Beruflich Dokumente
Kultur Dokumente
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp
.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
American Economic Association is collaborating with JSTOR to digitize, preserve and extend access to The
American Economic Review.
http://www.jstor.org
When we leave our closet, and engage in visible "good") of the ... small band or
the common affairsof life, (reason's) con- troop, or ... our families ... to the (ex-
clusions seem to vanish, like the phan- tended orderof cooperationthroughmar-
toms of the night on the appearanceof the kets), as our instincts and sentimental
morning;and 'tis difficult for us to retain yearnings often make us wish to do, we
even that conviction, which we had at- would destroy it. Yet if we were to always
tained with difficulty ... (David Hume, apply the (competitive) rules of the ex-
1739, 1985, p. 507). tended order to our more intimate group-
ings, we would crush them (FriedrichA.
... we must constantlyadjustour lives, our Hayek, 1988, p. 18; italics are his, paren-
thoughts and our emotions, in order to thetical reductionsare mine).
live simultaneouslywithin differentkinds
of orders according to different rules. If We have become accustomed to the idea
we were to apply the unmodified, un- that a naturalsystem like the humanbody
curbedrules (of caring interventionto do or an ecosystem regulates itself. To ex-
plain the regulation,we look for feedback
loops rather than a central planning and
t This article is a revised version of the lecture Vernon directing body. But somehow our intui-
L. Smith delivered in Stockholm, Sweden, on December 8, tions about self-regulation do not carry
2002, when he received the Bank of Sweden Prize in Eco- over to the artificial systems of human
nomic Sciences in Memory of Alfred Nobel. The article is
copyright (? The Nobel Foundation2002 and is published society. (Thus) ... the ... disbelief always
here with the permission of the Nobel Foundation. expressed by (my) architecturestudents
The title was suggested to the authorin the paperby Joel (about) ... medieval cities as marvelously
Norman, 'Two Visual Systems and Two Theories of Per- patterned systems that had mostly just
ception: An Attempt to Reconcile the Constructivistand "grown"in response to myriads of indi-
Ecological Approaches" (Behavioral and Brain Sciences, vidual decisions. To my studentsa pattern
2002). After finishing this paper I found that my use of the implied a planner.... The idea that a city
term below had been used by Gerd Gigerenzeret al. (1999), could acquireits patternas "naturally"as
for application to "fast and frugal decision making," by
individuals:"A heuristic is ecologically rational to the de-
a snowflake was foreign to them (Herbert
gree thatit is adaptedto the structureof an environment"(p.
Alexander Simon, 1981, 1996, p. 33).
13). Hayek, in the citations below, identifies both kinds of
rationality. Historically, a recurrenttheme in economics
* Departmentof Economics and Law,
Interdisciplinary
is that the values to which people respond are
Center for Economic Science, George Mason University, not confinedto those one would expect based on
4400 UniversityDrive, MSN 1B2, Fairfax,VA 22030. Doing the narrowly defined canons of rationality.
experimentaleconomics has changed the way I think about These roots go back to Adam Smith (1759,
economics. There are many reasons for this, but one of the
most prominent is that designing and conducting experi- 1776), who examinedthe moral sympathiesthat
ments forces you to think through the process rules and characterizenaturalhuman sociality.1 Contrary
proceduresof an institution.Few, like Einstein,can perform to vulgarimpressions,in Smith'sview, each indi-
detailed and imaginative mental experiments. Most of us vidualdefinedandpursuedhis own interestin his
need the challenge of real experiments to discipline our
own way, and individualswere mischaracterized
thinking. In this paper I hope to indicate how my thinking
has been changedin some detail. I am indebtedto Sid Siegel
for technical and conceptual inspiration;to John Hughes,
Stan Reiter, and the PurdueUniversityfaculty from 1955 to 1Economistsare
largely untouchedby Smith's firstgreat
1967 for warm, tolerant support beginning with my first work, which was eclipsed by the Wealthof Nations. Thus,
experiment;to CharlesPlott, CharlesHolt, and MartinShu- one of the profession's best-known historian of economic
bik for many valuable encountersover the years on institu- thought,"foundthese two works in some measurebasically
tional and experimental issues; to students, visitors, the inconsistent" (Jacob Viner, 1991, p. 250). For a contrary
currentICES team, and especially to my growing but tol- interpretationsee Smith (1998). Many of the references
eratingfamily who have made all these years the best years herein to my own and coauthoredwork have been reprinted
of my life. in Smith (1991, 2000).
