How Basic Are Behavioral Biases? Evidence from Capuchin Monkey Trading Behavior
Author(s): M. Keith Chen, Venkat Lakshminarayanan, and Laurie R. Santos
Source: Journal of Political Economy, Vol. 114, No. 3 (June 2006), pp. 517-537
Published by: The University of Chicago Press
Stable URL: http://www.jstor.org/stable/10.1086/503550 .
Accessed: 02/07/2015 08:58
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
.
The University of Chicago Press is collaborating with JSTOR to digitize, preserve and extend access to Journal
of Political Economy.
http://www.jstor.org
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
517
[ Journal of Political Economy, 2006, vol. 114, no. 3]
᭧ 2006 by The University of Chicago. All rights reserved. 0022-3808/2006/11403-0004$10.00
How Basic Are Behavioral Biases? Evidence
from Capuchin Monkey Trading Behavior
M. Keith Chen
Yale University and Cowles Foundation
Venkat Lakshminarayanan and Laurie R. Santos
Yale University
Behavioral economics has demonstrated systematic decision-making
biases in both lab and field data. Do these biases extend across contexts,
cultures, or even species? We investigate this question by introducing
fiat currency and trade to a colony of capuchin monkeys and
recovering their preferences over a range of goods and gambles. We
show that capuchins react rationally to both price and wealth shocks
but display several hallmark biases when faced with gambles, including
reference dependence and loss aversion. Given our capuchins’ inexperience
with money and trade, these results suggest that loss aversion
extends beyond humans and may be innate rather than learned.
Nobody ever saw a dog make a fair and deliberate exchange
of one bone for another with another dog. Nobody ever
saw one animal by its gestures and natural cries signify to
We would like to acknowledge financial support from Whitebox Advisors and the Russell
Sage Foundation, without which this research would not have been possible. Alisia Eckert
provided invaluable research assistance throughout this project. Many people gave generously
of their time on this and earlier drafts; we would especially like to thank Nick
Barberis, Dan Benjamin, Judy Chevalier, Steve Levitt, Barry Nalebuff, Emily Oster, Jesse
Shapiro, Peter Schott, and seminar participants at Berkeley, Brown, University of California
at San Diego, Caltech, the University of Chicago, the Stanford Institute for Theoretical
Economics, and Yale for their insightful comments and suggestions. These data and additional
materials are available at http://www.som.yale.edu/Faculty/keith.chen/.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
518 journal of political economy
another, this is mine, that yours; I am willing to give this
for that. (Adam Smith, Wealth of Nations)
I. Introduction
Over the past few decades, behavioral economists have identified that
human decision makers exhibit a number of systematic biases both in
the lab and in the field. Two of these biases, reference dependence and
loss aversion, have received a substantial amount of empirical attention,
from both economics and neighboring disciplines such as psychology
and sociology.1
Evidence that agents treat losses differently than comparable
gains has been found in experimental markets as the endowment
effect2
(Kahneman, Knetsch, and Thaler 1990), in the trades of
individual investors who are reluctant to realize losses (Odean 1998),
and in the behavior of house sellers who are unwilling to sell below
buying price (Genesove and Mayer 2001).
Despite the mounting evidence of the importance of this behavior,
less direct attention has been paid to understanding how basic or widespread
these biases are. Are biases such as loss aversion the result of
social or cultural learning and specific environmental experiences? Or
could they be more universal, perhaps resulting from mechanisms that
arise regardless of context or experience? The root cause of a behavioral
bias may affect how we think about both its potential scope and the
degree to which we believe that market incentives will act to reduce its
effects.
Traditionally, economists have remained agnostic as to the origins of
human preferences and usually assume their stability over both time
and circumstance. For example, Becker (1976, 5) writes that “generally
(among economists) . . . preferences are assumed not to change substantially
over time, nor to be very different between wealthy and poor
persons, or even between persons in different societies and cultures.”3
Indeed, coupled with maximizing behavior and market equilibrium,
Becker asserts that the assumption of stable preferences “forms the heart
of the economic approach.” If much of the fundamental structure of
our preferences were so deeply rooted as to extend to closely related
species, this would bolster the assumption of preference stability.
1
These biases, along with a probability-weighting function, make up “prospect theory,”
first introduced in Kahneman and Tversky (1979). For an excellent summary of the recent
empirical work on prospect theory, see Camerer (2000).
2
The endowment effect is the observation that the minimum amount subjects are willing
to accept to give up a randomly endowed good is often more than twice what they would
have been willing to pay for that good had they not been given it.
3
It must be stressed that Becker was referring not to preferences over market goods,
but to more primitive “underlying objects of choice” such as “health, prestige, sensual
pleasure, benevolence or envy” (1976, 5).
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 519
Indeed, early experimental work found support for the stability of
preferences and the applicability of economic choice theory far from
its usual subjects; several studies have demonstrated that rat and pigeon
behavior seems to obey the laws of demand (Kagel et al. 1975). Unfortunately,
while rats and pigeons are easy subjects to work with, their
limited cognitive abilities make it difficult to investigate more subtle
aspects of economic choice, including many important and systematic
human biases.
In this study we test for both adherence to the law of demand and
the presence of reference-dependent and loss-averse choice in a sophisticated
and closely related primate, the tufted capuchin monkey
(Cebus apella). To do this, we introduce a fiat currency to a colony of
capuchin monkeys, teaching them that small coin-like disks can be
traded with human experimenters for food rewards4
and are fungible
across a variety of possible trades. Using this new ability, we are able to
conduct a number of revealed-preference experiments analogous to
canonical human choice experiments.
Our first set of experiments investigate the purchasing behavior of
capuchins when they are asked to allocate a budget of tokens among a
set of possible foods. In response to both price and wealth shocks,
capuchins adjust their purchasing behavior in ways consistent with the
generalized axiom of revealed preferences (GARP).5
In this way, capuchins’
choice closely mirrors our own and admits the standard tools of
utility analysis and price theory. This closely mirrors the early economic
work on rats and pigeons and provides a context in which to interpret
the capuchins’ latter departure from rational choice.
