To be published in Behavioral
and Brain Sciences (in press)
© Cambridge University Press
2005
Below is the copyedited
final draft of a BBS target article that has been accepted for publication.
This updated preprint has been prepared for formally invited commentators.
Please DO NOT write a commentary unless you have been formally invited.
“Economic man” in cross-cultural perspective:
Behavioral experiments in 15 small-scale societies
Joseph Henrich
Department of Anthropology,
Robert Boyd
Department of Anthropology,
rboyd@anthro.ucla.edu
Samuel Bowles
Santa Fe Institute,
Colin Camerer
Div HSS 228-77, California Institute of Technology,
Ernst Fehr
Herbert Gintis
Santa Fe Institute,
Richard McElreath
Department of Anthropology,
Michael Alvard
Department of Anthropology,
Alvard@tamu.edu
Abigail Barr
Centre for the Study of African Economies,
abigail.barr@economics.ox.ac.uk
Jean Ensminger
Div HSS 228-77, Caltech,
Natalie Smith Henrich
ORC Macro.
Kim Hill
Department of Anthropology,
Francisco Gil-White
Department of Psychology,
fjgil@psych.upenn.edu
Michael Gurven
Department of Anthropology, University of
California-Santa Barbara,
gurven@anth.ucsb.edu
Frank W. Marlowe
Department of Anthropology,
John Q. Patton
Department of Anthropology,
pattonj@mail.wsu.edu
David Tracer
Department of Anthropology & Health &
Behavioral Sciences,
Abstract: Researchers from across the social sciences have
found consistent deviations from the predictions of the canonical model of
self-interest in hundreds of experiments from around the world. That research,
however, cannot determine whether the uniformity results from universal
patterns of human behavior or from the limited cultural variation available
among the university students used in virtually all prior experimental work. To
address this, we undertook a cross-cultural study of behavior in Ultimatum,
Public Goods, and Dictator Games in a range of small-scale societies exhibiting
a wide variety of economic and cultural conditions. We found, first, that the
canonical model – based on pure self-interest – fails in all of the societies
studied. Second, our data reveal substantially more behavioral variability
across social groups than has been found in previous research. Third,
group-level differences in economic organization and the structure of social
interactions explain a substantial portion of the behavioral variation across
societies: the higher the degree of market integration and the higher the
payoffs to cooperation in everyday life, the greater the level of prosociality
expressed in experimental games. Fourth, the available individual-level economic
and demographic variables do not consistently explain game behavior, either
within or across groups. Fifth, in many cases experimental play appears to
reflect the common interactional patterns of everyday life.
Keywords: self-interest; altruism; cooperation; Ultimatum Game;
Public Goods Game; cross-cultural research; experimental economics; game theory
1. Introduction
Since “Selfishness examined…” (Caporael
et al. 1989) appeared in these pages more than 15 years ago, many subsequent experiments
have strongly confirmed the doubts expressed by Caporael and her collaborators
concerning the adequacy of self-interest as a behavioral foundation for the
social sciences. Experimental economists and others have uncovered large,
consistent deviations from the textbook predictions of Homo economicus
(Camerer 2003; Fehr et al. 2002; Hoffman et al. 1998; Roth 1995). Hundreds of
experiments in dozens of countries, using a variety of game structures and
experimental protocols, have suggested that in addition to their own material
payoffs, students care about fairness and reciprocity and will sacrifice their
own gains to change the distribution of material outcomes among others,
sometimes rewarding those who act prosocially and punishing those who do not. Initial
skepticism about such experimental evidence has waned as subsequent studies
involving high stakes and ample opportunity for learning has repeatedly failed
to modify these fundamental conclusions.
This multitude of diverse experiments creates a powerful
empirical challenge to what we call the selfishness axiom – the
assumption that individuals seek to maximize their own material gains in these
interactions and expect others to do the same.1 However, key
questions remain unanswered. Do such consistent violations of the canonical
model provide evidence of universal patterns that characterize our species? Or
do individuals’ economic and social environments shape their behavior,
motivations, and preferences? If the latter, are there boundaries on the malleability
of human nature, and which economic and social conditions are most involved?
Are there cultures that approximate the canonical account of purely
self-regarding behavior? Are inclinations toward fairness (equity) and “tastes”
for punishing unfairness better explained statistically by individuals’
attributes such as their sex, age, education, and relative wealth, or by the
attributes of the individuals’ group?
Previous research could not answer such questions because
virtually all subjects have been university students. Although there are modest
differences among student populations throughout the world (Roth et al. 1991),
these differences in subjects and settings are small compared to the full range
of human social and cultural environments. To broaden this inquiry, we
undertook a large cross-cultural study using Ultimatum, Public Goods, and
Dictator Games. Twelve experienced field researchers, working in 12 countries
on four continents and New Guinea, recruited subjects from 15 small-scale
societies exhibiting a wide variety of economic and social conditions. Our
sample of societies consisted of three groups of foragers, six groups of slash-and-burn
horticulturalists, four groups of nomadic herders, and two groups of
small-scale agriculturalists.
Our overall results can be summarized in five major points:
first, there is no society in which experimental behavior is fully consistent
with the selfishness axiom; second, there is much more variation between groups
than previously observed, although the range and patterns in the behavior
indicate that there are certain constraints on the plasticity of human
sociality; third, differences between societies in market integration and the
local importance of cooperation explain a substantial portion of the behavioral
variation between groups; fourth, individual-level economic and
demographic variables do not consistently explain behavior within or across
groups; and fifth, experimental play often reflects patterns of interaction
found in everyday life. In this target article, we describe the experimental
methods used and give a comparative overview of the societies studied. We then
present and interpret our findings. More extensive details about each society, our
results, and our methods can be found in Henrich et al. (2004).
2. Experimental games and behavior in
student populations
The three experiments we deployed, the Ultimatum Game (UG),
Dictator Game (DG), and Public Goods Game (PGG), have been extensively studied among
students in complex market societies. In this section, we lay out the basic
games and briefly summarize the typical findings in such student populations.
For extensive reviews see Kagel and Roth (1997) and Camerer (2003).
2.1.
The Ultimatum and Dictator Games
The UG is a simple bargaining game in which subjects are
paired. The first player, often called the “proposer,” is provisionally
allotted a divisible “pie” (usually money). The proposer then offers a portion
of the total pie to a second person, called the “responder.” The responder, knowing
both the offer and the amount of the pie, can either accept or reject the
proposer’s offer. If the responder accepts, he receives the offer and the
proposer keeps the remainder (the pie minus the offer). If the responder
rejects the offer, then neither receives anything. In either case, the game
ends; the two subjects receive their winnings and depart. Players are typically
paid in cash and are anonymous to other players but not to the experimenters
(although experimentalists have manipulated these variables). In all of the
experiments we conducted, players were anonymous to each other and the games
used substantial sums of money (in the appropriate currency). For this game,
the canonical model (all participants maximize their income and this is known
by all of them) predicts that responders, faced with a choice between zero and
a positive payoff, should accept any positive offer. Knowing this, proposers
should offer the smallest nonzero amount possible. In every experiment yet
conducted the vast majority of participants violated this prediction.
The DG is the same as the UG, except that responders are not
given an opportunity to reject – they simply get whatever the proposer
dictates.
In student populations, modal offers in the UG are almost
always 50%, and mean offers are between 40 and 45%. Responders reject offers of
20% about half the time, and rejection is associated with emotional activation
in the insula cortex (Sanfey et al. 2003). In the DG, modal offers are
typically 0% and means usually fall in the 20 to 30% range, although DG results
are more variable than in the UG.
2.2. The Public Goods Game
The PGG shows how people behave when individual and
group-interests conflict. We used two variants: the voluntary contributions
(VC) and the common-pool resources (CPR) formats. In the VC version, players
receive some initial monetary endowment, and then have the simultaneous
opportunity to anonymously contribute any portion of their endowment (from zero
to the full endowment) to the group fund. Whatever money is in the group fund
after players have contributed is augmented by 50% (or sometimes doubled), and
then distributed equally among all players regardless of their contribution.
The payoff structure of the CPR format is identical, except that instead of
receiving an endowment, players can make limited withdrawals from the group
fund. Whatever remains in the fund (the common pool) after everyone has
withdrawn is increased by 50%, or doubled, and distributed equally among all
group members. The game is not repeated. Selfish subjects may calculate that,
independent of the actions taken by the other players, contributing as little as
possible (in the VC version) or withdrawing as much as possible (in the CPR
version) maximizes their monetary payoffs: Free riding is thus the dominant
strategy for selfish subjects.
Students in one-shot Public Goods Games contribute a mean
amount between 40% and 60%, although there is a wide variance, with most
contributing either everything or nothing (Henrich & Smith 2004; Ledyard
1995; Sally 1995). Although this is fairly robust, participants are sensitive
to the costs of cooperation and repeated play. Raising the augmentation
percentage of the common pool produces an increase in contributions (Andreoni
& Miller 2002). When the PGG is played
repeatedly with the same partners, the level of contribution declines toward
zero, culminating in most subjects refusing to contribute to the common pool
(Andreoni 1988; Fehr & Gächter 2000; 2002).
The two major concerns with interpreting experimental data –
stake size and familiarity with the experimental context – have now largely
been put to rest. Some have argued that as the stakes increase, the costs of
being nonselfish also increase, so selfish behavior should increase. Were this
true, it would show that in behaving unselfishly, people respond to costs and
benefits (as they do in many games; cf., e.g., Ledyard 1995; Andreoni &
Miller 2002). But evidence of responding
to the cost of being nonselfish does not suggest that unselfish behavior is
unimportant or extinguished at high stakes. Indeed, in the UG, raising the
stakes to quite high levels (e.g., three months’ income) does not substantially
alter the basic results (Camerer & Hogarth 1999; Cameron 1999; Hoffman et
al. 1996; List & Cherry 2000; Slonim & Roth 1998). In fact, at high
stakes, proposers tend to offer a little more, and responders remain willing to
reject offers that represent small fractions of the pie (e.g., 20%) even when
the pie is large (e.g., $400 in the United States; see List & Cherry 2000).
Similarly, the results do not appear to be due to a lack of familiarity with
the experimental context. Subjects often do not change their behavior in any
systematic way when they participate in several replications of the identical
experiment (Fehr & Gächter 2002; Knez & Camerer 1995; List & Cherry
2000).
Several researchers have tested the effects of demographic
variables on behavior in experimental games (Camerer 2003). The general result
is that demographic effects are nonexistent, or are inconsistent, weak or both.
In the UG, female students reject somewhat less often, but no differences
emerge for offers. In the DG, no gender differences have been found. Similarly,
the age of adult subjects was not an important predictor in any of our games,
or among the handful of results from nonstudent populations in the United
States (Carpenter et al. 2005; Henrich & Henrich in press, chap 7). Thus,
our cross-cultural results are consistent with existing findings on demographic
variables. However, there is intriguing evidence that younger children behave more
selfishly, but gradually behave more fair-mindedly as they grow older, up to
age 22 or so (Harbaugh & Krause 2000; Harbaugh et al. 2002; Murnighan &
Saxon 1998). An important exception is that about one-third of autistic
children and adults offer nothing in the UG (Hill & Sally, 2004);
presumably their inability to imagine the reactions of responders leads them to
behave, ironically, in accordance with the canonical model.
Behavioral economists have been remarkably successful in
explaining the experimental behavior of students by adding social preferences
(especially those related to equity, reciprocity, and fairness) to game
theoretical models (Camerer 2003; Fehr & Schmidt 1999). Our endeavor aims
at the foundation of these proximate models by exploring the nature of
nonselfish preferences.
3. The
cross-cultural behavioral experiments project
Early cross-cultural economic experiments (Cameron 1999;
Roth et al. 1991) showed little variation among university students. However,
in 1996 a surprising finding broke the consensus: the Machiguenga,
slash-and-burn horticulturalists living in the southeastern Peruvian Amazon,
behaved much less prosocially than student populations around the world
(Henrich 2000). The UG “Machiguenga outlier” sparked curiosity among a group of
behavioral scientists: Was this simply an odd result, perhaps due to the
unusual circumstances of the experiment, or had Henrich tapped real behavioral
differences, perhaps reflecting the distinct economic circumstances or cultural
environment of this Amazonian society? In November 1997, the MacArthur
Foundation Research Network on the Nature and Origin of Preferences brought 12
experienced field workers and several behavioral economists together in a
three-day workshop at UCLA. During this meeting we redesigned the experiments –
typically conducted in computer labs at universities – for field implementation
in remote areas among nonliterate subjects. Two years later, when all of our
team had returned from the field, we reconvened to present, compare, and
discuss our findings. Here we summarize the findings to this point (a second
phase is currently underway).