465
from Descartes (also Bacon and Hobbes),6who Alvin E. Roth, 1998; Camerer and Teck-Hua
believed and argued that all worthwhile social Ho, 1999).
institutionswere and should be createdby con- An alternative and perhaps complimentary
scious deductiveprocesses of humanreason.7In explanation of some of these contradictionsto
economics the SSSM leads to rational predic- theory is that people may use social-grown
tive models of decision that motivate research norms of trust and reciprocity9(including eq-
hypotheses thatexperimentalistshave been test- uity, meaning to each according to his justly
ing in the laboratory since the mid-twentieth earneddessert;i.e., equality of opportunity,not
century. Although the test results tend to be outcome) to achieve cooperativestates superior
confirmingin impersonalmarketexchange, the to individuallyrationaldefection outcomes. We
results are famously and recalcitrantlymixed in will reportsome experimentaltests designed to
"personalexchange," notably in a great variety separate competing preference and reciprocity
of two-person extensive-form games where theories of behavior in personal exchange. Al-
some half of the people attemptand frequently though reciprocity seems to be a leader in the
succeed when risking cooperation, even when comparisons we summarize,its strengthis not
anonymouslypaired. These results have moti- uniformacross all tests, and much remainsto be
vated constructivistextensions of game theory learnedaboutthe hidden recesses of meaningin
based on other-regarding,in addition to own- humanbehaviorand the circumstancesin which
regarding, preferences (e.g., Gary E Bolton, cooperativeor noncooperativebehavioris man-
1991; Matthew Rabin, 1993), and on ifest.10Technically, the issue is how most pro-
"learning"-the idea that the predictionsof the ductively to model agent "types"by extending
SSSM might be approachedover time by trial- game theory so that types are an integralpartof
and-erroradaptationprocesses (Ido Erev and its predictive content, rather than merely im-
ported as an ex post technical explanation of
experimentalresults. For example, moves can
6 In the nineteenth
signal types, and effect decision, which explains
century, Bentham and John Stuart why game form matters,and why payoffs avail-
Mill were among the leading constructivists.Bentham (and
the utilitarians)sought to "... remake the whole of ... (Brit- able, but forgone, can effect outcomes. These
ish) ... law and institutions on rational principles"(Hayek, elements must be partof the internalstructureof
1960, p. 174). Mill introducedthe much-abusedconstruc- the theory such that outcomes become predic-
tivist concept of (but not the name) "naturalmonopoly."To tions conditionalon the elementarycharacteris-
Mill it was transparentlywasteful and duplicative to have
two or more mail carriersoperatingon the same route. He
tics of players who read each other's intentions.
is the intellectual father of the U.S. and other postal mo- If successful, many of the basic results in game
nopolies around the world, their resistance to innovation,
and their demise in the face of the privatizationmovement
in some countries and the growth of superiorsubstitutesin 9 Dissatisfied with the utilitarianapproach because its
others. Mill could not imagine that it would be efficient for predictions fail to account for the observed importanceof
two cities to be connected by two parallel railroadtracks instructions/procedures,we began investigating the reci-
(Mill, 1899, Vol. 1, pp. 131, 141-42; Vol. 2, p. 463). Mill procity hypothesis in Elizabeth Hoffman et al. (1994). Me-
died in 1873. I would conjecturethat by that date, or soon chanically, utilities can serve as intermediateplaceholders
thereafter,men with grade-school educations had become for reciprocaltrust, but, as surface indicators,serve poorly
rich constructing the first parallel-route railroads. These to generatenew hypotheses designed to understandinterac-
emergent contradictionsto constructivistnaturalmonopoly tive processes. Good theory must be an engine for generat-
are examples of what we shall call ecological rationality,as ing testable hypotheses, and utility theory runs out of fuel
detailed below. quickly. Utility values are seen as providing the ultimate
7 ... Descartes contended that
all the useful human in- "given" data, and the conversationstops.