Our second set of experiments demonstrates that when faced with
decisions involving simple gain-loss frames, capuchins demonstrate both
reference dependence and loss aversion. In these experiments, a capuchin
must choose between trades in which one amount of food is initially
displayed (serving as a frame) but another amount may be delivered.
Specifically, in our main experiments, capuchins express a strong preference
for gambles in which good outcomes are framed as bonuses (the
subject sometimes receives more than was initially displayed) rather than
payoff-identical gambles in which bad outcomes are framed as losses.
4
For similar trading methodologies with capuchins, see Westergaard, Liv, Chavanne,
and Suomi (1998), Liv, Westergaard, and Suomi (1999), Brosnan and de Waal (2003,
2004), and Westergaard et al. (2004); for a very early example of primates’ capacity to
trade, see Wolfe (1936).
5
Early papers by Samuelson (1938), Houthakker (1950), and Afriat (1967) established
the revealed preference approach to evaluating whether any given set of choices is consistent
with rational behavior. Varian (1982) generalized this approach and showed that
GARP is a necessary and sufficient condition for any set of choices to arise from the
maximization of a continuous, concave, weakly monotonic, and locally nonsatiated utility
function.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
520 journal of political economy
Furthermore, capuchins seem to weigh those losses more heavily than
comparable gains, displaying not just reference dependence but loss
aversion. These experiments also allow us to reject most competing
models of naive or unsophisticated choice. In particular, several of our
results require a capuchin to overcome the impulse to try to obtain an
initially larger food reward and instead to trade with an experimenter
who initially displays a smaller food reward. Numerous studies have
shown that this type of inhibition problem is difficult for both monkeys
and even great apes.6
Arguing that loss aversion is not an acquired characteristic of capuchin
preferences is the novelty of the situation. Abstract gambles were first
introduced to these capuchins by our experiments, and subjects encountered
them alone, away from others; subjects had no prior trading
experience that could have led to the development of these biases. As
such, our results suggest that loss-averse behavior is a very general feature
of economic choice and extends to some of our closely related neighbors.
A large body of research in biology suggests that since humans
and capuchins are closely related, any shared cognitive systems are likely
to have a common origin. In light of this, our results may suggest that
loss aversion is an innate and evolutionarily ancient feature of human
preferences, a function of decision-making systems that evolved before
the common ancestors of capuchins and humans diverged.
The remainder of the paper is organized as follows. Section II reviews
the set of papers in behavioral economics that speak to the origin and
scope of behavioral biases. Section III describes capuchins both as a
species and as our test subjects, and lays out some of the trade-offs
inherent in experimenting on nonhuman primates. Section IV provides
details on our laboratory setup and our method for eliciting preferences
from capuchin trading behavior. Section V describes our initial compensated
price shift experiments and presents those results. Section VI
describes our method for inducing gain/loss frames and the setup of
our loss-aversion experiments. Section VII presents the results of these
experiments, which we discuss before presenting conclusions in Section
VIII.
6
Both new-world monkeys and great apes fail to solve reverse contingency tasks (games
in which an experimenter presents a large reward whenever the agent reaches for a smaller
treat and presents a small reward whenever the agent reaches for the larger treat) (Boysen
and Berntson 1995; Kralik, Hauser, and Zimlicki 2002). However, when the game is modified
such that the large reward is presented whenever the agent reaches for a symbol of
the small reward (and the small reward is presented whenever the agent reaches for a
symbol of the large reward), many primates succeed (e.g., Boysen, Mukobi, and Berntson
1999). A contribution of our work to the psychology literature is that we report the
surprising result that capuchins have no trouble solving a reverse contingency task when
treats are obtained by exchanging fiat currency for the rewards rather than simply reaching
for them. That is, token-mediated exchange allows primates to overcome the impulse to
simply reach for the greater reward—just as symbol-mediated choice does.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 521
II. Related Literature
Several recent papers have shed light on the foundations of behavioral
biases. A growing literature in the field of neuroeconomics has attempted
to use imaging technology to map brain activity as subjects
make economic decisions and correlates these measures of brain activity
to subjects’ decisions. For example, McClure et al. (2004) show that the
spatial distribution of brain activity is correlated with decisions involving
intertemporal choice. While this approach is extremely useful in shedding
light on the mechanisms of decision making, the ability of this
approach to address questions of universality and stability is limited by
the scope of activities that can be scanned (subjects must be securely
restrained inside a large magnet) and by the difficulty of translating
neural correlates of behavior into causal statements.
A few theory papers have also tried to understand what types of evolutionary
forces might have selected for organisms that display behavioral
biases. These papers typically model preferences as a means by
which nature incentivizes an organism to maximize its evolutionary fitness;
various constraints on nature’s ability to achieve first-best incentives
give rise to behavioral biases. Most notably, Rayo and Becker (2005)
explore what types of evolutionary pressures would have produced both
past payoff and social reference point effects, and Samuelson and Swinkels
(2006) explore under what conditions choice set effects might be
evolutionarily optimal.
More similar in goals to our approach, Henrich et al. (2001) perform
behavioral experiments in 15 small-scale societies, all of which are relatively
isolated and have had relatively limited market contact. Essentially,
their approach exploits the extreme cultural variation between
these societies and finds large differences in how they play an ultimatum
game. We also hope to shed light on the origins of human economic
behavior and the role of environmental experience, but exploit a very
different source of variation than Henrich and his colleagues. Our experiments
can be seen as exploring which aspects of our behavior are
not confined to the heavily socialized human species, but extend to
primates that lack any previous market experience. Specifically, if loss
aversion emerged in our evolutionary past, we would expect that closely
related species would exhibit analogous behavior and may better understand
the origins of our biases by understanding their expression in
our close evolutionary neighbors.
Economic experiments with children and animals.—Harbaugh, Krause, and
Berry (2001) also conduct experiments with goals similar to our own,
exploiting age instead of cultural or species variation. They conduct
numerous simple budgeting experiments on children between the ages
of 7 and 11 and find that violations of GARP are relatively rare. HarThis
content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
522 journal of political economy
baugh, Krause, and Vesterlund (2001), in contrast, find evidence of the
endowment effect in children as young as 5 and find no evidence that
the effect diminishes with age up through college.7
This suggests that
the endowment effect is not reduced by market exposure, though it
leaves open the possibility that children learn this behavior through
experience they receive before age 5.