3.1.
The experiments
Overall, we performed 15 Ultimatum, 6 Public Goods, and 3 Dictator
Games, as well as 2 control experiments in the United States at UCLA and at the
University of Michigan. All of our games were played anonymously, in one-shot
interactions, and for substantial real stakes (the local equivalent of one or
more days’ wages). Because the UG was administered everywhere (n = 564
pairs), we concentrate on these findings and their implications, and make only
some references to our other games (see Henrich et al. 2004).
3.2.
Ethnographic description
Figure 1 shows the location of each field site, and Table 1
provides some comparative information about the societies discussed here. In
selecting these, we included societies both sufficiently similar to the
Machiguenga to offer the possibility of replicating the original Machiguenga
results, and sufficiently different from one another to provide enough social,
cultural, and economic diversity to allow an exploration of the extent to which
behaviors covary with local differences in the structures of social
interaction, forms of livelihood, and other aspects of daily life.
Figure 1. Locations of the 15 small-scale societies.
Table 1. (Please click link below for Table 1)
http://www.bbsonline.org/Preprints/Henrich/Tables/Table01.pdf
In Table 1, the Economic Base column provides a general
classification of the production system of each society. Horticulturalists rely primarily on
slash-and-burn agriculture, which involves clearing, burning, and planting
gardens every couple of years. All the horticulturalist societies included here
also rely on some combination of hunting, fishing, and gathering. We have
classified the Aché’s economic base as horticulture/foraging because
they were full-time foragers until about three decades ago, and still
periodically go on multiweek foraging treks, but have spent much of the last
few decades as manioc-based horticulturalists. The Au and Gnau of Papua New
Guinea are classified as foraging/horticulture because, despite planting
small swidden gardens, they rely heavily on harvesting wild sago palms for
calories and hunting game for protein. Unlike foragers and horticulturalists, pastoralists
rely primarily on herding. Agro-pastoralists rely on both small-scale
sedentary agriculture and herding. We labeled the Orma, Mongols, and Kazakhs as
pastoralists because many people in these societies rely completely on herding,
although some members of all three groups do some agriculture. The Sangu are
labeled agro-pastoralists because many people in this society rely
heavily on growing corn, but others rely entirely on animal husbandry
(consequently, we sometimes separate Sangu herders and Sangu farmers).
The Residence column in Table 1 classifies societies
according to the nature and frequency of their movement. Nomadic groups
move frequently, spending as few as a couple of days or as long as a few months
in a single location. Seminomadic groups move less frequently, often
staying in the same location for a few years. Horticultural groups are often
seminomadic, moving along after a couple of years in search of more abundant
game, fish, wild foods, and fertile soils. Transhumant herders move
livestock between two or more locales in a fixed pattern, often following the
good pasture or responding to seasonal rainfall patterns. Bilocal
indicates that families maintain two residences and spend part of the year at
each residence. The Machiguenga, for example, spend the dry season living in
villages along major rivers but pass the wet season in their garden houses,
which may be located three or more hours from the village. The bilocal/seminomadic
classification given to the Machiguenga indicates that traditionally they were seminomadic
but have more recently adopted a bilocal residence pattern. Similarly, the Aché
are classified as sedentary/nomadic because of their recent transition
from nomadic foragers to sedentary horticulturalists.
The Language Family column provides the current linguistic
classification for the language traditionally spoken by these societies, and is
useful because linguistic affinity provides a rough measure of the cultural
relatedness of two groups. The classification of the Mapuche, Hadza, Tsimane, and New Guinean languages demand comment.
There is no general agreement about how to classify Mapudungun (the
Mapuche’s language) with the other language groups of South America. Similarly,
although Hadza was once considered a Khoisan language, distantly related to the
San languages of southern Africa, agreement about this is diminishing. The
Tsimane language resembles Moseten (the language of a Bolivian group similar to
the Tsimane), but otherwise these two seem unrelated to other South American
languages, except perhaps distantly to Panoan. Finally, because of the
linguistic diversity found in New Guinea, we have included for the Au and Gnau both
the language phylum, Torricelli, and their language family, Wapei.
The Complexity column refers to the anthropological
classification of societies according to their political economy (Johnson &
Earle 2000). Family-level societies consist of economically independent
families that lack any stable governing institutions or organizational decision-making
structures beyond the family. Societies classified as family plus extended
ties are similar to family-level societies, except that such groups also
use extended kin ties or nonkin alliances for specific purposes such as
warfare. In these circumstances, decision-making power remains ephemeral and usually
diffuse. Bands are composed of both related and unrelated families who
routinely cooperate in economic endeavors. Decision-making relies substantially
on group consensus, although the opinions of prestigious males often carry
substantial weight. Villages
and clans are both corporate groups of the same level of
complexity, and both are typically larger than bands. Clans are
organized around kinship, tracked by lineal descent from a common ancestor.
Decision-making power is often assigned according to lineage position, but
achieved status plays some role. Villages operate on the same scale of
social and political organization as clans, but usually consist of several
unrelated extended families. Decision making is often in the hands of a small
cadre of older, high-status men. At a larger scale of organization, multiclan
corporate groups are composed of several linked clans, and are
governed by a council of older high-status men – assignment to such councils is
often jointly determined by lineal descent, age, and achieved prestige.
Multiclan corporations sometimes act only to organize large groups in times of
war or conflict, and may or may not play an important economic role. Often
larger than multiclan corporations, chiefdoms are ruled by a single
individual or family and contain several ranked clans or villages. Both
individual ranks and that of clans/villages usually depend on real or customary
blood relations to the chief. Political integration and economic organization
in chiefdoms is more intense than in multiclan corporate groups, and chiefs
often require subjects to pay taxes or tribute.
The final two columns in Table 1, payoffs to cooperation
(PC) and aggregate market integration (AMI), refer to
rankings we constructed on the basis of ethnographic investigations; we explain
these in section 6.
4. Experimental results
4.1.
Substantial cross-cultural variability
The variability in Ultimatum Game behavior across the groups
in our study is larger than that previously observed in large-scale, industrialized
societies (e.g., Camerer 2003, chap. 2). Prior work comparing UG behavior among
university students from Pittsburgh, Ljubljana (Slovenia), Jerusalem, Tokyo
(Roth et al. 1991), and Yogyakarta (Indonesia; Cameron 1999) revealed little
group variation. In contrast, our UG results from 15 small-scale societies show
substantial variation, as is illustrated in Figure 2. Whereas mean UG offers in
standard experiments in industrialized societies are typically between 40% and
50% (see Table 2.2. in Camerer 2003), the mean offers from proposers in our
sample range from 26% to 58% – both below and above the “typical” behavior (Fig.
2; Table 2 presents additional details). Similarly, modal UG offers are
consistently 50% among university students, but our sample modes vary from 15%
to 50%, though the 50/50 offer is clearly popular in many groups. As a student
benchmark, we have included UG data from Roth et al.’s (1991) Pittsburgh study.2
Figure 2. A bubble plot showing the distribution of UG
offers for each group. The size of the bubble at each location along each row
represents the proportion of the sample that made a particular offer. The right
edge of the lightly shaded horizontal gray bar gives the mean offer for that
group. Looking across the Machiguenga row, for example, the mode is 0.15, the
secondary mode is 0.25, and the mean is 0.26.
Table 2. (Please click link below for Table 2)
http://www.bbsonline.org/Preprints/Henrich/Tables/Table02.pdf
On the responder side of the UG (Figure 3), rejection rates
are also quite variable. In some groups, rejections were extremely rare, even in the presence of low offers, but in others the
rejection rates were substantial and included frequent rejections of offers above
50%. Among the Kazakh, Quichua, Aché, and Tsimane, we observed zero rejections
out of 10, 14, 51, and 70 proposer offers, respectively. And although offers
among the Aché were mostly at or near 50%, they were at or below 30% for 57% of
the offers to Quichua and for 47% of offers to Tsimane – yet all were accepted.
Similarly, Machiguenga responders rejected only one offer, despite the fact
that more than 75% of their offers were below 30% of the pie. At the other end
of the rejection scale, Hadza rejected 24% of all offers and 43% (9/21) of
offers 20% and below. Unlike the Hadza and other groups who preferentially
rejected low offers, the Au and Gnau of Papua New Guinea rejected offers both
below and above 50%, with nearly equal frequency. University student
responders fall toward the upper end of the rejection scale (with more
rejections than average), but still they rejected less often than the Au, Gnau,
Sangu farmers, and Hadza, all of whom rejected positive offers with greater
frequency than did, for example, the Pittsburgh subjects in the study by Roth
et al. (1991).
Figure 3. Summary of responder’s behavior.
The lightly shaded bar represents the fraction of offers that were less than
20% of the pie. The length of the darker shaded bar gives the fraction of all
Ultimatum Game offers that were rejected, and the gray part of the darker
shaded bar gives the number of these low offers that were rejected as a
fraction of all offers. The low offers plotted for the Lamalera were sham
offers created by the investigator.
As in the UG, our data from Public Goods Games, which
include both VC and CPR versions, show much greater variation than previous
experiments in industrialized societies. Typical distributions of PGG contributions
from university students have a U-shape with the mode at full defection (zero
given to the group) and a secondary mode at full cooperation (everything to the
group). The format of the games does impact the results (e.g., people tend to
give more in the CPR version than in the VC version); nevertheless, the mean
contributions still usually end up between 40% and 60%. Table 3 shows that our
cross-cultural data provide some interesting contrasts with this pattern. The
Machiguenga, for example, have a mode at full defection but lack any fully
cooperative contributions, which yields a mean contribution of 22%. By direct
comparison, students at the University of Michigan (where the protocol and
experimenters were identical to those for the Machiguenga experiment) produced
the typical bimodal distribution, yielding a mean contribution of 43%. Both the
Aché and Tsimane experiments yielded means similar to those found in
industrialized societies, but the shape of their distributions could not have
been more different: they have unimodal, not bimodal, distributions. Their
distributions resemble inverted American distributions with few or no
contributions at zero or 100%. Like the Aché and Tsimane, the Huinca and Orma
show modes near the center of the distribution, at 40% and 50% respectively,
but they also have secondary peaks at full cooperation (100%) – and no
contributions at full defection.
Table 3. (Please click link below for Table 3)
http://www.bbsonline.org/Preprints/Henrich/Tables/Table03.pdf
4.2
Violations of the selfishness axiom
The selfishness axiom was violated in some way in every
society we studied, across all three experimental games (DG, UG, and PGG). Focusing
on the UG, either proposer or responder behavior, or both, violated the axiom.
Yet, responder behavior was consistent with selfish motives in several groups,
unlike typical university students. As shown in Table 2, responders from the
Aché, Tsimane, Machiguenga, Quichua, Orma, Sangu herders, and Kazakhs all have
rejection rates of less than 5%, roughly consistent with the canonical model.
For some groups these low rejection rates are not informative because all the
offers were near 50/50 (e.g., the Aché and Sangu), so no one in these groups
received a low offer. However, proposers in several societies made numerous low
offers that were not rejected. The selfishness axiom accurately predicts
responder behavior for about half of our societies, even though it generally
fails to predict the responder behavior of university students. Like university
students, the Au, Gnau, Sangu farmers, and Hadza subjects rejected positive
offers and thereby violated the axiom.
The issue of whether proposer behavior is consistent with
the selfishness axiom is more complicated. Table 2 and Figure 2 show that
proposers were not making offers consistent with the standard game theoretical
prediction based on the selfishness axiom, which requires that proposers offer
the smallest positive amount – because they believe that the responders are
seeking to maximize only their income from the game. In none of our societies
was this behavior common.
Perhaps, however, proposers’ behavior can be understood as
income maximizing given the presence of responders willing to reject low
offers. Among university subjects, it is generally thought that offers are
fairly consistent with expected income-maximizing strategies given the empirical distribution of
actual rejections across offers (Roth et al. 1991). Our results and analyses
suggest that this is unlikely to be the case in several of the groups studied.
For the groups in which at least one offer was rejected, we used the responder
data to estimate an income-maximizing offer (IMO), then
we compared this estimate to the group’s mean offers. Intuitively, the IMO is
the offer that an income-maximizing proposer would make assuming he knows the
distribution of what responders in his group will accept (and is neutral toward
economic risk, an important qualification we will return to shortly).
