Press.
Donkers, B., Melenberg, B., & van Soest, A. (2001). Estimating risk attitudes using lotteries; a large sample approach. Journal of Risk and Uncertainty, 22,
165–195.
Etchart-Vincent, N. (2004). Is probability weighting sensitive to the magnitude of consequences? An experimental investigation on losses. Journal of Risk and
Uncertainty, 28, 217–235.
Fehr-Duda, H., de Gennaro, M., & Schubert, R. (2006). Gender, financial risk, and probability weights. Theory and Decision, 60, 283–313.
Fellner, G., & Maciejovsky, B. (2007). Risk attitude and market behavior: Evidence from experimental asset markets. Journal of Economic Psychology, 28,
338–350.
Fennema, H., & van Assen, M. (1998). Measuring the utility of losses by means of the tradeoff method. Journal of Risk and Uncertainty, 17, 277–295.
Fishburn, P. C., & Kochenberger, G. A. (1979). Two-piece von neumann-morgenstern utility functions. Decision Sciences, 10, 503–518.
Friend, I., & Blume, M. E. (1975). The demand for risky assets. American Economic Review, 65, 900–922.
Gächter, S. E., Johnson, J., & Herrmann, A. (2007). Individual-level loss aversion in riskless and risky choices. CeDEx Discussion Paper 2007-02. University of
Nottingham.
Halevy, Y. (2007). Ellsberg revisited: an experimental study. Econometrica, 75, 503–536.
Harrison, G. W., Johnson, E., McInnes, M. M., & Rutström, E. E. (2005). Risk aversion and incentive effects: Comment. American Economic Review, 95, 897–901.
Harrison, G. W., & Rutström, E. E. (2009). Expected utility theory and prospect theory: One wedding and a decent funeral. Experimental Economics, 12,
133–158.
Hartog, J., Ferrer-i-Carbonell, A., & Jonker, N. (2002). Linking measured risk aversion to individual characteristics. Kyklos, 55, 3–26.
Hershey, J. C., & Schoemaker, P. J. H. (1985). Probability versus certainty equivalence methods in utility measurement: Are they equivalent? Management
Science, 31, 1213–1231.
Hertwig, R., & Ortmann, A. (2001). Experimental practices in economics: A challenge for psychologists? Behavioural and Brain Sciences, 24, 383–403.
Holt, C. A., & Laury, S. K. (2002). Risk aversion and incentive effects. American Economic Review, 92, 1644–1655.
Holt, C. A., & Laury, S. K. (2005). Risk aversion and incentive effects: New data without order effects. American Economic Review, 95, 902–904.
Johnson, E., Hershey, J., Meszaros, J., & Kunreuther, H. (1993). Framing, probability distortions, and insurance decisions. Journal of Risk and Uncertainty, 7,
35–51.
Kachelmeier, S., & Shehata, M. (1992). Examining risk preferences under high monetary incentives: Experimental evidence from the People’s Republic of
China. American Economic Review, 82, 1120–1141.
Kahneman, D. (2003). Maps of bounded rationality: Psychology for behavioral economics. American Economic Review, 93, 1449–1475.
Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47, 263–291.
Kahneman, D., Knetsch, J. L., & Thaler, R. H. (1991). The endowment effect, loss aversion, and status quo bias: Anomalies. Journal of Economic Perspectives, 5,
193–206.
Köbberling, V., & Wakker, P. P. (2005). An index of loss aversion. Journal of Economic Theory, 122, 119–131.
Kühberger, A., Schulte-Mecklenbeck, M., & Perner, J. (2002). Framing decisions: Hypothetical and real. Organizational Behavior and Human Decision Processes,
89, 1162–1176.
Lopes, L. L. (1987). Between hope and fear: The psychology of risk. Advances in Experimental Psychology, 20, 255–295.
Luce, R. D. (2000). Utility of gains and losses: Measurement-theoretical and experimental approaches. London: Lawrence Erlbaum Publishers.
McCord, M. R., & de Neufville, R. (1986). Lottery equivalents: Reduction of the certainty effect problem in utility assessment. Management Science, 32, 56–60.
Mehra, R., & Prescott, E. (1985). The equity premium: A puzzle. Journal of Monetary Economics, 15, 145–162.
Pälsson, A. (1996). Does the degree of relative risk aversion vary with household characteristics? Journal of Economic Psychology, 17, 771–787.
Plott, C. R., & Zeiler, K. (2005). The willingness to pay-willingness to accept gap, the endowment effect, subject misconceptions, and experimental
procedures for eliciting valuations. American Economic Review, 95, 530–545.
Rabin, M. (2000a). Risk aversion and expected-utility theory: A calibration theorem. Econometrica, 68, 1281–1292.
Rabin, M. (2000b). Diminishing marginal utility of wealth cannot explain risk aversion. In D. Kahneman & A. Tversky (Eds.). Choices, values, and frames (pp.
202–208), New York: Cambridge University Press.
Read, D. (2005). Monetary incentives, what are they good for? Journal of Economic Methodology, 12, 265–276.
Rubinstein, A. (2006). Dilemmas of an economic theorist. Econometrica, 74, 865–883.
Savage, L. (1954). The foundations of statistics. New York: Wiley.
Schmidt, U., & Traub, S. (2002). An experimental test of loss aversion. Journal of Risk and Uncertainty, 25, 233–249.
Schunk, D., & Betsch, C. (2006). Explaining heterogeneity in utility functions by individual difference in decision modes. Journal of Economic Psychology, 27,
386–401.
Smith, V., & Walker, J. (1993). Monetary rewards and decision costs in experimental economics. Economic Inquiry, 31, 245–261.
Starmer, C. (2000). Developments in non-expected utility theory: The hunt for a descriptive theory of choice under risk. Journal of Economic Literature, 38,
332–382.
Tversky, A., & Kahneman, D. (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty, 5, 297–323.
Tu, Q. (2005). Empirical analysis of time preferences and risk aversion. CentER PhD thesis 142. Tilburg University
Von Gaudecker, M., van Soest, A., & Wengstrom, E. (2008). Selection and mode effects in risk preference elicitation. IZA Discussion Paper 3321. University of
Bonn.
Wakker, P. P. (2005). Formalizing reference dependence and initial wealth in Rabin’s calibration theorem. Working paper. Erasmus University.
Wakker, P. P. (2008). Explaining the characteristics of the power (CRRA) utility function. Health Economics, 17, 1329–1344.
Wakker, P. P., & Deneffe, D. (1996). Eliciting von Neumann–Morgenstern utilities when probabilities are distorted or unknown. Management Science, 42,
1131–1150.
Wakker, P. P., Köbberling, V., & Schwieren, C. (2007). Prospect-theory’s diminishing sensitivity versus economic’s intrinsic utility of money: How the
introduction of the Euro can be used to disentangle the two empirically. Theory and Decision, 63, 205–231.
Wakker, P. P., & Tversky, A. (1993). An axiomatization of cumulative prospect theory. Journal of Risk and Uncertainty, 7, 147–176.
Weber, E., Shafir, S., & Blais, A. R. (2004). Predicting risk sensitivity in humans and lower animals: Risk as variance or coefficient of variation. Psychological
Review, 111, 430–445.