While the use of animal subjects is widespread in psychology, their
use as subjects in economics is relatively scarce. A notable exception is
the work of Kagel, Battalio, Green, and their colleagues (Kagel et al.
1975; Battalio, Green, and Kagel 1981; Battalio, Kagel, Rachlin, and
Green 1981; Kagel, Battalio, Rachlin, and Green 1981; Battalio, Kagel,
and MacDonald 1985; Kagel, MacDonald, and Battalio 1990; Kagel, Battalio,
and Green 1995). These researchers systematically explored a variety
of economic decisions (e.g., consumer demand, labor supply, risk
aversion, and intertemporal choice) in two classic exemplars of associative
learning: rats and pigeons. Having been trained that different
levers each delivered a unique reward at an experimentally variable rate,
subjects signaled preferences via their lever choices. Kagel and his colleagues
then employed a simple revealed preference method in which
they examined their subjects’ choices when presented with a “budget”
of limited lever presses.
Most applicable to our work, Kagel et al. (1975) explore how rats and
pigeons respond to a compensated price shift. They find that subjects’
choices during such a shift largely respected GARP; in fact, utility maximization
does a much better job of explaining their data than any other
available choice theory (including the canonical nonhuman psychological
choice model, the matching law).8
In later experiments involving
gambles, Kagel and his colleagues observed that, on balance, rats and
pigeons obeyed expected utility theory but do display some systematic
biases. However, in contrast to results on human (and our capuchin)
subjects, Kagel and his colleagues find that prospect theory does not
explain the deviations from expected utility theory that are present in
rats and pigeons.9
We depart from the important work of Kagel and his coauthors in
7
Closely related to loss aversion, the endowment effect is the observation that consumers
often seem to value goods more after possessing them than they do when they do not
have them. This is often characterized by a set of people randomly endowed with an object
exhibiting a higher willingness to accept (the price for selling the good) than the control
group’s willingness to pay. For a good overview of this bias and its connection to loss
aversion, see Kahneman et al. (1991).
8
For a good summary of the psychological literature on the matching law and its relationship
to more modern theories of choice, see Herrnstein and Prelec (1991).
9
Instead, they find evidence of nonstandard probability weighting that is best represented
by some mix of fanning out and fanning in (see Kagel, MacDonald, and Battalio
1990).
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 523
two key ways. First, since rats and pigeons are very distantly related to
humans, experiments on them are of limited use in answering questions
about high-level human decision making, since most of the relevant
neural architecture presumably emerged after our common evolution.
The contribution of Kagel and his coauthors is more closely akin to that
of Becker (1962), demonstrating the robustness of price theory to large
variation in the sophistication of agents.
Second, since rats and pigeons lack the cognitive sophistication of
humans, researchers working with these species can carry out only relatively
simple choice experiments (i.e., choice between trained levers).
These tasks seem unlikely to lead to the classic biases observed in humans,
such as framing or reference point effects. Capuchin monkeys,
on the other hand, are socially sophisticated organisms whose native
ecology requires successful management of scarce resources and risky
trade-offs. This sophistication and their evolutionary proximity to humans
make capuchins far better-suited subjects with which to study the
mechanisms that enable economic decision making; yet, since our subjects
have all been raised in captivity, we can limit the possibility that
behavior analogous to human behavior developed in response to similar
environments.
III. Subjects: The Tufted Capuchin
The tufted (or brown) capuchin is a new-world monkey native to tropical
climates within South America. A cohabiting capuchin breeding group
is usually characterized by a male-dominance hierarchy. A single alphamale
and several sub-alpha-males and females normally live together,
with the alpha-male holding sexual monopoly over the females within
the group. Capuchins are often referred to as “extractive foragers”; they
prefer easy-to-eat fruit but when pressed are capable of pounding apart
hard nuts, stripping tree bark, raiding beehives, and even killing small
vertebrates. For an excellent survey of the species covering all aspects
of their native ecology, see Fragaszy, Visalberghi, and Fedigan (2004).
As a species, tufted capuchin monkeys (Cebus apella) have been widely
studied in psychology and anthropology. They make excellent subjects
since they are relatively quick and adept problem solvers, skilled tool
users, and a close evolutionary neighbor to humans.10
10
Within the set of primates, though, capuchins are actually very distantly related to
humans. Capuchins diverged from our common ancestral line in what biologists call the
new-world primate radiation, when all the primates who inhabit the new world split off
from the old-world primates, the line humans emerged from. While the exact date of this
split is not known, molecular-clock estimates suggest that capuchins split off as a genus
around 23 million years ago. Estimates of our latest common ancestor date around 40
million years (Schneider et al. 2001).
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
524 journal of political economy
Despite this history, conducting economic experiments with capuchins
carries with it several trade-offs compared to conventional human
subjects. Experiments with human subjects must inevitably assume some
independence between the effect being studied in the laboratory and
such things as subjects’ selection into the subject pool, as well as preand
postexperimental conditions outside the laboratory. With nonhuman
subjects, we can control selection and directly manipulate various
features of their daily environment and social interactions. However,
because of the difficulties involved in housing and maintaining a rewarding
environment for capuchin subjects, it is prohibitively costly to
achieve sample sizes to which economists are accustomed in testing
humans. As such, we have chosen a sample size typical for comparative
cognition studies with primates (e.g., Brosnan and de Waal 2003, 2004).
The experimental subjects.—Our test subjects were all born in captivity
and live in a single social group. Six adult capuchins, three male (AG,
FL, and NN) and three female (MD, HG, and JM) ranging from 7 to
8 years old, participated in this experiment. All were genetically unrelated
with the exception of JM (mother of MD). Individuals were isolated
from the rest of the group during each trial in order to minimize the
effects of social interaction on experimental performance.
All subjects had previously participated in experiments concerning
visual cognition, social cognition, and tool use but had never before
participated in a study involving trade. In all our experiments we used
sweet foods, which are highly valued by capuchins, as food rewards.
Outside our experiments all subjects have ample access to water and
calories (in the form of fruits and monkey chow); what motivates our
capuchins is payment of desirable foods, not hunger per se.