Figure 4 compares the actual mean offers from proposers (on
the y-axis) with their corresponding IMOs (calculated from responder data, on
the x-axis) for the various
societies. The mean offer/IMO pairs for each society are plotted as points next
to the societies’ names. Look first at the midpoint and ignore the ellipses
around them. Every group is above the unity line where mean UG offer =
IMO. This unity line is where the average offer would lie if the average offer
in each group were perfectly calibrated to that group’s empirical IMO. When the
mean UG offer is above the unity line, proposers are being “generous” given the likelihood of
rejection at each offer level (i.e., they are offering more than selfishness
alone would motivate them to offer).
Figure 4. Two-dimensional 90% confidence intervals of the mean
UG offers in various groups plotted against the expected income maximizing
offers (estimated from observed distribution of rejections). Intervals show
loci of possible mean offers and expected IMOs randomly resampled
(bootstrapped) from samples. We were unable to estimate the IMO for societies
with no rejections (Quichua, Tsimane, Ache, Kazakhs) or for societies in which
rejections bore no systematic relationship to offers (Au, Gnau).
To assess the statistical significance of how far mean
offers depart from the estimated IMO, each point in Figures 4a and 4b is
surrounded by an elliptical two-dimensional
90% “confidence interval”.3 A one-dimensional 90% confidence
interval is a range of numbers that has a 90% chance of containing the true
value of the statistic of interest. A two-dimensional interval is the same idea
extended to a pair of statistics. Using a statistical method called
“bootstrapping,” we can use the data we gathered to judge how differently the
results might have turned out if the experiment had been done (hypothetically)
over and over. The interval of bootstrapped values that results enables us to
judge how confident we can be that the mean offer would almost always be above
the IMO if our experiments were repeated.
Now we return to the question of whether the average offer
is above the IMO – that is, did proposers offer significantly more than they
had to, to maximize their earnings (given that some responders rejected low
offers)? That question is answered at a glance for a particular group by simply
observing whether the entire two-dimensional ellipse for that group lies above
and left of the 45-degree unity line.
The two graphs plot separately those societies in which we can be quite
confident the mean offer is clearly above the IMO (Figure 4a), and those for
which we cannot be fully confident the mean offer is truly above the IMO
(Figure 4b). Roughly half of the societies clearly lie to the upper left, with
their mean offers above their IMOs. The others also lie in the upper left, but
we cannot be too confident that their means are above their IMOs, even though the ellipses only
slightly overlap the 45-degree unity line for the Machiguenga and the Sangu
herders.4
It is possible that such high offers are consistent with a
more conventional extension of the selfishness axiom – an aversion toward
taking a chance on either getting a high or a low money payoff (“risk aversion”
in economic language). It is a common (though not universal) observation that
people prefer a certain amount of
money to a gamble with the same
expected payoff. Economists model this behavior by assuming that people seek to
maximize their expected utility, and that utility is a concave function of
income (diminishing psychophysical returns – earning an extra dollar is worth
less in utility terms on top of a lot of other dollars, compare to a smaller
number of dollars).
For example, suppose a subject estimates that an offer of
40% of the pie will be accepted for sure (leaving 60% for the proposer), and
that an offer of 10% will be accepted with probability 2/3. If she were risk
averse, she might value the certainty of keeping 60% of the pie more than the
2/3 chance of keeping 90% (and a 1/3 chance of getting nothing). In this case
the expected monetary gain is the same for the two offers (namely, 60% of the
pie), but the expected utility of the certain outcome is greater.
Thus, a highly risk averse subject might make a high offer even if the
probability of rejection of a low offer were small.
To explore whether risk aversion can explain the fact that
average offers are so much higher than IMOs in most of our samples, we measured
the degree of risk aversion both indirectly and directly. The indirect
measurement asks what degree of aversion toward risk is necessary to make the
risk-adjusted IMO equal to the mean offer. To answer this we transformed the
game payoffs into utilities, by assuming that the utility function for money is
a power function xr of the money amount x,
with. r the standard measure of the degree of risk aversion. For each group we estimated the value of r that would make
the observed mean offer a utility maximizing offer given the distribution of actual
rejection frequencies.5
As noted in Figure 4b, the Hadza and the Sangu farmers were
approximately expected income maximizers – that is, their average offers were consistent
with expected utility maximization for risk-neutral individuals. But for the
other groups – Orma, Sangu herders, Machiguenga, Mapuche, and Shona – the
implied levels of risk aversion were implausibly high. Even for the least
extreme case, the Shona, the degree of risk aversion necessary to make their
behavior consistent with expected utility maximization implies that they would
be indifferent between an even chance that an offer of 1 out of 10 dollars
would be accepted (an expected payoff of $4.5) and getting only $.04 for sure.6
Clearly, an individual with this degree of risk aversion would be unable to
function in variable environments.
Risk aversion was also measured directly among the
Mapuche and the Sangu. Subjects were
offered a series of risky choices between gambles with different probabilities
of monetary payoffs to numerically calibrate their degree of aversion toward
economic risk (Henrich & McElreath, 2002; Henrich & Smith, 2004). In
neither society did measured risk preferences predict offers. Moreover, in both
societies, subjects were risk preferring (formally, r >
1) rather than risk averse, a fact that casts further doubt on the risk-aversion
interpretation. We conclude that our offers are not explained by risk aversion in
the usual sense (i.e., concave utility functions defined over gamble
income, xr with r < 1).
Instead, high offers may reflect a desire to avoid rejections because of an
aversion to social conflict or a fear that a rejection is an awkward insult,
rather than because of an aversion to variance in monetary outcomes (as in the
economic model).
Alternatively, perhaps because proposers are not sure how
likely responders are to reject, they offer more to be on the safe side. This
tendency to behave cautiously in the face of unknown odds (“ambiguity” in
economic language) is consistent with many other types of experimental data and
economic phenomena (Camerer & Weber 1992). In our settings,
ambiguity-aversion toward rejection is plausible because the proposers do not
see all the rejection frequencies that we observe. Whether
ambiguity aversion can explain the high mean offers can be judged using the
bootstrapping results shown in Figures 4a and 4b. That exercise produces
1,000 different estimates of IMOs. Think of these as expressing the range of
possible beliefs about rejections which an uncertain proposer might entertain,
and the optimal offers those wide-ranging beliefs imply. We can then ask: How
pessimistic would proposers have to be to justify the mean offer as expected–income
maximizing given pessimistic beliefs? A simple way to answer this question is
to ask what fraction of the IMOs is above the mean offer. For most of the
groups for which we can estimate IMOs at all, the results are striking: For the
Achuar, Shona, Orma, Sangu herders, Machiguenga, and Mapuche, the mean offer is
just slightly above the most pessimistic IMO among the 1,000 simulated
ones (which occurs when all the resampled offers are rejected). The mean
offers/maximum IMO pairs are, respectively, 0.42/0.30, 0.44/0.40, 0.43/0.40,
0.41/0.33, 0.26/0.25, and 0.335/0.33. It is as if subjects have a good guess
about the highest offer that could be rejected, act as if that offer
will be rejected for sure, and offer just above it to avoid rejection.
Therefore, although the gap between mean offers and IMOs visible in Figures 4a
and 4b cannot be explained by risk aversion because of the concavity of the
utility function for money, it can be explained as the result of pessimism
about rejection frequencies and aversion to ambiguity.
For four groups (the Aché, Tsimane, Kazakhs, and Quichua) we
could not estimate the IMOs because there were no rejections. Nevertheless, as we have discussed, it seems
likely that substantially lower offers would have been accepted. On that ground,
offers in these groups cannot be explained by narrow self interest. Among the
Au and Gnau, the IMO could not be established because responders from these
groups did not preferentially accept higher offers, which is perhaps an even
more striking violation of the selfishness axiom.
Additional evidence against the selfishness axiom comes from
our three Dictator Games: the results here are more transparent than for the UG
because the proposer is simply giving money away, anonymously, with no
possibility of rejection. In each of the three groups in which the DG was
played, offers deviated from the typical behavior of university students and
from the predictions of self-regarding models. Mean offers among the Orma,
Hadza, and Tsimane were 31, 20, and 32 percent, respectively, of the stake.
These mean Dictator offers were 70, 60, and 86 percent of the corresponding
mean UG offers for these groups. Few or none of the subjects in these societies
offered zero, whereas the modal offer among university students is typically
zero (Camerer 2003).7
Finally, the results from all six of our Public Goods Games
also conflict with the selfishness axiom, with means ranging from 22% among the
Machiguenga to 65% among the Aché – see Table 3. Even the Machiguenga data show
62% of the sample violating the income-maximizing prediction of 0%. Among the
other groups, no group had more than 5% of the sample making contributions of
zero. To our knowledge, this is never seen in one-shot PGGs among students,
where a large percentage of players (usually the mode) give zero.
4.3.
Methodological variations between sites
Because our experiments were conducted at remote field sites
with diverse, largely uneducated participants, we used some discretion in
conducting the experiments to ensure comprehension and internal validity. The
result was some methodological variation across sites. For the UG, Table 4
documents the potentially important dimensions of variation in the
administration of the experiments—note, we have grouped by the “researcher”
here, rather than the “society,” as this is locus of methodological variation.
These variations fall into eight categories. First, column two lists the three
ways that the instructions used by different experimenters explained the
allocation of the initial sum of money between the proposer and responder. In nine
of our societies, the instructions stated that the money was allocated “to the
pair”; in five societies the money was allocated “to the first person” (the
proposer). Experimental economists have used both of these versions in their
many UG experiments, and the results do not show any significant variation.
Finally, instructions among the Shona (Barr 2004) left the allocation of money
ambiguous.
Table 4. (Please click link below for Table 4)
http://www.bbsonline.org/Preprints/Henrich/Tables/Table04.pdf
A second kind of variation is outlined in column 3, which
shows that most of our researchers stuck to entirely abstract explanations of
the game and experimental context, using no explicit (and intentional) framing,
but two ethnographers did use some contextualization or framing in the games.
To ensure comprehension among the Aché, Hill created an analogy between the UG
and the process used by the Aché for apportioning the subcutaneous fat of game
animals (Hill & Gurven 2004). More indirectly, to attract Achuar and
Quichua to the game, Patton called for a Minga, which, among these
groups, is called to bring people together for cooperative work projects such
as cutting a field for planting (Patton 2004).
In a third kind of variation, five researchers read the
instructions to a group first, and then brought the individuals into a gaming
area to have their comprehension tested and make their decisions. Six other researchers
explained the games to individuals only after they had entered the gaming area,
and explained nothing to the group. Among the Machiguenga, Henrich (2000) used
both methods and found no difference. Among university students this
modification makes no difference.
Fourth, the difficulty of bringing all players together at
the same time led four researchers to conduct their experiments from
house-to-house or one-by-one, sometimes spreading the games out over a few
weeks. However, in nine other societies everyone was brought together in a
single gaming area. Among the Machiguenga, both methods were used and no
difference was found. Among students, this procedural variation does not impact
the results (Henrich 2000; Henrich & Smith 2004).
Fifth, in all of our UG experiments, participants divided
sums of cash, except in Lamalera. There, to avoid the appearance of
gambling, packs of cigarettes (which can be readily traded) were used as the
medium of exchange (Alvard 2004).
Sixth, a few of our ethnographers, desiring to explore
whether low offers would be rejected, fabricated offers for responders.9
Seventh, along with the money from the game itself, players
in seven groups were paid a flat fee for ‘showing up’ to the experiment (which
subjects received regardless of what happened in the game). The eight other
groups received only money from decisions in the games. U.S. research suggests that show-up fees do
not have an important impact on UG play (Henrich & Smith 2004; Henrich
& Henrich in press).
Finally, one-on-one post-game interviews (to explore what
people thought of the games, and why they did what they did, etc.) were conducted
extensively in six societies, somewhat in five, and not at all in three groups.
In one group, the Shona, Barr (2004) used post-game focus groups.
Three lines of argument suggest that these methodological variations cannot account for the broad patterns of variation we observed. First, there is no reliable correspondence between methodological variations across groups in the UG and their game behavior (compare Tables 2 and 4).
Second, as noted, many of these variations do not produce
substantial differences in the populations where they have been tested.10
Third, in several cases in which the identical protocols and experimenters were
used in different places, the results still show substantial variation. The
following subsets faced the identical experimenters and protocols and
still showed substantial variation: (1) Machiguenga, UCLA students (a student
control; see Henrich 2000) and the Mapuche (Henrich: these three yielded UG
mean offers of 26, 48, and 34%, respectively), (2) the Quichua and Achuar
(Patton: UG mean offers of 25 and 43%). The same can be said of the PGG data,
where the same experimenters and protocols were used for comparisons of the
Machiguenga with the Michigan students, and of the Huinca with the Mapuche.