stitutions were and ought to be deliberate creation(s) of 10I am reminded of a
departmenthead from Hewlett-
conscious reason ... a capacity of the mind to arrive at the Packardvisiting our lab. I naively assumedthathe would be
truth by a deductive process from a few obvious and un- most interested in demonstrationsof some of our market
doubtablepremises" (Hayek, 1967, p. 85). experiments.Not so. He was more interestedin the "trust"
8 Behavioraleconomists have made a cottage industryof
experiments.Why? He saw the HP managementproblemas
showing that the SSSM assumptionsseem to apply almost one of getting teams to cooperateinternallyby buildingtrust
nowhere to real decisions. This is because their research and trustworthiness,while being vigorous competitors ex-
programhas been a deliberatesearch in the tails of distri- ternally.Could the trustgames serve as a measurementand
butions for "Identifyingthe ways in which behaviordiffers teaching tool for helping to solve this problem?This nicely
from the standard model..." (Mullainathan and Thaler, illustrates the tension in Hayek's two-worlds quote in the
2001, Vol. II, p. 1094), a search that can only succeed. text.
theory would become special cases of the ex- about what we observe. Having sharpenedour
tended theory. understandingon Cartesiancomplete informa-
In market experiments-where cooperation tion parableswe carrythese tools into the world
can occur throughthe coordinationfunction of for applicationwithout all the necessarycaveats
prices produced by, but simultaneouslyresult- that reflect the tractabilityconstraintsimposed
ing from, interaction with individual choice by our bounded professional cognitive capaci-
behavior-the results are more commonly in ties as theorists.
accord with standardcompetitive models that In summary, constructivism uses reason to
maximize group welfare. This professionalvic- deliberately create rules of action, and create
tory is hollowed by the failure of standardthe- human socioeconomic institutions that yield
ory to predict the "surprisingly"11weak outcomes deemed preferable, given particular
conditions under which the results obtain.12 circumstances,to those producedby alternative
Thus, for tractability,Cartesian rationalism arrangements.Although constructivism is one
provisionally assumes or "requires"agents to of the crowning achievements of the human
possess completepayoff and otherinformation- intellect, it is importantto remain sensitive to
far more than could ever be given to one mind. the fact that human institutionsand most deci-
In economics the resulting exercises are be- sion makingis not guided primarily,if at all, by
lieved to sharpeneconomic thinking,as if-then constructivism.Emergentarrangements,even if
parables.Yet, these assumptionsare unlikely to initially constructivistin form, must have sur-
approximatethe level of ignorancethathas con- vival propertiesthattake accountof opportunity
ditioned either individual behavior, or our costs and environmentalchallenges invisible to
evolved institutions,as abstractnorms or rules our modeling efforts.
independent of particular parameters, which
have survived as part of the world of experi- B. Limitationsand Distractions of
ence.13The temptationis to ignore this reality ConstructivistRationality
because it is poorly understood, and does not
yield to our familiar but inadequatemodeling Since our theories and thought processes
tools, and to proceed in the implicit belief that about social systems involve the conscious and
our parables capture what is most essential deliberateuse of reason, it is necessary to con-
stantly remind ourselves that human activity is
diffused and dominated by unconscious, auto-
1 Robert B. Wilson (1992,
p. 256) discusses an effi- nomic, neuropsychologicalsystems that enable
ciency theorem,and suggests that the phenomenonis "per-
people to function effectively without always
haps unsurprising."It is, nowadays,but few believe it; also,
theory has lagged well behind the evidence, and yields calling upon the brain's scarcest resource-
inadequatetestable insight into the process dynamics oper- attentional and reasoning circuitry. This is an
ating in different institutions. important economizing property of how the
12I want to acknowledge correspondencewith Charles brain works. If it were otherwise, no one could
Plott and add the following: Althoughthis is a giant victory
for the economic theory of marketsit simultaneouslydem- get through the day under the burden of the
onstrates that the theory is incomplete. The unexpectedly self-conscious monitoring and planning of ev-
weak conditions under which the results obtain are good ery trivial action in detail.14Also, no one can
news for marketperformance,but not such good news for express in thoughts,let alone words, all that he
the scientific communitybecause it demonstratesthatwe do or she knows, and does not know but might call
not understandwhy marketswork as they do. You do not
have to have large numbersof agents, each an insignificant upon, or need to discover, for some purposive
partof the whole-three or four buyers and as many sellers action. Imagine the strain on the brain's re-
are entirely adequatein a wide range of economic environ- sources if at the supermarketa shopper were
ments; they do not have to have complete or perfect or required to explicitly evaluate his preferences
common information-each can have only privateinforma-
for every combinationof the tens of thousands
tion; nor is it required that individuals make decisions
systematically or be economically sophisticated. of grocery items that are feasible for a given
13
Throughout the paper I will use "environment"to
mean the collection of agent values (preferences)thatdefine
the gains from trade; "institution"to refer to the language
14
"If we stopped doing everything for which we do not
(messages), rules of message exchange, and contract in a know the reason, or for which we cannot provide a justifi-
market;and "behavior"for agent message choices condi- cation ... we would probably soon be dead" (Hayek, 1988,
tional on the environmentand institution(Smith, 1982). p. 68).