IV. Methods: Setting and Apparatus
In all the following experiments subjects were allowed to trade tokens
with one of two experimenters. Each experiment is composed of several
sessions, each session constitutes 12 trials, and each trial is an opportunity
to trade a token for one of two possible food rewards. Every
capuchin was endowed with a budget of tokens at the beginning of each
session and was allowed to allocate this budget however it saw fit; however,
trades had to be conducted one at a time. Identical inch-wide
aluminum discs were used as tokens in all exchanges.
Trading was conducted in a cubical testing chamber (28 inches wide)
that was adjacent to the main cage and into which subjects entered
voluntarily. Two panels on opposite sides of the chamber allowed participants
to interact with experimenters through rectangular openings,
large enough for the capuchins to reach out of (and experimenters to
reach into) the testing chamber (see fig. 1). In each trial, two potential
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 525
Fig. 1.—A capuchin must decide how to spend a budget of coins. The tray at the front
of the testing chamber holds the monkey’s budget, and each of the two experimenters
displays in one hand a food reward in a small tray; his other hand is empty and outstretched.
The subject enters the testing chamber (frame A), takes a token from the tray (frame B),
places it in the hand of an experimenter (frame C), and receives a food reward from a
tray in his other hand (frame D). The film clip from which these pictures are drawn is
available from Chen on request.
trades were offered on opposite sides of the cube, and the subject made
its choice between these two options by choosing which experimenter
to exchange a token with. All sessions were videotaped; in addition, a
research assistant recorded each actor’s string of choices. The experimental
trading protocol is pictured in figure 1.
General methods.—Before each session of 12 trials, two experimenters
( and , wearing different colors)11
arranged an endowment of tokensE E1 2
on a tray, which was in view but out of reach of the subject. To begin
each session, an experimenter pushed the tray within reach of the subjects
through the front of the testing chamber. Then to begin each trial
the two experimenters simultaneously positioned themselves in front of
opposite side panels (frame A in fig. 1). Each experimenter held a dish
with a food reward (in clear view of the subjects) approximately 6 inches
above the opening closest to the interior of the cage and extended an
11
For expositional simplicity each experimenter is denoted by , where x is the numberEx
of pieces of food the experimenter would initially display before possibly adding or taking
away pieces.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
526 journal of political economy
empty hand into the other panel opening. If the capuchin took a token
from the tray (frame B) and placed it in an experimenter’s hand (frame
C), then the experimenter would lower his food dish and present the
capuchin with the food reward (frame D).
In later conditions the experimenters presented capuchins with risky
choices: before lowering the food dish, the experimenter would sometimes
alter the amount of food, either taking away or adding to the
amount of food in the dish. Between each trial the experimenters
swapped positions (replenishing the food in their dish if necessary) and
resumed their initial stance, with the food reward held several inches
above the opening closest to the main cage and an empty hand extended
into the cube through the opening nearest the tokens. The session
ended after the subject exchanged all 12 of its budgeted tokens for food
rewards. Nonstandard trades (including those in which tokens were
thrown from the enclosure or those in which multiple tokens were
pressed into an experimenter’s hand) were not rewarded, and the subject
was allowed to make that choice again. So as to minimize subject
confusion, each experimenter represented a consistent choice for the
capuchin throughout each experiment. No capuchin was allowed to
participate in more than two experimental sessions on the same day
but could participate every day if it wished to.
Subjects each participated in experiments 1, 2, and 3 in sequence,
moving from one experiment to the next when their choices in the
previous one had stabilized. In each of our experiments this criterion
was set as five consecutive sessions in which a capuchin allocated its
tokens in near-constant proportion.12
We took these final choices to
express each actor’s preferred split between the choices each experiment
affords. In our data analysis we use only the last five sessions for
each actor; each capuchin took between six and 12 sessions to stabilize.
Once a subject was finished with an experiment, we transitioned it to
the next by running several days’ worth of “forced trials.” These trials
were identical to the subject’s next experiment, except that only one
of its future choices (randomly selected each trial) was available at any
given time. In this way the capuchin both became aware that the trading
environment had changed and was “forced” to become equally familiar
with both of its new options.
V. Preliminary Experiments: Capuchins Obey Price Theory
Our preliminary experiments closely mirror those of Kagel and his coauthors
and allow us to directly test that capuchin choice looks broadly
12
This meant that the token allocations moved no more than one out of each session’s
12 trials, for five consecutive sessions.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 527
rational and admits standard price theory. In order to do so, we first
found a set of two goods for the subjects between which they were
roughly indifferent and then elicited their choice over a simple budget
set between these two goods. We then subjected each capuchin to a
compensated price shift and examined how it responded.
A. Methods: Identifying Preferences
Before beginning the pricing experiment, each participant was tested
to identify two food rewards between which the subject was roughly
indifferent. That is, starting with apples as the first good, we looked for
another good such that when allocating a budget of 12 disks, the capuchin
would reliably consume at least some of each good. Each experimenter
was assigned a different good to display and exchange for a
single token. When a subject reliably consumed a positive quantity of
both apples and the other food over at least 10 sessions, it was determined
that the subject was roughly indifferent between the two goods
offered (these foods ended up being either grapes or gelatin cubes).
Until this combination was found, the nonapple experimenter changed
the good he or she offered until this interior budgeting condition was
satisfied.
Baseline and compensated price shift.—Once an appropriate good was
found, the next steps of our price shift experiment were very straightforward.
To establish a baseline measurement, each one of three subjects
was repeatedly asked to allocate a budget of 12 disks between food 1
(apples) and food 2 (either grapes or gelatin). This was done exactly
as described in the general methods above, with each experimenter
trading one token for one piece of the respective food reward.
Each capuchin was run on this baseline condition until its choices
stabilized, that is, until its choices did not change by more than one
token for a span of five sessions. Once an actor had stabilized, its behavior
over the next week was averaged into a representative consumption
bundle (see the solid budget line in fig. 2). Using this bundle, we
assigned a new budget of disks to each actor for use in a compensated
price shift.