Moreover, within our linguistic groups, individual researchers found
substantial variation between communities (Tsimane, Sangu, Shona, and Hadza),
which is discussed further in section 7. By the same token, however, the same
experimenters and protocols did not always find between-group variation, as
these comparisons attest: (1) Kazaks and Mongols (Gil-White) and (2) the Au and
Gnau (Tracer).
Third, it is also important to realize that UG results from
industrialized societies are generally quite robust against a wide range of
procedural variations (which is why we selected the UG for the project!).11
Many experimentalists have highlighted significant differences in framing effects
for the UG, but the size of these differences is almost always small compared
to the kinds of differences we found cross-culturally (Camerer 2003: chap. 2).
Thus, “significant” effects should not be confused with big effects (and one
should also consider that treatments that result in nonsignificant differences
will rarely see the light of day). The largest of these effects (among
university students) involves substantial manipulations, such as including a
pregame trivia contest to determine who is to be the proposer. Under these
conditions, proposers offer less, and responders accept lower offers (Hoffman
et al. 1994). Certain contextualizations (e.g., a monopoly seller choosing a
price) have a modest effect on offers, shifting the mean by about 10% of the
pie (Camerer 2003, chap. 2; Hoffman et al. 1994). Other seemingly important
variations actually have little effect on offers (Larrick & Blount 1997). For
example, playing repeatedly (with feedback about one’s own results) or
increasing stakes by up to a factor of 25 changes offers by only 1–2% of the
stake, and does not effect the modal offer. In contrast, moving the identical
protocol from the Machiguenga to UCLA increases offers by 24% of the stake, and
moves the mode from 15% to 50%.
It is important to realize that the few variations in UG
instructions or procedures that have shown a substantial impact on past results
were deliberately designed by researchers because they suspected that
such variations might cause a big effect. In contrast, our researchers tried to
avoid any modifications that might have an effect, and our variations were
typically ad hoc procedures created by field researchers in adapting to the
field situation, or inadvertent nuances due to translation. Such variations do
not result, for example, in accidentally slipping a trivia contest, which
determines who the proposer, into the instructions.
A final methodological concern in interpreting the
cross-cultural results comes from possible experimenter bias. The relationships
between our experimenters and the participants are typically much closer, more
personal, and longer lasting than in university-based experiments.
Consequently, it is possible that ethnographers may bias the results of our
experiments in ways different from those found in standard situations. However,
two pieces of data argue against this interpretation. First, Henrich (2000)
attempted to control for some of this effect by replicating the Machiguenga UG
protocol with UCLA graduate students. In this control, Henrich and his subjects
knew one another, had interacted in the past, and would interact again in the
future. His results were quite similar to typical UG results in high-stakes
games among adults in the United States, and substantially different from the
Machiguenga. This is certainly not a complete control for experimenter bias,
but it does confront some elements of the bias. Second, to test for
experimenter bias across our samples, we examined the relationship between the
time each experimenter had spent in the field prior to administering the games
and the mean UG of each group, but found no consistent pattern in the data.
Finally, since most people would predict that having some longer term
relationship with the experimenter would bias offers toward generosity, and
most of our variation is more selfish than university student results, it is
difficult to argue that such a bias is driving the results. Nonetheless, we
cannot entirely exclude the possibility that some of the observed between-group
differences result from differences among the experimenters and the details of
how the experiments were implemented.
5. Explaining group differences in behavior
To examine the variation between groups, we first examined
whether any attributes of individuals were statistically associated with
proposer offers across our sample. Among the measured individual
characteristics that we thought might explain offers were the proposer’s sex,
age, level of formal education, and wealth relative to others in their group.12
In pooled regressions across all offers none of these individual-level
variables predicted offers once we allowed for group-level differences in
offers (by introducing group dummy variables). Since the group dummy variables
account for approximately 12% of the variance in individual offers, we conclude
that group differences are important. However, for the moment, we remain
agnostic about the role of individual differences. Our pooled regression tested
for common effects of these variables across all the groups and hence does not
exclude the possibility that the individual differences we have measured may
predict behaviors in different ways from group to group. We return to this in
section 6.
In proposing this project, we hypothesized that differences
in economic organization and independence, social organization (complexity),
and market integration may influence cultural transmission and create
between-group differences in notions of fairness and punishment.13 To test these initial hypotheses, we rank ordered our
societies along five dimensions. First, payoffs to cooperation (PC): To what
degree does economic life depend on cooperation with nonimmediate kin? In a
sense, PC measures the presence of extrafamilial cooperative institutions.
Groups like the Machiguenga and Tsimane ranked the lowest because they are
almost entirely economically independent at the family level. In contrast, the
economy of the whale hunters on Lamalera depends on the cooperation of large
groups of nonkin. Second, market integration (MI):– Do people engage
frequently in market exchange? Hadza foragers were ranked low because their
life would change little if markets suddenly disappeared. Others, like the
Orma, were ranked higher because they frequently buy and sell livestock and
work for wages. Third, anonymity (AN): How important are anonymous roles
and transactions? Many Achuar of the Ecuadorian Amazon never interact with
strangers, unlike the Shona of Zimbabwe who frequently interact with people
they do not know and may never see again. Fourth, privacy: How well can
people keep their activities secret from others? In groups like the Au, Gnau,
and Hadza, who live in small villages or bands and eat in public, it is nearly
impossible to keep secrets and quite difficult to hide anything of value. Among
the Hadza, simply having pants increases privacy because they have pockets. In
contrast, Mapuche farmers live in widely scattered houses and maintain strict
rules about approaching another’s house without permission, so privacy is
substantial. Fifth, sociopolitical complexity (SC): How much decision making
occurs above the level of the household? Because of the importance in the
anthropological literature of the classifications of societies by their
political complexity (Johnson & Earle 2000), we ranked our societies from
family level through chiefdoms and states. Finally, settlement size (SS)
– the size of local settlements, which ranged from fewer than 100 members among
the Hadza to more than 1,000 on Lamalera.
Before beginning the data analysis we ranked the groups
along these dimensions using the following procedures. First, during a meeting
of the research team, we had a lengthy discussion of the underlying attributes
that each dimension was designed to capture. Then the field researchers lined
up and sorted themselves by repeatedly comparing the group they studied with
those of their two neighbors in line, switching places as necessary, and
repeating the process until no one needed to move. The subjective nature of the
resulting ordinal measures is evident.14 Second, our complexity
rankings were generated by both Henrich (who was not blind to our experimental
results) and Allen Johnson, an outside expert on societal complexity, who was
blind to the results. Henrich’s and Johnson’s rankings correlated 0.9, and
explain about the same amount of variation in mean UG offers.
We have no way of knowing the direction of causality between
the measures of social structure and offers. An ideal way to disentangle
causality is to have an exogenous variation in structural conditions and
correlate it with offers (what econometricians call an “instrumental
variable”). The time course of history
in these societies does not permit such an inference.
As can be seen in Table 5, four of the indices – market integration,
anonymity, social complexity, and settlement size – are highly correlated across groups,
suggesting that they may all result from the same underlying causal process.
The correlation of each of these variables with the potential payoffs to
cooperation is very small, suggesting that this ranking measures a second set
of causal factors. This is not surprising. An increase in social scale is
associated with a shift to a market-based economy and an increase in anonymity.
Within small-scale societies with similar levels of social complexity, there is
a wide range of economic systems with varying levels of cooperation. To capture
the causal effects of this nexus of variables, we created a new index of
“aggregate market integration” (AMI) by averaging the ranks of MI, SS, and SC.
(We did not include AN because it is so similar to MI,
and including it has only a slight effect.)
Table 5. (Please click link below for Table 5)
http://www.bbsonline.org/Preprints/Henrich/Tables/Table05.pdf
We estimated ordinary least squares regression equations for
explaining group mean UG offers using the PC and AMI. Both of their normalized regression coefficients are highly significant and indicate
that a standard deviation difference in either variable is associated with
roughly half of a standard deviation difference in the group mean offers (Table
6; Figure 5). Together these two variables account for 47% (adjusted R2) of the variance among societies
in mean UG offers. The magnitude of these coefficients, and their significance,
is robust to three different checks on the analysis.15
Table 6. (Please click link below for Table 6)
http://www.bbsonline.org/Preprints/Henrich/Tables/Table06.pdf
Figure 5.
Plots Mean UG offers as a function of the PC and AMI indices. Because AMI and
PC are almost uncorrelated (r = .09),
these bivariate plots give a useful picture of their effects.
All regressions using PC and one of the other predictors
(AN, MI, SC, or SS) yielded a significant positive coefficient for PC and a
positive, nearly significant, coefficient for the other variable. If we use the
income maximizing offer as a predictor of the UG offers along with PC and AMI, we
find that the IMO’s coefficient is smaller (in magnitude), negative, and
insignificant, whereas the coefficients of PC and AMI remain large and close to
significance at conventional levels (even though for IMO n = 9),
suggesting that the effects of economic structure and cultural differences
captured by PC and AMI do not substantially influence offers through the IMO.
The same two variables (PC and AMI) also predict the group
average IMO; the effect sizes are large (normalized regression coefficients
about one half) but very imprecisely estimated (significant only at the 20%
level). Taken at face value, these estimates suggest that subjects'
expectations about the likelihood that low offers will be rejected covaries
with both the benefits of cooperation and aggregate market integration.
Our analysis of the individual-level responder data across
all groups reveals some of the same basic patterns observed in the proposer
data. The age, sex, and relative wealth of a responder does
not affect an individual’s likelihood of rejecting an offer across our entire
sample. What does matter is the proportion of the stake offered and the
responders’ ethnolinguistic group.
6. Explaining individual differences within
groups
In contrast to the power of our group-level measures in
statistically explaining between-group differences in experimental behaviors,
our individual-level variables explain little of the variation within or across
groups. With a few group-specific exceptions, nothing we measured about
individuals other than their group membership (society, village, camp, or other
subgroup membership) predicted experimental behavior. Here we summarize our
findings concerning individual attributes and experimental play in within-group
analyses. Sex, wealth, and age do not generally account for any significant
portion of the variance in game play. However, in the UG, sex was marginally
significant among the Tsimane, where males offered 10% more than females (Gurven
2004). Among the Hadza, women’s UG offers strongly increased with camp
population size, but camp size was not important to men’s offers. Conversely,
in the DG, it was the offers of Hadza men that increased with camp size
(Marlowe 2004). As in the UG, Public Goods Game data from five societies also
reveal no significant effects of sex, except among Aché men who contributed a
bit more than did the women (Hill & Gurven 2004). Similarly, wealth in any
form (e.g., cash, cows, land) fails to predict game behavior. In several
circumstances, multiple measures of wealth (e.g., animal wealth, cash, and land
wealth) were gathered and analyzed, as well as an aggregate measure. In these
within-group analyses, wealth arose as significant only once in 12 different
data sets, including both UG and PGG games. The exception comes from an
all-male Public Goods Game among the Orma. Controlling for age, education,
income, and residence pattern (sedentary vs. nomadic), wealth was the only
significant predictor of contributions in a multivariate linear regression,
with a standard deviation difference in wealth predicting well over half a
standard deviation increase in contributions (Ensminger 2004). We make sense of
this finding below.
Several researchers also analyzed measures of formal
education. Analyzing UG data from the Sangu, Orma, Mapuche, Au, and Gnau, we
find that the extent of schooling does not account for any significant portion
of the variation in offers in either bivariate analyses or multivariate
regressions that controlled for sex, age, and wealth. Among the Tsimane, the
extent of formal education emerges as marginally significant in a multivariate
regression involving age, village, sex, Spanish-speaking ability, trips to the
nearest market town, and wage labor participation. Less-educated Tsimane offered
more in the UG game. However, we find no effect of formal education on
PGG play among the Tsimane. Therefore, although schooling effects may exist,
they are neither particularly strong nor consistent across games or societies.
Although our group-level measure of aggregate market
integration has solid statistical power, individual-level measures of market
exposure do not explain any significant proportion of the variation within
groups. To assess market integration, some researchers gathered data on
individuals’ participation in wage labor, their reliance on cash cropping, and
their competence in the national language. Wage labor participation shows no
significant relation to UG offers in six groups: the Tsimane, Aché, Gnau, Au,
Machiguenga, and Mapuche. PGG data from the Orma, Aché, Machiguenga, and
Tsimane also show that wage labor does not influence game play. The only clear
exception to the wage labor pattern occurs in the Orma UG data, where
individuals who participate in wage labor (to any degree) make significantly
higher offers than those who do not (Ensminger 2004).