budget. Such mental processes are enormously thoroughly understood this proposition, what
opportunity-costly and implicitly our brain was the "fatalconceit"?"The idea that the abil-
knows, if our conscious mind does not know, ity to acquire skills stems from reason." The
that we must avoid incurringopportunitycosts constructivistmind makes a fatal "error,"blind-
that are not worth the benefit." The challenge ing itself to understanding,as we are warned,
of any unfamiliaraction or problemappearsfirst "one should never suppose that our reason is in
to trigger a search by the brain to bring to the the higher critical position and that only those
conscious mind what one knows that is related moral rules are valid that reason endorses"
to the decision context. Context triggers auto- (Hayek, 1988, p. 21). But the anthropocentric
biographic experiential memory, which ex- (-morphic) mind routinely makes this signifi-
plains why context surfaces as a nontrivial cant error.
treatment, particularlyin small group experi- Most of our operatingknowledge17we do not
ments. The brain (including the entire neuro- remember learning. Natural language is the
physiological system) takes over directly in the most prominent example, but also music and
case of familiar, mastered tasks, and plays the virtually everything that constitutes our devel-
equivalent of lightening chess when the "ex- opmental socialization. We learn the rules of a
pert"trades,plays Beethoven's Fifth piano con- language and of efficient social intercourse,
certo, or connects with a 95-mile/hour withoutexplicit instruction,simply by exposure
fastball-all without self-aware "thinking"by to family and extended family social networks
the mind. (Jerome Kagan and Sharon Lamb, 1987; Alan
We fail utterlyto possess naturalmechanisms Page Fiske, 1991; Kagan, 1994; Steven Pinker,
for reminding ourselves of the brain's off-line 1994). That the brain is capable of off-line
activities and accomplishments.This important subconscious learningis shown by experiments
proposition has led Michael S. Ga7zzaniga with amnesiacswho are taughta new task. They
(1998) to ask why the brainfools the mind into learn to perform well, but memory of having
believing it is in control.16And to Hayek, who learned the task escapes them (Barbara J.
Knowlton et al., 1996).
principlesof action, norms,traditions,and "mo- sciously to create it to serve its perceived in-
rality."19Ecological rationality uses reason- tended purposes.20
rationalreconstruction-to examinethe behavior In experimental economics the eighteenth-
of individuals based on their experience and century Scottish traditionis revealed in the ob-
folk knowledge, who are "naive"in their ability servationof emergentorderin numerousstudies
to apply constructivist tools to the decisions of existing marketinstitutionssuch as the con-
they make; to understandthe emergentorderin tinuous double auction (CDA).21To paraphrase
human cultures;to discover the possible intel- Adam Smith, people in these experiments are
ligence embodied in the rules, norms, and insti- led to promote group welfare-enhancingsocial
tutions of our cultural and biological heritage ends that are not part of their intention. This
thatare createdfrom humaninteractionsbut not principle is supportedby hundreds of experi-
by deliberatehumandesign. People follow rules ments whose environments and institutions
without being able to articulatethem, but they (sealed bid, posted offer, and others besides
can be discovered. This is the intellectualheri- CDA) may exceed the capacityof formalgame-
tage of the Scottishphilosophers,who described theoretic analysis to articulatepredictive mod-
and interpretedthe social and economic order els. But they do not exceed the functional
they observed. capacity of collectives of incompletely in-
An eighteenth-centuryprecursorof Herbert formed human decision makers, whose auto-
Simon, David Hume was concerned with the nomic mental algorithms coordinate behavior
limits of reason, the bounds on human under- through the rules of the institution-social
standing,and with scaling back the exaggerated algorithms-to generate high levels of mea-
claims of Cartesianconstructivism.To Hume, sured performance.
rationalitywas phenomenathat reason discov- Acknowledging and investigating the work-
ers in emergentinstitutions.Thus, "the rules of ings of unseen processes are essential to the
morality ... are not conclusions of (our) reason."