In this compensated price shift, the experimenter who trades for
apples changed the amount he was willing to trade for a token. Instead
of trading a token for one piece of apple, the experimenter would now
always display and trade two pieces of apple for each token. This represented
a fall in the price of apples by a half, and in order to compensate
for this, each subject’s budget was reduced. The budget was
reduced from 12 to either nine or 10 tokens, depending on which most
closely shifted back the new budget set such that the bundle the subject
originally consumed was close to still lying on the new budget line (see
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
528 journal of political economy
Fig. 2.—Subjects’ choices over baseline and compensated price shift regimes. Capuchins
satisfy GARP; each shaded point represents the purchasing behavior of a single subject
in either a baseline or compensated price shift condition. We represent their average
choice behavior from the week after they had learned the trading regime and their choices
stabilized.
the dashed budget lines in fig. 2). Each subject’s preferences were again
allowed to stabilize, and then another week’s worth of sessions were
elicited under this new price regime.
B. Results: Preliminary Price Theory Experiments
The results of the preliminary pricing experiment are summarized in
table 1 and figure 2. In figure 2, the solid line represents the initial
budget set each actor was presented in the baseline condition, and the
dashed line represents each actor’s compensated budget after the price
of apples falls in half. In order to satisfy GARP, an actor must consume
(weakly) more apples after the shift than before. All subjects’ choices
are aggregated over at least 10 sessions, and every actor’s choices easily
satisfy GARP at the 1 percent level.
Note though that with only one compensated price shift, several naive
models of choice will satisfy GARP. Indeed as Becker (1962) points out,
many forms of random behavior can satisfy GARP in response to compensated
price changes. In this setting, note that since we have experimenters
switch sides of the testing chamber after each trial (to eliminate
side bias), constant budget rationing could arise from complete inattention.
To account for this we also test whether each subject’s choices
are inelastic enough to have arisen from random behavior, using as our
null that neither price nor wealth shocks affect a capuchin’s choices.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 529
TABLE 1
Price Theory and Compensated Price Shifts
Actor
FL NN AG
A. Baseline Experimenta
Food used for good 1 Apples Apples Apples
Food used for good 2 Gelatin Grapes Gelatin
Amount spent on good 1 47% 42% 51%
Number of trials 132 72 144
B. Compensated Price Shift Experimentb
New budget of disks 10 10 9
Amount spent on good 1 69% 64% 50%
Good 1 consumed before r
after shift 5.6r13.8 5.0r12.8 6.1r9.0
Number of trials 140 100 216
Choices satisfy GARPc
p ! .001 p ! .001 p ! .001
Choices respond to shiftd
p ! .001 p ! .004 p ! .966
Note.—Tests of significance are reported as p-values of a two-sided test.
a
Subjects were given a budget of 12 disks, and both goods have price 1.
b
Good 1’s price falls from 1 to .5, good 2’s price stays 1, and the budget shrinks.
c
Our test of GARP is a two-sided p-test that the number of pieces of apple that the actor consumes weakly increases
after the compensated price shift.
d
This is a two-sided p-test that the fraction of trials the subject chooses apples responds to the price shift; NN and
FL change significantly but AG does not.
Accordingly, in table 1 we also examine the percentage of trials each
capuchin spends trading for apples before and after the price shift and
test whether these percentages differ. Two out of three subjects showed
a significant response to this test at the 1 percent level. Note that our
third subject, AG, does not pass this test, spending 50 percent of its
time (and, consequently, its budget) trading for apples both before and
after the price shift. AG’s behavior suggests that it either is maximizing
Cobb-Douglas preferences (with a price elasticity of minus one) or is
inattentive to prices when choosing which experimenter to trade with.
VI. Main Experiments: Are Capuchins Reference-Dependent?
Once our original subjects completed these initial experiments, an additional
three subjects were recruited for our main set of experiments;
AG was dropped from the study for becoming unresponsive. In this set
of experiments the same budgeting procedure was used to elicit choices,
with each session composed of 12 opportunities to trade a token for
one of two possible food rewards. In contrast to the initial experiment,
though, only apples were used as a food reward, and the experimenters
no longer automatically presented the capuchin with the apples displayed
in their tray when given a token. Now, experimenters sometimes
altered the food in the presentation tray before making that tray availThis
content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
530 journal of political economy
able to the subject. In this way, we were able to independently vary what
the capuchin was initially shown and what the capuchin would receive
in exchange for a token, which sometimes consisted of a gamble.
A. Methods: Experiment 1, Stochastic Dominance
In experiment 1, a capuchin could trade its tokens with one of two
experimenters. Experimenter represented a random payoff of oneE2
or two apple pieces each with equal probability, and experimenter E1
represented a sure payoff of one piece. Experimenters and alsoE E1 2
differed in how many pieces they initially showed the capuchin: dis-E2
played two squares of apple, and displayed only one square of apple.E1
This experiment tests whether capuchins’ choice respects first-order
stochastic dominance, that is, if they prefer gambles that weakly dominate
another option.
Specifically, after being given a token, experimenter always loweredE1
his dish to present the subject with one apple piece—exactly as many
as he had displayed. In contrast, started every trial displaying twoE2
apple pieces in her tray but would deliver both pieces only half the time
she was traded with. The other half of the time would remove oneE2
of her two apple pieces and deliver only the remaining piece to the
subject. A random number generator determined beforehand whether
any given trade would result in a payoff of two or one; when an apple
piece was removed, it was placed into an opaque receptacle underneath
the testing table that was both out of sight and out of reach of the
subject.
B. Methods: Experiment 2, Reference Dependence
In experiment 2, subjects chose between experimenters who both delivered
identical gambles, differing only in whether they added to or
subtracted from their initial displayed offering of one or two apple
pieces. This was designed to test whether capuchins would respond to
a simple framing manipulation, presenting some payoffs as gains and
some as losses, while holding constant the underlying payoffs.
Specifically, and would stand on opposite sides of the testingE E1 2
chamber displaying one and two apple pieces, respectively. Upon being
presented with a token, either would present the subject with theE2
two apple pieces he had displayed or he would visibly remove one piece
and deliver only the remaining apple piece to the subject. When an
apple piece was removed, it was placed into an opaque receptacle underneath
the testing table that was out of sight and out of reach of the
subject.
When the subject traded with , however, she would either presentE1
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 531
the single piece she displayed or add one apple piece and deliver two
pieces. When this bonus piece was added, it was drawn from an identical
receptacle.