In societies based on agriculture, another measure of market
integration is the amount of land an individual (or household) devotes to cash
cropping, as opposed to subsistence cropping. We obtained cash cropping data
from three societies. Among the Machiguenga, land (in hectares or as a
proportion of total land) devoted to cash cropping is positively correlated
with UG offers; its normalized partial regression coefficient when age, sex,
and wage labor are controlled remains substantial, though its significance
level is marginal (Henrich & Smith 2004). Neither total cash-cropping land
nor the proportion of land devoted to cash cropping is significantly related to
UG offers for the Au and Gnau. However, among the Au (but not the Gnau)
multivariate regressions show that land devoted to subsistence cropping
positively predicts UG offers, controlling for sex, age, cash cropping land,
and wage labor (Tracer 2003; 2004).
In many places, an individual’s degree of competence in the
national language may also represent a measure of market integration, or at
least of market exposure. We have language data only from the Tsimane, and
though it is significant in bivariate analyses, multivariate regressions that
control for village membership, sex, age, visits to San Borja, years of formal
education, and participation in wage labor show no relationship between
Spanish-speaking ability and UG offers. Using the same controls, competence in
the national language also fails to predict PGG contributions (Gurven 2004).
As is the case for all of our individual-level data, except
for age and sex, these measures capture individual behaviors that may well be
endogenous with respect to the beliefs or preferences our experiments measure.
Because it is possible that these measures are the consequence rather than the
cause of individual behavioral differences, we also sought to use geographical
measures of proximity to market opportunities as exogenous instruments for
measuring market exposure in three groups: the Tsimane, Au, and Gnau. None of
these were significant predictors of proposer behavior.
It is possible, of course, that the unexplained within-group
variance in experimental behaviors reflects a lack of comprehension of the game
or errors in experimental play that are unrelated to measures like age, wealth,
or wage labor participation. Overall, we have little reason to suspect that
game comprehension significantly influenced the results (although see Gil-White
2004). In most cases experimenters tested subjects for game comprehension
before the experiments were implemented, and excluded those who had difficulty
grasping the game. In several studies, experimenters used post-game interviews
to probe for possible misunderstandings and faulty assumptions. Among the
Mapuche, the players who passed the basic tests were ranked according to how
well they understood the strategic nature of the game and how well they were
able to do the monetary calculations involved. Neither measure predicts game
behavior or deviation from mean game behavior. Similarly, among the Hadza (F.
Marlowe 2004), players were scored according to the number of practice examples
it took for them to learn the game. Among Hadza males this measure is unrelated
to both UG proposer and responder behavior, but for females comprehension is
positively and significantly correlated with offer size. We do not know if the
covariation of comprehension and experimental behavior among Hadza women
represents the effect of comprehension per se or results from the association
of comprehension with other correlates of game play for women, such as camp
size (a strong predictor of Hadza women’s offers). Finally, as noted above, education,
which might be thought to correlate with degree of game comprehension, did not
predict game behavior.
Given that we sought individual-level statistical
associations for a number of variables in 15 societies and found just a handful
of estimates suggesting substantial effects, we conclude that, other than group
membership, the individual-level facts we have collected about our subjects do
not consistently predict how individuals will behave. This does not mean that
within-group variation in subjects’ behavior cannot be explained; rather it
suggests that the explanation may be group-specific or that we may not have
collected the appropriate information, or both.
7. Local group effects
Our analysis suggests that group-effects may be important,
and this opens the question of how to define a group. In the above analyses,
ethnolinguistic markers were used to define group membership, but
nonethnolinguistic regional groupings or smaller local clusters (e.g.,
villages) may be more appropriate. Our data allow some comparisons. Such
small-scale tests permit us to control for a number of variables, including
climate, language, regional/national economy, local buying power of the game
stakes, and local history. In the Bolivian Amazon, the effects of market integration
on local groups were examined by performing the UG and PGG in five villages at
different distances from the market town of San Borja, the only source of
commercial goods, medicines, and wage labor opportunities. Like the
Machiguenga, the Tsimane live in small communities scattered along a major
riverine drainage system. Under these circumstances, physical distance (in
travel time along the river) from San Borja provides a proxy measure for the
extent of market contact of different Tsimane communities. As noted, the
results indicate that a community’s distance from San Borja is unrelated to UG
or PGG behavior. Interestingly, however, the best predictor for UG offers and
PGG contributions is what community one is from, independent of the
community’s distance from San Borja and population size. So where a Tsimane
lives matters, but differences in both individual-level measures of market
integration and community-level market variables apparently do not. Among the
Tsimane, the relevant group for predicting UG and PGG behavior appears to be
smaller than the ethnolinguistic group.
We found a number of other cases in which group membership
effects were strong even in the absence of geographical isolation, suggesting
that the processes that generate and maintain behavioral differences among
groups can maintain differences between frequently interacting, and even
intermarrying, groups. In Chile, Mapuche farmers and non-Mapuche Chilean
townspeople, locally called Huinca, have lived side-by-side, intermarried, and
interacted for over 100 years. Yet, the Mapuche and the Huinca behaved quite
differently in a single-shot PGG game. The Mapuche contributed a mean of 33% to
the pot, the Huinca offered an average of 58%. In
Ecuador the Achuar and Quichua of Conambo, who interact and intermarry
frequently, played the UG quite differently: Achuar proposers offered a mean of
43% while Quichua proposers offered only 25%. This difference is especially
notable as Quichua and Achuar subjects were randomly paired, so the proposers
from the two groups faced the same probability of rejection. In Tanzania, Hadza
from the biggest camp (which was three times larger than the next largest camp)
played the UG much more like university students than like Hadza from the four
smaller camps, despite the fact that camps are ephemeral social units and camp
membership is quite fluid. For the Hadza, camp population size turns out to be
the best predictor of UG offers – the larger the camp, the higher the mean UG
offer. Finally, although Sangu herders and farmers made similar UG offers,
farmers rejected offers more frequently than herders. Yet, Sangu often change
from herder to farmer and back again over the course of one lifetime.
In contrast, local groups in some locations showed little or
no between-group variation. In Mongolia, the Torguud Mongols and Kazakhs are
separated by deep cultural and historical differences, yet they played the UG
similarly. In Papua New Guinea, the Au and Gnau, who speak mutually
unintelligible languages and show differing degrees of market incorporation,
played the UG in the same unusual manner (making and rejecting offers over
50%). Similar comparisons in Zimbabwe between resettled and unresettled Shona
reveal only slight differences.
In general, the micro level variation we observed contrasts
with the UG results from the United States and Europe in which university
students, who speak different languages and live thousands of miles apart,
behave quite similarly. Of course, it is possible that variation exists within
contemporary societies, but that this variation is not represented in
university populations (Ferraro & Cummings 2005). Nevertheless, recent UG
experiments with adult subjects outside of universities have failed to uncover
behavioral patterns in the UG much different from those observed among
university students (Carpenter et al. 2005; Henrich & Henrich in press).
8. Experimental behavior and everyday life
The fact that group-level measures of economic and social
structure statistically explain much of the between-group variance in
experimental play suggests that there may be a relationship between game
behavior and patterns of daily life in these places. In several cases the
parallels are striking, and in some cases our subjects readily discerned the
similarity and were able to articulate it. The Orma,
for example, immediately recognized that the PGG was similar to the harambee,
a locally initiated contribution that Orma households make when their community
decides to pursue a public good, such as constructing a road or school. They
dubbed the experiment “the harambee game” and contributed generously
(mean 58% with 25% full contributors).
Recall that among the Au and Gnau
of Papua New Guinea many proposers offered more than half the pie, and many of
these offers were rejected. The making and rejection of seemingly generous
offers, of more than half, may have a parallel in the culture of
status-seeking through gift giving found in Au and Gnau villages and throughout
Melanesia. In these societies, accepting gifts, even unsolicited ones, implies
a strong obligation to reciprocate at some future time. Unrepaid debts
accumulate and place the receiver in a subordinate status. Further, the giver
may demand repayment at times or in forms (political alliances) not to the receiver’s liking, but the receiver is still strongly
obliged to respond. As a consequence, excessively large gifts, especially
unsolicited ones, will frequently be refused. Together, this suggests that as a
result of growing up in such societies, individuals may have acquired values,
preferences, or expectations that explain both high offers and the rejection of
high offers in a one-shot game. Interestingly, it may turn out that what is
unique here is not the rejection of high offers (ethnographically, many
societies disdain excess generosity), but the willingness to make offers of
more than 50%.
Among the whale hunting peoples on the island of Lamalera
(Indonesia), 63% of the proposers in the Ultimatum Game divided the pie
equally, and most of those who did not, offered more than half (the mean offer
was 58% of the pie). In real life, when a Lamalera whaling crew returns with a
large catch, a designated person meticulously divides the prey into
predesignated parts allocated to the harpooner, crewmembers, and others
participating in the hunt, as well as to the sailmaker, members of the hunters’
corporate group, and other community members (who make no direct contribution
to the hunt). Because the size of the pie in the Lamalera experiments was the
equivalent of 10 days’ wages, making an experimental offer in the UG may have
seemed similar to dividing a whale.
Similarly, in Paraguay the Aché regularly share meat. During
this sharing, the hunters responsible for the meat forgo their share, and the
prey is distributed equally among all other households. There is no consistent
relationship between the amount a hunter brings back and the amount his family
receives (Kaplan & Hill 1985). Successful hunters often leave their prey
outside the camp to be discovered by others, carefully avoiding any hint of
boastfulness. When asked to divide the UG pie, Aché proposers may have
perceived themselves as dividing the game (meat) they or a male member of their
family had acquired, thereby leading 79% of the Aché proposers to offer either
half or 40%, and 16% to offer more than 50%, with no rejected offers.
By contrast, the low offers and high rejection rates of the
Hadza, another group of small-scale foragers, are not surprising in light of
the numerous ethnographic descriptions (F. Marlowe 2004; Woodburn 1968). While
the Hadza extensively share meat (and other foods to a lesser degree),they do not do so without complaint and many look for
opportunities to avoid sharing. Hunters sometimes wait on the outskirts of camp
until nightfall so they can sneak meat into their shelter (F. W. Marlowe 2004).
The Hadza share because they fear the social consequences that would result
from not sharing. Cooperation and sharing are enforced by a fear of punishment
that comes in the form of informal social sanctions, gossip, and ostracism
(Blurton Jones 1984; 1987). Many Hadza proposers tried to avoid sharing, and
several of them were punished by rejection. Thus, we find two foraging peoples,
the Aché and the Hadza, at opposite ends of the UG spectrum in both offers and
rejections, with each seeming to reflect their differing patterns of everyday
life.
Similarly, both the Tsimane and Machiguenga live in
societies with little cooperation, sharing, or exchange beyond the family unit.
Ethnographically, both groups demonstrate little fear of social sanctions and
seem to care little about local opinion. The Machiguenga, for example, did not
even have personal names until recently – presumably because there was little
reason to refer to people outside of one’s kin circle (Johnson 2003).
Consequently, it is not very surprising that in an anonymous interaction both
groups made low UG offers. Given that Tsimane UG offers vary across villages,
it would be interesting to know if these differences reflect village-level
differences in real prosocial behavior.
Although methodological discussions commonly address the
correspondence of experimental regularities to behavior in naturally occurring
economic interactions (Camerer 1996; Loewenstein 1999), our concern here is
more modest: to explore the possibility of a connection between patterns of
behavior in the experiments and those in the daily lives of our subjects. In
many societies it appears that there may be such a connection, and that
sometimes our subjects were able to verbalize those parallels.
9. Discussion: Theoretical implications
Understanding the patterns in our results calls for
incorporating proximate-level decision-making models from behavioral economics,
which have increasingly drawn insights on human motivation and reasoning from
psychology and neuroscience (Camerer 2003; de Quervain et al. 2004; Sanfey et
al. 2003), under the ultimate-level evolutionary umbrella created by
culture-gene coevolutionary theory (Baldwin 1896; Boyd & Richerson 1985;
Campbell 1965; Cavalli-Sforza & Feldman 1981; Durham 1991; Pulliam &
Dunford 1980). Coevolutionary theory treats genes and culture as intertwined
informational systems subject to dual evolutionary forces. In our species,
cultural capacities are best understood as sophisticated social learning
mechanisms (Tomasello et al., in press) for acquiring, at low cost, locally
adaptive behaviors or decision information. Because these forms of social
learning create cumulative evolutionary products over generations (e.g.,
technologies), multiply stable equilibria in social interactions (e.g.,
institutional forms), and operate on much shorter time scales than genetic
evolution (Boyd & Richerson 1996; Gintis 2003a; Tomasello 1999), cultural
evolution and its products have undoubtedly influenced the human genotype
(Bowles & Gintis 2004). This theoretical avenue predicts that humans should
be equipped with learning mechanisms designed to accurately and efficiently
acquire the motivations and preferences applicable to the local set of
culturally evolved social equilibria (institutions).