(Hume, 1985, p. 235). Adam Smith developed
the idea of emergentorderfor economics. Truth 20In culturaland
biological evolution, orderarises from
is discovered in the form of the intelligence mechanisms for generating variation to which is applied
embodied in rules and traditions that have mechanismsfor selection. I am indebted to Todd Zywicki,
formed,inscrutably,out of the ancienthistoryof who, at a recent Liberty Fund conference on "Hayek, Ex-
human social interactions.This is the antithesis periment and Freedom," observed that reason is good at
of the anthropocentricbelief that if an observed providing variation, but not selection. Constructivism is
indeed an engine for generating variation, but is far too
social mechanism is functional, somebody in limited in its ability to comprehendand apply all the rele-
the unrecordedpast must have used reason con- vant facts to serve the process of selection, which is better
left to ecological processes.
21
Whatexperimentalistshave unintentionallybroughtto
the table is a methodology for objectively testing the
twins (Nancy L. Segal, 1999). Deconstructivistreports ar- Scottish-Hayekian hypotheses under scientific controls.
gue that these studies exhibit many of the usual data and This answers the question Milton Friedmanis said to have
statistical identification problems (Arthur S. Goldberger, raised concerningthe validity of Hayek's theory/reasoning:
1979), but the need is for positive revisionist analysis. "How would you know?" (I am unable to find/providethe
19"Morality"refers to any maxim of cohesive social reference). Remarkably,Hayek stated on the very edge of
behavior that survives the test of time, and is prominently recognizing what experimentscould do for testing his the-
represented by the great "shalt not" prohibitions of the ory, then dismissed it. Thus, "We can test it ('competitionas
leading world religions: thou shalt not: (1) steal, (2) covet a discovery procedure') on conceptual models, and we
the possessions of others, commit (3) murder,(4) adultery, might conceivably test it in artificially created real situa-
or (5) bear false witness. The first two define and defend tions, where the facts which competition is intended to
property rights in the product of one's labor, and all re- discover are already known to the observer. But in such
sources accumulatedby such labor,enabling the emergence cases it is of no practical value, so that to carry out the
of the extended order of mind through markets. The last experiment would hardly be worth the expense" (Hayek,
three commandments protect the sanctity of social 1984, p. 255; also see Smith, 2002, pp. 95-96). Economic
exchange-the external order of the mind. These modest historians,e.g., North (1981), and political economists, e.g.,
exclusionary constraintsleave an immense scope for free- Elinor Ostrom (1990), have long explored the intelligence
dom within their bounds. Corollaries, like the Buddhist and efficacy embodied in emergent socioeconomic institu-
live-and-let-live version of the golden rule, are explicit in tions that solve, or fail to solve, problems of growth and
this respect:"do not unto othersas you would have them not resource management.They study "natural"ecological ex-
do unto you." perimentsfrom which we have learned immeasurably.
growth of our understandingof social phenom- II. Impersonal Exchange: The Extended Order
ena, and enable us to probe beyond the anthro- of the Market
pocentric limitations of constructivism.
Both kinds of rationalityhave influenced the A. How Are the Two Concepts of a Rational
design and interpretation of experiments in Order Related?
economics. Thus, if people in certain contexts
make choices that contradictour formal theory Constructivism takes as given the social
of rationality, rather than conclude that they structures generated by emergent institutions
are irrational,some ask why, reexamine main- that we observe in the world, and proceeds to
tained hypotheses including all aspects of the model it formally. An example would be the
experiments-procedures, payoffs, context, in- Dutch auction or its alleged isomorphic equiv-
structions, etc.-and inquire as to what new alent, the sealed-bid auction (William Vickrey,
concepts and experimentaldesigns can help us 1961; Paul Milgrom and Robert J. Weber,
to better understandthe behavior. What is the 1982). Constructivistmodels need not ask why
subjects' perceptionof the problemthatthey are or how an auctioninstitutionarose or what were
trying to solve? the ecological conditions that createdit, or why
Finally,understanding decisionrequiresknowl- there are so many distinct auction institutions.
edge beyond the traditionalbounds of econom- In some cases it is the other way around.Thus,
ics,22 a challenge to which Hume and Smith revenue equivalence theorems show that the
were not strangers.23This is manifest in the standard auctions generate identical expected
recent studies of the neural correlatesof strate- outcomes, which, if taken literally, leave no
gic interaction(McCabe calls it neuroeconom- modeled economic reasonfor choosing between
ics) using fMRI and other brain-imaging them.