Essentially then, both experimenters represented a 50-50 lottery of
one or two apple pieces. They differed only in whether they initially
had displayed one or two apples, framing for the marginal apple piece
as either a gain or a loss. A random number generator determined
beforehand whether any given trade would result in a payoff of two or
one; we call these the bonus versus penalties conditions.
C. Methods: Experiment 3, Loss Aversion
In experiment 3, subjects chose between experimenters who both delivered
a payoff of one apple piece, differing only in whether they initially
displayed one or two pieces. This experiment was designed to test
for the presence of reference effects in riskless situations and, when
combined with experiment 2, allows us to measure loss aversion.
Specifically, and would stand on opposite sides of the testingE E1 2
chamber, displaying one and two apple pieces, respectively. Upon being
presented a token, always removed one apple piece and deliveredE2
the remaining piece to the subject. The removed square was placed into
the opaque receptacle underneath the testing table. In contrast, if the
subject traded with , she always presented the single square sheE1
displayed.
Essentially then, trading with either or delivered identical pay-E E1 2
offs. However, on all trades gave exactly the quantity of apple heE1
displayed, whereas displayed a quantity of apple that was always re-E2
duced from two to one before it was made available to the subject.
VII. Results: Main Experiments
The results of all our experiments are reported in tables 2, 3, and 4,
broken down by subject. We will first discuss what can be learned from
our results without imposing any significant parametric assumptions; in
the next section we fit a simplified version of the standard prospecttheoretic
utility function to our subjects that allows more precise analysis.
The results of experiment 1 are summarized in table 2, which shows
how the subjects behaved over five sessions (60 trials), after an initial
set of sessions in which their choices stabilized as they learned about
the experimental choices. The capuchins express a clear preference (87
percent of trades) for , the experimenter who displays two apple piecesE2
and delivers either one or two pieces with equal probability. This is of
course not surprising; the second option stochastically dominates the
first and gives on average a half piece more of apple.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
532 journal of political economy
TABLE 2
Experiment 1: Gambles and Stochastic Dominance
Subject Name
FL HG JM MD NN
Trials chosen*E1 10% 18% 12% 22% 5%
Sessions till stable 7 6 7 6 11
Note.—Experimenter shows one and gives one; shows two and then gives one or two with probability one-half.E E1 2
* All subjects’ choices are different from 50 percent at the 1 percent level in a two-sided p-test.
TABLE 3
Experiment 2: Reference Dependence in Gambles
Subject Name
FL HG JM MD NN
Trials chosen*E1 68% 70% 70% 70% 78%
Sessions till stable 11 9 9 9 13
Note.—Experimenter shows one and shows two. Then both give one or two with probability one-half.E E1 2
* All subjects’ choices are different from 50 percent at the 1 percent level in a two-sided p-test.
Given this result, though, the results of experiment 2 are quite surprising.
All subjects left experiment 1 conditioned to favor , the ex-E2
perimenter who displays two pieces of food. Despite this, in experiment
2 the capuchins quickly reverse this preference and trade much more
with the experimenter who displays only one piece of food. Table 3
summarizes these results. Contrary to both their conditioning and the
intuition that naive subjects would favor greater initial displays of food
(experimenter ), capuchins express a preference for , the experi-E E2 1
menter who frames the gamble as a 50 percent chance of a bonus rather
that a 50 percent chance of a loss. Pooled, subjects traded with inE1
71 percent of trials in their last five sessions (again, measured after each
subject’s choices stabilized). For all five subjects this change was significantly
different not just from experiment 1 but from random (50-50)
behavior.
Note that any theory of choice that does not take into account reference
dependence fails to predict this pattern of behavior. Indeed,
since our experimenters switch sides of the testing chamber between
each trial, in order to express a preference between and , a cap-E E1 2
uchin has to actively follow its preferred experimenter from side to side,
expending both time and attention.
Experiment 3 shows that this effect is not confined to risky choices
and, when combined with experiment 2, suggests that capuchins are
not just reference-dependent but loss-averse. The results are summarized
in table 4. Subjects strongly preferred experimenter over ex-E1
perimenter (who initially displayed one and two pieces of apple,E2
respectively), despite the fact that both always provided the same, sure
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 533
TABLE 4
Experiment 3: Riskless Reference Dependence
Subject Name
FL HG JM MD NN
Trials chosen*E1 73% 75% 80% 82% 87%
Sessions till stable 9 10 10 8 10
Trials greater than in experiment
2 p ! .27 p ! .27 p ! .10 p ! .07 p ! .11
Pooled, trials greater than
in experiment 2 p ! .023 p ! .023 p ! .023 p ! .023 p ! .023
Note.—Experimenter shows one and shows two. Then both give one.E E1 2
* All subjects’ choices are different from 50 percent at the 1 percent level in a two-sided p-test.
TABLE 5
Expected Gains, Losses, and Values for Each Experimental Choice
Experiment 1 Experiment 2 Experiment 3
E1 E2 E1 E2 E1 E2
Gamble offered (1, 1, 1) (2, 1, 2) (1, 1, 2) (2, 1, 2) (1, 1, 1) (2, 1, 1)
Gains 0 0 .5 0 0 0
Losses 0 .5 0 .5 0 1
Expected value 1 1.5 1.5 1.5 1 1
Trials chosen 13% 87% 71% 29% 79% 21%
Note.—The table is constructed pooling all subjects’ last five sessions after choices stabilize (60 trials).
payoff of one apple piece. For all subjects this preference (percentage
of trials trading with ) was stronger than in experiment 2, suggestingE1
that this was not due to conditioning from the previous experiment.
Since the only difference between the two experimenters was that E2
showed more than he eventually gave, these results suggest that our
capuchins are reference-dependent even in riskless choice settings.