Behavioral game theory – the subdiscipline from which our
methods derive – is rooted in the notion that individuals will select among
alternatives by weighing how well the possible outcomes of each option meet
their goals and desires. Theoretically, this is operationalized by assuming
agents maximize a preference function subject to informational and
material constraints. Behavioral game theory shows that by varying the
constraints and the rewards, as assessed by the agent’s preference function, as
we do in such games as the UG and PGG (Charness & Rabin 2000; Fehr &
Schmidt 1999), we can determine the arguments of the agent’s preference
function and how the agent trades off among desired rewards. We call this the
preferences, beliefs, and constraints approach.
It is often thought that this preferences, beliefs, and
constraints approach presumes that individuals are self-regarding, or that they
have very high levels of reasoning or omniscience, or both. However, though this
has often been true of many models, these assumptions are certainly not
necessary. Indeed our research (along with much other work) shows that such
considerations as fairness, sympathy, and equity are critical for understanding
the preference functions of many humans, and can be effectively integrated with
such things as pleasure, security, and fitness to produce a more complete
understanding of human behavior. Similarly, these models do not necessarily
presume anything in the way of reasoning ability beyond that required to
understand and perform in everyday social contexts.
The relationship between culture-gene coevolutionary theory
and the preferences, beliefs, and constraints approach is straightforward,
although rarely illuminated. As background, evolutionary game theory has shown
that social interactions among populations of individuals with adaptive
learning mechanisms often produce multiple stable social equilibria (Fudenberg
& Levine 1998; Gintis 2000; Weibull 1995; Young 1998). As different human
ancestral groups spread across the globe and adapted their behavioral
repertoire to every major habitat from the malarial swamps of New Guinea to the
frozen tundra of the Siberian Arctic, they would have, over time, culturally
evolved different social equilibria (forms of social organizations and
institutions).16 As a consequence, ancestral humans would have
needed to adapt themselves ontogenetically to the vast range of potential
social equilibria that one might encounter upon entering the world. The result
of dealing with this adaptive problem, we argue, is that humans are endowed
with cultural learning capacities that allow us to acquire the beliefs and
preferences appropriate for the local social environment; that is, human
preferences are programmable and are often internalized, just as are
aspects of our culinary and sexual preferences. The preferences become
part of the preference function that is maximized in preferences, beliefs, and
constraint models. It is in this manner that norms such as “treat strangers
equitably” become valued goals in themselves, and not simply because they lead
to the attainment of other valued goals.
The theory sketched above has two immediate empirical
entailments. First, people should rely on cultural learning to acquire
significant components of their social behavior. If they do not, the theory
cannot even get off the ground. Second, as a consequence of these adaptive
learning processes, societies with different historical trajectories are likely
to arrive at different social equilibria. As such, people from different
societies will tend to express different preferences and beliefs: one should be
able to measure between-group variation. With regard to this second entailment,
we submit the above results from our cross-cultural project.
For the first entailment, there is ample evidence from
psychology and sociology that humans acquire much of their social behavior
through cultural learning. Psychologists have amassed evidence showing that
children spontaneously (without incentives) acquire social behavior by
observing and imitating others (Bandura 1977; Rosenthal & Zimmerman 1978).
More to the point, studies of prosociality in children show that children
readily imitate models demonstrating either costly altruism or selfishness
(Bryan 1971; Bryan & Walbek 1970; Grusec 1971; Presbie & Coiteux 1971).
Additional work demonstrates that (1) this effect is not ephemeral and can be
seen in retests months later (Rice & Grusec 1975; Rushton 1975), (2) the
effect is increased somewhat if values are strongly voiced along with actions
(Grusec et al. 1978; Rice & Grusec 1975; Rushton 1975), (3) sometimes these
imitation patterns are generalized to other quite different contexts (Elliot
& Vasta 1970; Midlarsky & Bryan 1972), and (4) children use learned
standards of altruism to judge and punish others (Mischel & Liebert 1966).
Some of the details of how norms get internalized have been studied in
socialization theory (Grusec & Kuczynski 1997; Parsons 1967).
Integrated with these basic cultural processes, the
preferences and beliefs of new members are influenced by the economic and
social institutions that structure the tasks people perform to make a living
and to remain in good standing in their communities. Indeed, evidence from
experiments, industrial sociology, and ethnography suggest that commonly
performed tasks affect the basic values incorporated in the individual’s
preference function, and hence will be expressed far beyond the limits of the
workplace or the specific institutional structure responsible for their social
prominence. In experimental work, Sherif (1937) and others have shown that the
performance of cooperative tasks (in which success depends on the efforts of
many and the rewards are shared) induces positive sentiments toward those with
whom one cooperates. Competitive tasks produce the opposite effect.
Sociological and ethnographic studies show that the degree of autonomy one
exercises, for example in making a living, is strongly associated with child-rearing
values in industrial (Kohn 1990) and small-scale (Barry et al. 1959) societies.
That these values are widely internalized and expressed is exemplified by the
fact that group-level average UG offers and PGG contributions are highly correlated
across the societies in which both games were played (r = 0.79, p =
0.06, n = 6).
Consistent with this view is evidence from UG, DG, and PGG
experiments among children and adults in the United States showing that
preferences related to altruism, conditional cooperation, and equity are
acquired slowly over the first two decades of life (second graders are pretty
selfish) and subsequently change little after this (Harbaugh & Krause 2000;
Harbaugh et al. 2002; Henrich 2003).
Because of the nature of our adaptive learning processes,
individuals in experiments bring the preferences and beliefs that they have
acquired in the real world into the decision-making situation. The social
relations of daily life may lead individuals to generalize about how others are
likely to act in novel situations. Thus, if there is a high level of
cooperation in work or community, people may expect others to behave in a
similarly cooperative manner in novel situations such as those provided by
experimental games. If people prefer to cooperate when others cooperate (as
shown by experimental data from Fehr & Gächter 2000; 2002, and in
cross-cultural data from Henrich &
Smith 2004), and if they have reason to believe others will cooperate, they
themselves will likely cooperate, leading to a high level of cooperation in the
experimental situation. If subjects believe others will not cooperate, and even
if they prefer to cooperate as long as others do so as well, a low level of
cooperation will likely result. For example, participants in a market-oriented
society may develop distinct cognitive capacities and habits. Moreover,
extensive market interactions may accustom individuals to the idea that
strangers can be trusted (i.e., expected to cooperate). This idea is consistent
with the fact that UG offers and the degree of market integration are strongly
correlated across our groups.
Demonstrating the effect of contextual interpretation on
beliefs and expectations, experiments with students in industrialized societies
have shown that contextual cues can change contributions in social dilemmas.
This dramatizes the importance of expectations in strategic cooperative
behavior. For example, Ross and Ward (1996) and Pillutla and Chen (1999) used
two versions of a public goods game, one construed as a joint investment or
“Wall Street game,” and
the other as a contribution to a social event or “community
game.” Players contributed significantly less to the investment than to the
social event, holding their payoff structures constant (also see Hayashi et al.
1999).17
For some cues, culture and context interact. Cues that
create an effect in one place do not create the same effect elsewhere. For
example, in a public goods experiment comparing Canadian, mainland Chinese, and
The details of how daily life enters the experimental
situation to influence behavior remain unclear.
Two nonexclusive possibilities deserve note. It may be that different
social, cultural, and physical environments foster the development of differing
generalized behavioral dispositions (equity, altruism, etc.) that are
applicable across many domains, as might be the case using the above reasoning
concerning task performance or investment in reputation building. For example,
the Lamalera may be generally more “‘altruistic” or “fair-minded” than
Machiguenga or Quichua. In our experimental situations, such dispositions could
account for the statistical relationships between group characteristics and
experimental outcomes. Alternatively, the abstract game structures, which are
standard in such experiments, may cue one or more highly context-specific
behavioral rules (or sets of preferences), as is suggested by the situational
framing examples above. In these situations, subjects in some places were first
identifying the kind of situation they were in, seeking analogues in their
daily life, and then acting appropriately. In this case, individual differences
result from the differing ways that individuals frame a given situation, not
from generalized dispositional differences. Given what is known about how
generalized values develop, it is plausible that both are going on to differing
degrees in different societies.
One of our cases allows a distinction between the two.
Recall that the Orma made a connection between the public goods game and their
local practice, the harambee. The Orma believe that wealthier households
should make larger contributions to the harambee than poorer households.
The Orma did not perceive a similar connection between the harambee and
the UG. Multivariate regressions involving wealth, age, education, and income
indicate that wealth is the only significant predictor of PGG contributions among
Orma individuals. The more wealth a person has the more they contribute to the
common pool, just as in the real harambee. Wealth, however, is not a
significant predictor of UG offers in either multivariate or bivariate
analyses. The importance of wealth for PGG games, but not for UG, is consistent
with predictions from the context-specific approach, assuming that the
resemblance of the public goods game to the familiar harambee cues
appropriate behavior in that game but does not generalize to the uncued.
Combining a preferences, beliefs, and constraints approach
with culture-gene coevolutionary theory produces a framework that endogenizes
both the cultural and genetic aspects of human preferences and beliefs, and at
the same time retains analytically tractable models that permit quantitative
predictions of behavior (Camerer 2003; Fehr & Schmidt 1999; Fischbacher et
al. 2002). Coevolutionary approaches provide a firm theoretical foundation for
studying the psychological mechanisms that permit us to rapidly and accurately
acquire the locally adaptive preferences, norms, and beliefs (Gintis 2003a;
2003b; Henrich & Gil-White 2001; Richerson & Boyd 2000). Cultural
evolutionary game theory allows us to explore the conditions and processes that
generate the range of different preferences and beliefs that underpin the
diversity of human institutions and social norms observed in our species (Boyd
et al. 2003; Henrich & Boyd 2001; McElreath et al. 2003). Each of these
evolutionary processes helps us to understand where the preferences and beliefs
– the critical ingredients of the decision-making models – come from, and how
they have evolved over human history, on both shorter and longer time scales
(Bowles 1998).18
NOTES
1. We extend this axiom to cover cases in which individuals
maximize the expected utility of their material gains to address the
question of risk aversion, but use this simpler formulation otherwise.
2. Most of this group-level variation is not likely to be explained
by differences in sample size between our efforts and those of laboratory
experimentalists. First, our experiments used mostly sample sizes on a par
with, or larger than, university-based experiments. The robust UG pattern that
motivated us is based on numerous samples of 25 to 30 pairs. For example, Roth
et al.’s (1991) four-country study used samples of 27, 29, 30, and 30 pairs.
Comparably, the Machiguenga, Hadza, Mapuche, and Tsimane studies used 21, 55,
34, and 70 pairs. Overall, our mean sample size was 38, compared to 29 for Roth
et al. Second, the regressions
on UG offer shown below explain a substantial portion of the between-group
variation (which is unlikely to arise via sample variation). Third, we compared
this standard regression to a weighted regression (using 
as the weight) and found little difference in
the results, thereby demonstrating that the sample size variation is likely not
having important effects. Fourth, we regressed sample size on the groups’
deviations from the overall mean (across groups) and found no significant
relationship (p = 0.41).
3. The two-dimensional intervals were calculated using the
following procedure: For a sample of n data points, we created a
randomized “bootstrap” sample by sampling n times from the offer
distribution with replacement. For each randomly sampled offer, we
randomly sampled a rejection (e.g., if we sampled an offer of 40%, and two out
of three 40% offers were rejected, we sampled whether an acceptance or
rejection occurred with probability 2/3). This yielded a single “pseudosample”
of n offers and an associated rejection profile of 0’s or 1’s for each
offer. We then used the rejection profile to estimate an IMO (explained in the Appendix
of Henrich et al. 2004). This single resampling produced a mean offer and
IMO. This procedure was repeated 1,000
times. Each repetition generated a mean offer/IMO pair. The two-dimensional intervals
drew an ellipse around the 900 pseudosamples (out of the 1000 samples) that
were closest to the mean – that is, the smallest circle that included all 900
pseudosampled (mean offer, IMO) pairs. Small samples generate large confidence
intervals because the means of pseudosample of n draws, made with
replacement, can be quite different from the mean of the actual sample.