technologies. That researchexplores the neuro- More generally, using rational theory, one
correlatesof intentions or "mind reading,"and representsan observed socioeconomic situation
otherhypotheses aboutinformation,choice, and with an abstractinteractivegame tree. Contrar-
own versus other payoffs in determininginter- ily, the ecological concept of rationality asks
active behavior. from whence came the structurecapturedby the
The above themes will be illustratedand dis- tree? Why this social practice, from which we
cussed in a wide variety of examples drawn can abstracta particulargame, and not another?
from economics, law, experimentaleconomics, Were there otherpractices and associated game
and psychology. I will begin with impersonal trees that lacked survival properties and were
exchange through markets, drawing on the successfully invaded by what we observe?
learning from experiments and field observa- There is a sense in which ecological systems,
tions to illustratehow the contrastbetween con- whetherculturalor biological, must necessarily
structive and ecological rationality informs be, or are in the process of becoming, rational:
learningfrom observation.Then I will examine they serve the fitness needs of those who unin-
personalexchange, particularlyin the context of tentionally created them through their interac-
two-person extensive-form games, asking why tions. Constructivistmental models are based
constructivistmodels are of limited success in on assumptions about behavior, structure,and
predictingbehavior in single-play games, even the value-knowledge environment. These as-
when subjects are anonymously matched. sumptions might be correct, incorrect, or irrel-
evant, and the models may or may not lead to
rational action in the sense of serving well the
22 needs of those to whom the models apply. As
I importunestudentsto read narrowlywithin econom-
ics, but widely in science. Within economics there is essen- theorists the professional charge for which we
tially only one model to be adapted to every application: are paid is to formulateand prove theorems. A
optimization subject to constraintsdue to resource limita- theorem is a mapping from assumptions into
tions, institutionalrules, and/orthe behaviorof others, as in testable or observable implications. The de-
Coumot-Nashequilibria.The economic literatureis not the
best place to find new inspirationbeyond these traditional mandsof tractabilityloom large in this exercise,
technical methods of modeling. and to get anything much in the way of results
23 it is necessary to consider both the assumptions
Thus, for Hayek, "an economist who is only an econ-
omist cannot be a good economist." and their implications as variables. Few game
This example illustratesthe use of the labo- bed to examine the performanceof proposed
ratory in economic systems design. In these new institutions, and modifies their rules and
exercises we can test-bed alternative market implementationfeatures in the light of the test
auctionrules and the effect of transmissioncon- results. The proposed designs are initially con-
straintson generatorsupply behavior(Steven R. structivist,although most applications, such as
Backerman et al., 2001), vary the degree of the design of electricity marketsor auctions for
marketconcentration,or "power"in a noncon- spectrum licenses, are far too complicated for
vex environment (Michael J. Denton et al., formalanalysis (JeffreyBanks et al., 2003; Ras-
2001), compare the effect of more or less stra- senti et al., 2003).
tegic demand responsiveness (Rassenti et al., But when a design is modified in the light of
2003), study network and multiple market ef- test results, the modifications tested, modified
fects also in a nonconvex environment (Mark again, retested, and so on, one is using the
Olson et al., 2003), and test-bed markets to laboratoryto effect an evolutionary adaptation
inform, but not finalize, market liberalization as in the ecological concept of a rationalorder.
policy (Rassenti et al., 2002). For a survey of If the final result is implementedin the field, it
many examples, see McCabe et al. (1991). certainlyundergoesfurtherevolutionarychange
The two types of rationalorder are both ex- in the light of practice,and of operationalforces
pressedin the experimentalmethodologydevel- not tested in the experimentsbecause they were
oped for economic systems design. This branch unknown,or beyond currentlaboratorytechnol-
of experimentaleconomics uses the lab as a test ogy.27 In fact this evolutionary process is es-
sential if institutions, as dynamic social tools,
are to be adaptive and responsive to changing
conditions. How can such flexibility be made
power (Wilson, 2002, p. 1332). Beware this simplistic pop-
ularexplanation:it is a two-wrongs-make-a-rightargument: part of their design? We do not know because
yes, of course, given that you were going to protect the no one can foresee what changes will be
monopoly power of the distributionutilities to tie the rental needed.