To investigate whether capuchins display loss aversion, we now compare
their behavior across experiments. Table 5 summarizes the aggregate
behavior of our subjects in all three experiments and summarizes
what each available gamble represented with respect to expected gains,
losses, and food rewards. Looking at experiments 2 and 3, note that in
experiment 2, subjects chose between , who gave a half chance of aE2
loss, and , who gave a half chance of a gain, both of these forces pushingE1
the subject to choose . In experiment 3, however, always gave exactlyE E1 1
what he showed whereas delivered a sure loss. An interpretation ofE2
the fact that subjects show a stronger preference in experiment 3 than
in 2 (79 percent to 71 percent) is that a sure loss has a stronger effect
than the combined effects of a half loss and a half gain, or that losses
affect a subject’s choices more than gains. In other words,
1 1
FlossF 1 FlossF ϩ FgainF ⇔ FlossF 1 FgainF.2 2
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
534 journal of political economy
Thus the pooled p-test of experiment 3 being stronger than in experiment
2 ( in a two-sided p-test) can be taken to confirm thep ! .023
presence of loss aversion in capuchin choice.
A similar intuition suggests that expected value calculations weigh
more heavily than framing effects on capuchin choice. Comparing experiments
1 and 2, note that in both choices, a half chance of a loss
pushes the subject to prefer to . However, these experiments differE E2 1
in what force would attract a subject to : in experiment 2, it is theE1
half chance of a possible gain, whereas in experiment 1, it is the expectation
of an extra half piece of apple. That the capuchins strongly
prefer in experiment 1 and in experiment 2 suggests that increasesE E2 1
in expected rewards greatly outweigh the effects of potential losses.
VIII. Discussion and Conclusion
Taken together our results suggest that capuchins’ choice both is very
sensitive to changes in prices, budgets, and expected payoffs and, to a
lesser degree, displays both reference dependence and loss aversion.
That expected rewards carry a much larger effect than gain/loss frames
in our experiments is perhaps not surprising; our analysis examines only
the long-run behavior of our subjects after facing the same choice many
times, and our reference point treatment is a relatively mild framing
intervention.
It is tempting to ask whether capuchins’ loss aversion resembles human
aversion in magnitude; for example, in much of the work on human
loss aversion, the ratio of the coefficients on losses and gains is remarkably
stable and is commonly used as a measure of loss aversion,
representing how “kinked” the utility function is at the reference point.
Tversky and Kahneman (1991) summarize a large body of survey evidence
on minimally acceptable gambles and find a ratio of roughly
2.5 : 1. Average ratios of willingness to pay to willingness to accept found
in most endowment effect experiments (see, e.g., Kahneman et al. 1990)
yield a ratio of around 2.7 : 1. Benartzi and Thaler (1995) calibrate a
ratio of approximately 2.3 from the aggregate risk preferences of stock
investors. Although the amount of price variation we achieve with only
three experiments is minimal, imposing additional structure on capuchin
preferences allows the estimation of a comparable but extremely
rough measure of losses to gains for our capuchins.13
We have argued that finding behavioral biases in capuchin choice
13
In a preliminary analysis, we fit a simple linear utility specification to our capuchins’
behavior. While parametric results are always to be taken with caution, the relative strength
of losses to gains in capuchin decisions (the coefficient of loss aversion) seems similar to
human estimates. For details, see the working version of this paper on Chen’s Web site
(http://www.som.yale.edu/Faculty/keith.chen/).
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 535
can suggest an early-evolutionary origin for these biases in humans. This
relies heavily on the fact that on questions of origin, it is widely accepted
in both cognitive science and evolutionary psychology that a mechanism
is most likely evolutionarily ancient if it explains analogous behavior in
both humans and primates. That is, since primates and humans are
closely related, it is unlikely that a common trait evolved in parallel
between our two species, and much more likely that common traits
evolved once during our common evolutionary heritage. While our results
are by no means definitive proof that loss aversion is innate in
humans, to the degree that they make us more likely to believe that
some amount of this behavior has a biological component, they may
have implications for how we treat loss-averse tendencies in human
behavior.
For example, if these biases are innate, we may be more inclined to
believe that they will persist in both common and novel settings, will
be stable across time and cultures, and may endure even in the face of
large individual costs, ample feedback, or repeated market disciplining.
This would greatly constrain both the potential for successful policy
intervention and the types of remedies available. In contrast, while a
learned, noninnate heuristic may arise in many (if not all) cultures, we
may not expect it to persist in settings in which it was highly suboptimal
or in which market forces strongly discipline behavior. This would limit
the potential scope and scale for welfare losses and may suggest that
policy interventions that increase feedback or learning may eliminate
what losses do exist.
Short of this attributional leap, however, that loss-averse behavior is
not confined to humans can be seen as adding scope to the growing
evidence of loss-averse behavior in many aspects of human economic
behavior. Understanding how broadly these biases manifest themselves
may influence how we should incorporate them into an adequate model
of individual decision making. It is hoped that our paper also suggests
the utility of methodological exchanges between economics and the
closely related behavioral sciences. Bringing the analytic framework of
revealed preference to bear on questions at the intersection of economics,
biology, and psychology carries the possibility of insights useful
to each.
References
Afriat, Sydney N. 1967. “The Construction of Utility Functions from Expenditure
Data.” Internat. Econ. Rev. 8 (February): 67–77.
Battalio, Raymond C., Leonard Green, and John H. Kagel. 1981. “Income-Leisure
Tradeoffs of Animal Workers.” A.E.R. 71 (September): 621–32.
Battalio, Raymond C., John H. Kagel, and Don N. MacDonald. 1985. “Animals’
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
536 journal of political economy
Choices over Uncertain Outcomes: Some Initial Experimental Results.” A.E.R.
75 (September): 597–613.
Battalio, Raymond C., John H. Kagel, Howard Rachlin, and Leonard Green.
1981. “Commodity-Choice Behavior with Pigeons as Subjects.” J.P.E. 89 (February):
67–91.
Becker, Gary S. 1962. “Irrational Behavior and Economic Theory.” J.P.E. 70 (February):
1–13.
———. 1976. The Economic Approach to Human Behavior. Chicago: Univ. Chicago
Press.
Benartzi, Shlomo, and Richard H. Thaler. 1995. “Myopic Loss Aversion and the
Equity Premium Puzzle.” Q.J.E. 110 (February): 73–92.
Boysen, Sarah T., and Gary G. Berntson. 1995. “Responses to Quantity: Perceptual
versus Cognitive Mechanisms in Chimpanzees (Pan troglodytes).” J. Experimental
Psychology: Animal Behavior Processes 21 (January): 82–86.