4. A simple measure of our confidence that the average offer is
above the estimated IMO is the percentage of resampled points that lie below
the 45-degree unity line (this is an exact numerical measure of “how much” of
the ellipse crosses right and below the 45-degree line). These percentages are
13.7% (Pittsburgh), 0.0% (Achuar), 0.0% (Shona), 58.9% (Sangu farmers), 0.0%
(Sangu herders), 1.5% (Mapuche), 1.2% (Machiguenga), 25.5% (Hadza), and 0.0%
(Orma). (These figures do not match up perfectly with the visual impression
from Figures 4a and 4b because the ellipses enclose the tightest cluster
of 900 points, so the portion of an ellipsis that overlaps the line may
actually contain no simulated observations, or may contain a higher density of
simulated observations across the 45-degree line). Note that the only group for
which this percentage is above half is the Sangu farmers. Even the Pittsburgh
(student) offers, which are widely interpreted as consistent with expected
income maximization (i.e., average offers are around the IMO; see Roth et al.
1991), are shown to be too high to be
consistent with expected income maximization.
The ellipses are flat and elongated because we are much less
confident about the true IMOs in each group than we are about the mean offers.
This is a reflection of the fact that small statistical changes in the
rejections lead to large differences in our estimates of the IMOs. Since
rejections may be the tail that wags the dog of proposer offers, our low
confidence in what the true IMOs are is a reminder that better methods are
needed for measuring what people are likely to reject. The second phase of our
project addressees this directly.
5. An individual for whom r
< 1 is risk averse, r = 1 is risk neutral, and r > 1 is risk
preferring. We calculated the values of r for which the observed mean offer
maximized the expected utility of the proposers, where the expectation is taken
over all possible offers and the estimated likelihood of their being rejected.
See the Appendix of Henrich et al. (2004) for details on this calculation.
6. Because the numbers of rejections were small, some of our
estimates of risk aversion are imprecise. Accordingly, one concern is that more
reasonable estimates of risk aversion might fit the data nearly as well as the
best fit. To test for this possibility, we computed the difference between the
best-fit value of r and 0.81, the value estimated by Tversky and
Kahneman (1992) from laboratory data on risky decision making. The differences
were small for some data sets and quite large for others. In addition, there is
a positive but nonsignificant correlation between the deviation of observed
behavior from the IMO and this measure of the precision of the r
estimate. Therefore, it seems unlikely that risk aversion is an important
explanation of our observations.
7. Among nonstudent adults in industrialized societies, DG offers
are higher, with means between 40 and 50%, and modes at 50% (Carpenter et al.
2005; Henrich & Henrich in press, chap. 7).
9. Since completing this project, our research team has decided to
avoid any use of deception in future work. We also hope to set this as the
standard for experimental work in anthropology.
10. Of course, some variations might matter a lot in some places
but not in others. This kind of culture-method interaction is in itself an
important kind of cultural variation.
11. It is important to distinguish between classes of games in
assessing the impact of methodological variables. Many of the largest effects
of methodological and contextual variables have been observed in Dictator Games
rather than in Ultimatum Games (e.g., Camerer 2003, chap. 2; Hoffman et al.
1998). This is not surprising since the DG is a “weak situation.” Absent a
strong social norm or strategic forces constraining how much to give,
methodological and contextual variables have a fighting chance to have a large
impact. In contrast, UG offers are strategically constrained by the possibility
of rejection; that is, a wide range of rejection frequency curves will lead to
a narrow range of optimal offers. As a result, we should expect less empirical
variation in UGs than in DGs. Therefore,
one cannot simply say “context matters a lot” without referring to specific
games.
12. Relative wealth was measured by the in-group percentile ranking
of each individual, with the measure of individual wealth varying among groups:
for the Orma and Mapuche we used the total cash value of livestock; among the
Au, Gnau, and Machiguenga we used total cash cropping land. In the UG,
estimates of relative wealth were available only for seven groups.
13. The original MacArthur-funded proposal is available at
http://www.hss.caltech.edu/roots-of-sociality/phase-i/.
14. Abigail Barr suggested this procedure.
15. Three exercises were performed to test robustness. First, because
the sample sizes vary across groups by a factor of almost 10, it is possible
the results are disproportionately influenced by groups with small samples. To
correct for this, we ran weighted least squares in which observations were
weighted by 1/√n. This gives univariate standardized coefficients of 0.61
(t = 3.80, p < 0.01) for PC and 0.41 (t = 2.28, p
< 0.05) for MI, close to those from ordinary least squares in Table 5. Second, we reran the (univariate) regressions,
switching every pair of adjacent expressed ranks in the variables PC and MI,
one pair at a time. For example, the societies ranked 1 and 2 were artificially
reranked 2 and 1, respectively, then the regression
was reestimated using the switched ranks. This comparison tells us how
misleading our conclusions would be if the ranks were really 2 and 1 but were
mistakenly switched. For PC, this procedure gave standardized univariate values
of βPC ranging from 0.53 to 0.66, with t-statistics from 3.0–4.5 (all p < 0.01). For MI, the
corresponding estimates range from 0.37–0.45, with t-statistics from 2.0–2.6
(all p < .05 one-tailed).
These results mean that even if small mistakes were made in ranking
groups on PC and MI, the same results are derived as if the mistakes had not been
made. The third robustness check added quadratic and cubic terms (e.g., MI2
and MI3). This is an omnibus check for a misspecification in which
the ordered ranks are mistakenly entered linearly, but identical numerical
differences in ranks actually have larger and smaller effects (e.g., the difference
between the impacts of rank 1 and rank 2 may be smaller than between 9 and 10,
which can be captured by a quadratic function of the rank). The quadratic and
cubic terms actually lower the adjusted R2
dramatically for MI, and increase it only slightly (from 0.60 to 0.63) for PC,
which indicates that squared and cubic terms add no predictive power.
16. This is true even for situations of n-person
cooperation, if punishing strategies also exist (Boyd & Richerson 1992;
Henrich & Boyd 2001).
17. Hoffman et al. (1994) reported similar effects of “social
distance” and construal in the UG; for example, players offer less (and appear
to accept less) when bargaining is described as a seller naming a
take-it-or-leave-it price to a buyer rather than as a simple sharing of money.
18. It is a common misconception that decision-making models rooted
in the preferences, beliefs, and constraints approach are inconsistent with
notions of evolved modularity and domain-specificity. Such models, however, are
mute on this debate, and merely provide a tractable approach for describing how
situational (e.g., payoff) information is integrated with coevolved
motivations. This implies nothing about the cognitive architecture that infers,
formulates, and/or biases beliefs and preferences, nor about what kinds of
situations activate which human motivations. It is our view that the science of
human behavior needs both proximate models that integrate and weight
motivations and beliefs, and rich cognitive theories about how information is
prioritized and processed.
References
Alvard, M. (2004) The ultimatum
game, fairness, and cooperation among big game hunters. In: Foundations of
human sociality: Economic experiments and ethnographic evidence from fifteen
small-scale societies, ed. J.
Henrich, R. Boyd, S. Bowles, C. Camerer, E. Fehr & H. Gintis, pp. 413–35.
Andreoni, J. (1988) Why free ride? Strategies
and learning in public goods experiments. Journal of Public Economics
37:291–304. [aJH]
Andreoni, J. & Miller, J. H. (2002)
Giving according to GARP: An experimental test of the consistency of
preferences for altruism. Econometrica 70(2):737–53. [aJH]
Baldwin, J. M. (1896) Physical and social
heredity. American Naturalist 30:422–28. [aJH]
Bandura, A. (1977) Social learning
theory. Prentice Hall. [aJH]
Barr, A. (2004). Kinship, familiarity, and trust: An
experimental investigation. In: Foundations of human sociality: Economic experiments
and ethnographic evidence from fifteen small-scale societies, ed. J.
Henrich, R. Boyd, S. Bowles, C. Camerer, E. Fehr, & H. Gintis, pp. 305–34.
Barry, H., Child, I. L. & Bacon, M. K.
(1959) Relation of child training to subsistence economy. American
Anthropologist 61:51–63. [aJH]
Blurton Jones, N. (1984) Selfish origin of human food
sharing: Tolerated theft. Ethology & Sociobiology 5:1–3. [aJH]
(1987) Tolerated theft:
Suggestions about the ecology and evolution of sharing, hoarding, and
scrounging. Social Science Information 26:31–54. [aJH]
Bowles, S. (1998) Endogenous preferences: The cultural
consequences of markets and other economic institutions. Journal Economic
Literature XXXVI:75–111. [aJH]
Bowles, S. & Gintis, H. (2004) The
origins of human cooperation. In: Genetic and cultural origins of
cooperation, ed. P.
Hammerstein. MIT Press. [aJH]
Boyd, R., Gintis, H., Bowles, S. & Richerson, P. J.
(2003) The evolution of altruistic punishment. Proceedings
of the National Academy of Sciences
100(6):3531–35. [aJH]
Boyd, R. & Richerson, P. (1992) Punishment allows the
evolution of cooperation (or anything else) in sizable groups. Ethnology and
Sociobiology 13:171–95. [aJH]
(1996) Why culture is common, but
cultural evolution is rare. Proceedings of the British Academy 88:77–93. [aJH]
Boyd, R. & Richerson, P. J. (1985) Culture
and the evolutionary process.
Bryan, J. H. (1971) Model affect and children's imitative
altruism. Child Development 42(6):2061–65. [aJH]
Bryan, J. H. & Walbek, N. H. (1970) Preaching and practicing
generosity: Children's actions and reactions. Child Development 41(2):329–53. [aJH]
Camerer, C. (1996) Rules for experimenting
in psychology and economics, and why they differ. In: Essays in honor
of Reinhard Selten, ed. W. Guth & E. van Damme. Springer-Verlag. [aJH]
(2003). Behavior game theory:
Experiments in strategic interaction.
Camerer, C. & Hogarth, R. (1999) The
effects of financial incentives in experiments: A review of
capital-labor-production framework. Cal Tech Social
Science Working Paper, 1059. [aJH]
Camerer, C. & Weber, M. W. (1992) Recent
developments in modelling preferences: Uncertainty and ambiguity. Journal of
Risk and Uncertainty 5:325–70. [aJH]
Cameron, L. (1999) Raising the stakes in
the ultimatum game: Experimental evidence from
Campbell, D. T. (1965) Variation and
selective retention in socio-cultural evolution. In: Social change in
developing areas: A reinterpretation of evolutionary theory, ed. H. R.
Barringer, G. I. Glanksten & R. W. Mack, pp. 19–49. Schenkman. [aJH]
Caporael, L., Dawes, R., Orbell, J. & van de Kragt, A.
(1989) Selfishness examined: Cooperation in the absence of egoistic incentives.
Behavioral & Brain Sciences 12:683–97. [aJH]
Carpenter, J., Burks, S., & Verhoogen, E. (2005). Comparing Students to Workers: The Effects of Social Framing on
Behavior in Distribution Games. In J. Carpenter, G. Harrison & J. A.
List (Eds.), Field Experiments in
Economics.
Cavalli-Sforza,
L. L. & Feldman, M. (1981) cultural transmission and evolution.
Charness, G., & Rabin, M. (2002).
Social Preferences: Some Simple Tests of a New Model. Quarterly Journal of Economics, 117(3), 817-869 [aJH]
de Quervain, D. J., Fischbacher,
U., Treyer, V., Schellhammer, M., Schnyder, U., Buck, A. & Fehr, E. (2004)
The neural basis of altruistic punishment. Science 305:1254–58. [aJH]
Durham, W. H. (1991) Coevolution: Genes, culture, and
human diversity.
Elliot, R. & Vasta, R. (1970) The
modeling of sharing: Effects associated with vicarious reinforcement,
symbolization, age, and generalization. Journal of Experimental Child
Psychology 10:8–15. [aJH]
Ensminger, J. (2004) Market integration and fairness:
Evidence from ultimatum, dictator and public goods experiments in East Africa.
In: Foundations of human sociality: Economic experiments and ethnographic
evidence from fifteen small-scale societies, ed. J. Henrich, R. Boyd, S. Bowles, C. Camerer, E. Fehr & H.
Gintis, pp. 356–81.
Fehr, E., Fischbacher, U. & Gaechter,
S. (2002) Strong reciprocity, human cooperation and the enforcement of social norms.
Human Nature 13:1–25. [aJH]
Fehr, E. & Gächter, S. (2000) Cooperation and punishment
in public goods experiments. American Economic Review 90(4):980–95. [aJH]
(2002)
Altruistic punishment in humans. Nature 415:137–40. [aJH]
Fehr, E., & Schmidt, K. (1999).