of the wires to the metering and sale of energy at a fixed
regulatedprice, then one way to protect them temporarily MarketInstitutionsand Performance.-Non-
from the consequent wholesale price volatility might be to
encourage long-term contracts at a fixed average delivery cooperativeor Cournot-Nashcompetitive equi-
cost. But supplierswill want higherprices and/orshort-term librium (CE) theory has conventionally offered
contractsif they anticipateshortages-you cannotget blood two specificationsconcerningthe preconditions
out of a turnip;long contractswork to lower cost only to the for achieving a CE: (1) agents requirecomplete,
extent that suppliers are surprisedby high spot prices, but or "perfect,"informationon the equations de-
when it comes time to renegotiate expiring contracts they
will not replicatethe error.Californiadiscovered this when fining the CE; also common knowledge-all
they intervened to sign long-term contracts, and encoun- must know that all know that all know that they
tered high prices. This whole argument turns the design have this information. In this way all agents
problemon its head. You must (1) remove the legal power have common expectations of a CE and their
of the local wires monopoly to prevent competing energy
behavior must necessarily produce it; (2) an-
suppliersfrom contractingwith customers to discount off-
peak energy, charge premiumsfor peak energy, and install other tradition, popularly articulated in text-
the supportingcontrol devices; (2) let this competition de- books, and showing, perhaps, more sensitivity
termine the dynamic price structure, and investment re- for plausibility,has arguedfor a weakerrequire-
quired to implement it; (3) simultaneously, let financial ment thatagents need only be price-takersin the
instrumentsevolve to hedge whatever risk is left over as
prices become less volatile. Financialinstrumentscan hedge
market.
price volatility, not load volatility. Only demand-responsive
interruptibleloads can relieve supply stress and provide the
demand-sidereserves that reduce the risk of lost load. No
27
one mind or collective can anticipateand plan the needed People often ask, What are the limits of laboratory
mix of technologies to enable the market to manage de- investigation?I think any attempt to define such limits is
mand. Thereforeit is essential to remove all entry barriers, very likely to be bridged by the subsequentingenuity and
and allow firms to experimentthroughcompetition to dis- creativity (the primarybarriersat any one time) of some
cover and innovate efficient ways of organizing retail de- experimentalist.Twenty-five years ago I could not have
livery systems. Claims that short-run retail demand is imagined being able to do the kinds of experiments that
"notoriously"inelastic miss the point:how would you know today have become routinein our laboratories.Experimen-
if loan-sheddingtechnology is inflexible? Competitionand talists also include many of us who see no clear border
incentives to innovatehave never been partof the structure. separatingthe lab and the field.
all those thatwill not impose a loss on the agent. simultaneousinterdependentmarketsshow sim-
Thus, agents who are not rationalconstmructivist ilar patternsof convergence(Plott, 1988, 2001).
profit maximizers, and use no learning or up-
dating algorithms,achieve most of the possible The Iowa Electronic Market.-What evi-
social gains from trade using this institution. dence do we have that the laboratoryefficiency
Does this example illustrate in a small way propertiesof continuousdouble auction trading
those "super-individualstructureswithin which apply also in the field? One of the best sources
individuals found great opportunities... (and of evidence, I believe, is found in the Iowa
that) ... could take account of more factual cir- Electronic Market (IEM) used widely around
cumstancesthanindividualscould perceive, and the world (RobertForsythe et al., 1992, 1999).
in consequence ... is in some respects superior These marketsare used to study the efficacy of
to, or 'wiser' than, human reason ... "? (Hayek, futuresmarketsin aggregatingwidely dispersed
1988, pp. 77, 75). informationon the outcomes of political elec-
We do not know if the GS results generalize tions, or any well-defined extra-laboratory
to multiple market settings as discussed in the event, such as a change in the discount rate by
next paragraph.Ross M. Miller (2002), however, the Fed. The "laboratory"is the Internet.The
has shown thatin a very elementarytwo-market "subjects"are all who log on and buy an initial
environment-intertemporally separated mar- portfolio of claims on the final event outcomes;
kets for the same commodity-the GS results they consist of whomever logs in, and are not
are qualified. Complex price dynamics, includ- any kind of representativeor "scientific"sample
ing "bubbles,"appear,and there is loss of effi- as in the polls with which they are paired. The
ciency, althoughthe loss is not substantial.On institutionis the open-book double auction.
average the decline is apparentlyfrom around In the IEM, tradersmake a marketin shares
94 percent to 88 percent. representingpair-mutuelclaims on the popular
In multiple markettradingin nonlinearinter- vote (or winner-take-all)outcome of an elec-
dependentdemandenvironments,each individ- tion, referendum, etc. For example, the first
ual's maximumwillingness-to-payfor a unit of IEM was on the 1988 presidential election.