Boysen, Sarah T., Kimberly L. Mukobi, and Gary G. Berntson. 1999. “Overcoming
Response Bias Using Symbolic Representations of Number by Chimpanzees
(Pan troglodytes).” Animal Learning and Behavior 27 (May): 229–35.
Brosnan, Sarah F., and Frans B. M. de Waal. 2003. “Monkeys Reject Unequal
Pay.” Nature 425 (September 18): 297–99.
———. 2004. “Socially Learned Preferences for Differentially Rewarded Tokens
in the Brown Capuchin Monkey (Cebus apella).” J. Comparative Psychology 118
(June): 133–39.
Camerer, Colin F. 2000. “Prospect Theory in the Wild: Evidence from the Field.”
In Choices, Values, and Frames, edited by Daniel Kahneman and Amos Tversky.
Cambridge: Cambridge Univ. Press.
Fragaszy, Dorothy M., Elisabetta Visalberghi, and Linda Marie Fedigan. 2004.
The Complete Capuchin: The Biology of the Genus Cebus. Cambridge: Cambridge
Univ. Press.
Genesove, David, and Christopher Mayer. 2001. “Loss Aversion and Seller Behavior:
Evidence from the Housing Market.” Q.J.E. 116 (November): 1233–
60.
Harbaugh, William T., Kate Krause, and Timothy R. Berry. 2001. “GARP for Kids:
On the Development of Rational Choice Behavior.” A.E.R. 91 (December):
1539–45.
Harbaugh, William T., Kate Krause, and Lise Vesterlund. 2001. “Are Adults Better
Behaved than Children? Age, Experience, and the Endowment Effect.” Econ.
Letters 70 (February): 175–81.
Henrich, Joseph, et al. 2001. “In Search of Homo Economicus: Behavioral Experiments
in 15 Small-Scale Societies.” A.E.R. Papers and Proc. 91 (May): 73–
78.
Herrnstein, Richard J., and Drazen Prelec. 1991. “Melioration: A Theory of
Distributed Choice.” J. Econ. Perspectives 5 (Summer): 137–56.
Houthakker, Hendrik S. 1950. “Revealed Preference and the Utility Function.”
Economica 17 (May): 159–74.
Kagel, John H., et al. 1975. “Experimental Studies of Consumer Demand Behavior
Using Laboratory Animals.” Econ. Inquiry 13 (March): 22–38.
Kagel, John H., Raymond C. Battalio, and Leonard Green. 1995. Economic Choice
Theory: An Experimental Analysis of Animal Behavior. Cambridge: Cambridge
Univ. Press.
Kagel, John H., Raymond C. Battalio, Howard Rachlin, and Leonard Green.
1981. “Demand Curves for Animal Consumers.” Q.J.E. 96 (February): 1–16.
Kagel, John H., Don N. MacDonald, and Raymond C. Battalio. 1990. “Tests of
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions
behavioral biases 537
‘Fanning Out’ of Indifference Curves: Results from Animal and Human Experiments.”
A.E.R. 80 (September): 912–21.
Kahneman, Daniel, Jack L. Knetsch, and Richard H. Thaler. 1990. “Experimental
Tests of the Endowment Effect and the Coase Theorem.” J.P.E. 98 (December):
1325–48.
———. 1991. “The Endowment Effect, Loss Aversion, and Status Quo Bias:
Anomalies.” J. Econ. Perspectives 5 (Winter): 193–206.
Kahneman, Daniel, and Amos Tversky. 1979. “Prospect Theory: An Analysis of
Decision under Risk.” Econometrica 47 (March): 263–91.
Kralik, Jerald D., Marc D. Hauser, and Rasa Zimlicki. 2002. “The Relationship
between Problem Solving and Inhibitory Control: Cotton-Top Tamarin (Saguinus
oedipus) Performance on a Reversed Contingency Task.” J. Comparative
Psychology 116 (March): 39–50.
Liv, Chanya, Gregory C. Westergaard, and Stephen J. Suomi. 1999. “Exchange
and Value in Cebus apella.” American J. Primatology 49 (1): 74–75.
McClure, Samuel M., David I. Laibson, George Loewenstein, and Jonathan D.
Cohen. 2004. “Separate Neural Systems Value Immediate and Delayed Monetary
Rewards.” Science 306 (October 15): 503–7.
Odean, Terrance. 1998. “Are Investors Reluctant to Realize Their Losses?” J.
Finance 53 (October): 1775–98.
Rayo, Luis, and Gary S. Becker. 2005. “Evolutionary Efficiency and Happiness.”
Working paper, Univ. Chicago.
Samuelson, Larry, and Jeroen M. Swinkels. 2006. “Information, Evolution and
Utility.” Theoretical Econ. 1 (March): 119–42.
Samuelson, Paul A. 1938. “A Note on the Pure Theory of Consumer’s Behavior.”
Economica 5 (February): 61–71.
Schneider, Hora´cio, et al. 2001. “Can Molecular Data Place Each Neotropical
Monkey in Its Own Branch?” Chromosoma 109 (February): 515–23.
Tversky, Amos, and Daniel Kahneman. 1991. “Loss Aversion in Riskless Choice:
A Reference-Dependent Model.” Q.J.E. 106 (November): 1039–61.
Varian, Hal R. 1982. “The Nonparametric Approach to Demand Analysis.” Econometrica
50 (July): 945–73.
Westergaard, Gregory C., Chanya Liv, Tara J. Chavanne, and Stephen J. Suomi.
1998. “Token Mediated Tool-Use by a Tufted Capuchin Monkey (Cebus apella).”
Animal Cognition 1 (2): 101–6.
Westergaard, Gregory C., Chanya Liv, Andrea Rocca, Allison Cleveland, and
Stephen J. Suomi. 2004. “Tufted Capuchins (Cebus apella) Attribute Value to
Foods and Tools during Voluntary Exchanges with Humans.” Animal Cognition
7 (1): 19–24.
Wolfe, John Bascom. 1936. Effectiveness of Token-Rewards for Chimpanzees. Baltimore:
Johns Hopkins Press.
This content downloaded from 147.251.189.14 on Thu, 2 Jul 2015 08:58:02 AM
All use subject to JSTOR Terms and Conditions