A theory of fairness, competition, and cooperation. Quarterly Journal of Economics, 114,
817-868.
Ferraro, Paul J. and Ronald G. Cummings. 2005.
Cultural Diversity, Discrimination and Economic Outcomes: an experimental
analysis. Experimental Working Paper 2003-002 (revised 1/05), Georgia
State University, Atlanta, GA
Fischbacher, U., Fong, C. M. & Fehr,
E. (2002) Fairness and the power of competition. Unpublished manuscript,
Fudenberg, D. & Levine, D. (1998) The
theory of learning in games. MIT Press. [aJH]
Gil-White, F. (2004) Ultimatum game with an ethnicity
manipulation: Results from Khovdiin Bulgan Cum, Mongolia. In: Foundations of
human sociality: Economic experiments and ethnographic evidence from fifteen small-scale
societies, ed. J. Henrich, R. Boyd, S. Bowles,
C. Camerer, E. Fehr & H. Gintis, pp. 260–304.
Gintis, H. (2000) Game theory evolving.
(2003a) The
hitchhiker's guide to altruism: Gene-culture coevolution, and the
internalization of norms. Journal of Theoretical Biology 220(4):407–18. [aJH]
(2003b) Solving
the puzzle of prosociality. Rationality and Society 15(2):155–87. [aJH]
Grusec, J. E. (1971) Power and the internalization
of self-denial. Child Development 42(1):93–105. [aJH]
Grusec, J. E. & Kuczynski, L. (1997) Parneting and children's
internalization of values: A handbook of contemporary theory. Wiley. [aJH]
Grusec, J. E., Saas-Kortsaak, P. & Simutis, Z. M. (1978)
The role of example and moral exhortation in the
training of altruism. Child Development 49(3):920–23. [aJH]
Gurven, M. (2004) Does market exposure
affect economic game behavior? The ultimatum game and the public goods game
among the Tsimane of Bolivia. In: Foundations of human sociality: Economic experiments
and ethnographic evidence from fifteen small-scale societies, ed. J. Henrich, R. Boyd, S. Bowles,
C. Camerer, E. Fehr & H. Gintis, pp. 194–231.
Harbaugh, W. T. & Krause, K. (2000)
Children's altruism in public goods and dictator experiments. Economic
Inquiry 38(1):95–109. [aJH]
Harbaugh, W. T., Krause, K. & Liday, S. G. (2002) Bargaining
by children. Unpublished manuscript.
Hayashi, N., Ostrom, E., Walker, J. &
Yamagishi, T. (1999) Reciprocity, trust, and the sense of control. Rationality
and Society 11(1):27–46. [aJH]
Henrich, J. (2000) Does culture
matter in economic behavior: Ultimatum game bargaining among the Machiguenga. American
Economic Review 90(4):973–80. [aJH]
(2003) A cultural
species.Unpublished manuscript.
Henrich, J. & Boyd, R. (2001) Why
people punish defectors: Weak conformist transmission can stabilize costly
enforcement of norms in cooperative dilemmas. Journal of Theoretical Biology
208:79–89. [aJH]
Henrich, J., Boyd, R., Bowles, S.,
Camerer, C., Fehr, E., & Gintis, H., eds. (2004) Foundations of human sociality:
Economic experiments and ethnographic evidence from fifteen small-scale societies.
Henrich, J. & Gil-White, F. (2001) The
evolution of prestige: Freely conferred deference as a mechanism for enhancing
the benefits of cultural transmission. Evolution and Human Behavior 22(3):165–96. [aJH]
Henrich, J. & McElreath, R. (2002) Are peasants
risk-averse decision makers? Current Anthropology 43(1):172–81. [aJH]
Henrich, J., & Smith, N. (2004).
Comparative experimental evidence from Machiguenga, Mapuche,
and American Populations. In J. Henrich, R. Boyd, S. Bowles, H. Gintis,
E. Fehr & C. Camerer (Eds.), Foundations
of Human Sociality: Economic experiments and ethnographic evidence from fifteen
small-scale societies (pp. 125-167). Oxford: Oxford University Press. [aJH]
Henrich, J., & Henrich, N. S. (in
press). Culture and the
Evolutionary Origins of Cooperation: Kinship, Reciprocity, Social Norms, and
Ethnicity among the Chaldeans of metro Detroit. Oxford: Oxford University
Press. [aJH]
Hill, E. & Sally, D. (2004) Dilemmas and bargains:
Autism, theory of mind, cooperation and fairness. Unpublished
manuscript. University College
London [aJH]
Hill, K. & Gurven, M. (2004) Economic
experiments to examine fairness and cooperation among the Ache Indians of
Paraguay. In: Foundations in Human Sociality: Economic experiments
and ethnography from fifteen small-scale societies, ed. J. Henrich, R. Boyd, S. Bowles, C. Camerer, E. Fehr & H.
Gintis, pp. 382–412.
Hoffman, E., McCabe, K., Shachat, K. &
Smith, V. (1994) Preferences, property rights, and anonymity in bargaining
games. Game and Economic Behavior 7:346–80. [aJH]
Hoffman, E., McCabe, K. & Smith, V. (1998) Behavioral foundations
of reciprocity: Experimental economics and evolutionary psychology. Economic
Inquiry 36:335–52. [aJH]
Hoffman, E., McCabe, K. A. & Smith, V. L. (1996) On expectations and the monetary stakes in ultimatum games. International
Journal of Game Theory 25(3):289–301. [aJH]
Johnson, A. (2003) Families of the forest: Matsigenka
Indians of the Peruvian Amazon.
Johnson, A. & Earle, T. (2000) The
evolution of human societies (2nd edition).
Kachelmeier, S. & Shehata, M. (1997) Internal
auditing and voluntary cooperation in firms: A cross-cultural experiment. The
Accounting Review 72(3):407–31. [aJH]
Kagel, J. H. & Roth, A. E., eds. (1997) The handbook of experimental economics.
Kaplan, H. & Hill, K. (1985) Food sharing among Ache foragers:
Tests of explanatory hypotheses. Current Anthropology 26:223–45. [aJH]
Knez, M. J. & Camerer, C. F. (1995) Outside
options and social comparison in three-player ultimatum game experiments. Games
and Economic Behavior 10(1):65–94. [aJH]
Kohn, M. (1990) Position in the class
structure and psychological functioning in the
Larrick, R. P. & Blount, S. (1997) The
claiming effect: Why players are more generous in social dilemmas than in
ultimatum games. Journal of Personality and Social Psychology 72(4):810–25. [aJH]
Ledyard, J. O. (1995) Public goods: A survey of experimental
research. In: The handbook of experimental economics, pp. 111–94.
List, J. A. & Cherry, T. L. (2000) Learning
to accept in ultimatum games: Evidence form an experimental design that
generates low offers. Experimental Economics 3:11–31. [aJH]
Loewenstein, G. (1999) Experimental economics form the
vantage point of view of behavioral economics. Economic Journal 109:F25–F34. [aJH]
Marlowe, F. W. (2004) Dictators and ultimatums in an
egalitarian society of hunter-gatherers: The Hadza of Tanzania. In: Foundations
of human sociality: Economic experiments and ethnographic evidence from fifteen
small-scale societies, ed. J.
Henrich, R. Boyd, S. Bowles, C. Camerer, E. Fehr & H. Gintis, pp. 168–93.
Marlowe, F. W. (2004) What explains
Hadza food sharing? In: Socioeconomic aspects of human behavioral ecology, ed. M. Alvard, pp. 67–86, Research in Economic Anthropology, vol. 23. [aJH]
McElreath, R., Boyd, R. & Richerson, P. (2003) Shared
norms can lead to the evolution of ethnic markers. Current Anthropology 44(1):123–29. [aJH]
Midlarsky, E. & Bryan, J. H. (1972)
Affect expressions and children's imitative altruism. Journal of
Experimental Child Psychology 6:195–203. [aJH]
Mischel, W. & Liebert, R. M. (1966) Effects of
discrepancies between observed and imposed reward criteria on their acquisition
and transmission. Journal of Personality and Social Psychology 3:45–53. [aJH]
Murnighan, J. K. & Saxon, M. S. (1998)
Ultimatum bargaining by children and adults. Journal of Economic
Psychology 19(4):415–45. [aJH]
Parsons, T. (1967) Sociological theory
and modern society. Free Press.
[aJH]
Patton, J. Q. (2004) Coalitional effects on reciprocal
fairness in the ultimatum game: A case from the Ecuadorian Amazon. In: Foundations
of human sociality: Economic experiments and ethnographic evidence from fifteen
small-scale societies, ed. J.
Henrich, R. Boyd, S. Bowles, C. Camerer, E. Fehr & H. Gintis, pp. 96–124.
Pillutla, M. M. & Chen, X.-P.
(1999) Social norms and cooperation in social dilemmas: The effects of context
and feedback. Organizational Behavior and Human Decision Processes 78(2):81–103. [aJH]
Presbie, R. J. & Coiteux, P. F. (1971)
Learning to be generous or stingy: Imitation of sharing behavior as a function
of model generosity and vicarious reinforcement. Child Development 42(4):1033–38. [aJH]
Pulliam, H. R. & Dunford, C. (1980) Programmed to learn:
An essay on the evolution of culture.
Rice, M. E. & Grusec, J. E. (1975) Saying and doing:
Effects of observer performance. Journal of Personality and Social
Psychology 32:584–93. [aJH]
Richerson, P. & Boyd, R. (2000) Complex societies: The
evolutionary dynamics of a crude superorganism. Human Nature 10:253–89. [aJH]
Rosenthal, T. L. & Zimmerman, B. J.
(1978) Social learning and cognition.
Academic Press.
[aJH]
Ross, L. & Ward, A. (1996) Naive realism: Implications
for social conflict and misunderstanding. In: Values and knowledge, ed. T. Brown, E. Reed &
E. Turiel. Erlbaum.
[aJH]
Roth, A. E. (1995) Bargaining experiments. In: The
handbook of experimental economics, ed. J. H. Kagel & A. E. Roth, pp.
253–48.
Roth, A. E., Prasnikar, V., Okuno-Fujiwara, M. & Zamir,
S. (1991) Bargaining and market behavior in Jerusalem, Ljubljana, Pittburgh and
Tokyo: An experimental study. American Economic Review 81(5):1068–95. [aJH]
Rushton, J. P. (1975) Generosity in children: Immediate and
long term effects of modeling, preaching, and moral judgement. Journal of
Personality and Social Psychology 31:459–66. [aJH]
Sally, D. (1995) Conversation and cooperation in social
dilemmas: A meta-analysis of experiments from 1958 to 1992. Rationality and
Society 7:58–92. [aJH]
Sanfey, A. G., Rilling, J. K., Aronson, J. A., Nystrom, L.
E. & Cohen, J. D. (2003) The neural basis of economic decision-making in
the ultimatum game. Science 300:1755–58. [aJH]
Sherif, M.,
Slonim, R. & Roth, A. E. (1998) Learning
in high stakes ultimatum games: An experiment in the
Tomasello, M. (1999) The
cultural origins of human cognition.
Tomasello, M., Carpenter, M., Call, J., Behne, T. &
Moll, H. (in press) Understanding and sharing
intentions: The origins of cultural cognition. Behavioral
and Brain Sciences. [aJH]
Tracer, D. (2003) Selfishness and fairness in economic and
evolutionary perspective: An experimental economic study in Papua New Guinea. Current
Anthropology 44(3):432–38. [aJH]
Tracer, D. (2004). Market integration,
reciprocity, and fairness in rural Papua New Guinea: Results from a two-village
ultimatum game experiments. In: Foundations of human sociality: Economic experiments
and ethnographic evidence from fifteen small-scale societies, ed. J. Henrich, R. Boyd, S. Bowles,
C. Camerer, E. Fehr & H. Gintis, pp. 232–59.
Tversky, A. & Kahneman, D. (1992) Advances in prospect theory:
Cumulative representation of uncertainty. Journal of Risk and Uncertainty 5:297–323. [aJH]
Weibull, J. W. (1995) Evolutionary game
theory. MIT Press. [aJH]
Woodburn, J. (1968) An introduction
to Hadza ecology. In: Man the hunter, ed. R. B. Lee & I. DeVore, pp.
49–55. Aldine.
[aJH]
Young, H. P. (1998) Individual strategy and social
structure: An evolutionary theory of institutions.