commodity A depends on the price of B, and Each person wanting to trade shares deposits a
vice versa, and in this more complex economy minimumsum, $35, with the IEM and receives
double auction, markets also converge to the a tradingaccount containing $10 cash for buy-
vector of CE prices and trading volumes. A ing additionalshares, and ten elemental portfo-
two-commodity example is reported in Smith lios at $2.50 each, consisting of one share of
(1986), based on nonlineardemand (CES pay- each of the candidates-Bush, Dukakis,Jackson,
off function) and linear supply functions found and "rest-of-field."Tradingoccurs continuously
in Arlington Williams and Smith (1986); also in an open-book bid-ask market for several
see Williams et al. (2000). In these experiments, months, and everyone knows that the market
numericaltablesbased on the preferenceand cost will be called (tradingsuspended)in November
information defining the general-equilibrium on election day, when the dividendpaid on each
solution of four nonlinear equations in two share is equal to the candidate'sfraction of the
prices and two quantities are dispersed among popularvote times $2.50. Hence if the final two
the undergraduatesubjects. They buy and sell candidates and all others receive popular vote
units of each of the two commoditiesin a series shares (53.2 percent,45.4 percent, 1.4 percent),
of trading periods. Prices and trading volume these proportions (times $2.50) represent the
converge, after several trading periods, to the payoff to a traderfor each share held. Conse-
CE defined by the nonlinear equations. The quently, at any time t, normalizing on $1, the
subjects would not have a clue as to how to price of a share (+$2.50) reflects the market
solve the equationsmathematically.The exper- expectationof that candidate'sshareof the total
imenter applies the tools of constructivistrea- vote. A price, $0.43, means the marketpredicts
son to solve for the benchmarkCE, but in repeat that the candidate will poll 43 percent of the
play this "solution"emerges from the spontane- vote. Otherforms of contractthat can be traded
ous ordercreatedby the subjects tradingunder in some IEMs include winner-take-all,or num-
the rules of the double-auctionmarket institu- ber of seats in the House, and so on.
tion. Numerous other experiments with many The IEM data set includes 49 markets, 41
worldwide elections, and 13 countries. Several We have seen that markets economize on
results standout: the closing marketprices, pro- information, understanding, the number of
duced by a nonrepresentativesample of traders, agents, and individualrationality.Can they also
show lower average absolute forecasting error economize on the need for externalintervention
(1.5 percent) than the representativeexit poll to protect particularinterests,if all are empow-
samples (1.9 percent); in the subset of 16 na- ered by the trading institution to act in their
tional elections, the market outperforms the individual interests?
polls in 9 of 15 cases; in the course of several
months preceding the election outcome, the B. Strategy-proofness:Theoryand Behavior
market predictions are consistently much less
volatile than the polls; generally, larger and Preferences are private and unobservable,
more active marketspredictbetterthan smaller, and institutions have to rely on the messages
thinner markets;surveys of the market traders reportedby agents, not their true preferences.
show that their share holdings are biased in This follows from the fact that no one mind has
favor of the candidatesthey themselves prefer. all the informationknown togetherby all those
In view of this last result why do markets in the market. It is therefore possible for an
outperform the polls? Forsythe et al. (1992) agent to affect prices and outcomes in a market
argue that it's their marginaltraderhypothesis. by strategically misreportingpreferences. This
Those who are active in price "setting,"that is, prospect has motivated the literature seeking
in entering limit bids or asks, are found to be strategy-proof mechanisms: "An allocation
less subject to this bias, than those tradersac- mechanism is strategy-proof if every agent's
cepting (selling and buying "at market") the utility-maximizingchoice of what preferences
limit bids and asks. Polls record unmotivated, to report depends only on his own preferences
representative,average opinion. Marketsrecord and not on his expectationsconcerningthe pref-
motivated marginal opinion that cannot be de- erences that other agents will report" (Mark
scribed as "representative."This analysis helps Satterthwaite,1987, p. 519). This requireseach
to provide a good mechanical, if not ultimate, agent to have a dominantstrategyto reporttrue
understandingof how human interaction with preferences, and has led to impossibility theo-
the rules of a bid/ask CDA yield efficient rems establishing the nonexistence of such a
predictions.30 mechanism under certain conditions.
In view of such negative theoretical results
and the narrow conditions under which solu-
30
Other markets besides the IEM are known to have tions have been investigated, it is importantto
efficient information-aggregatingproperties. Pari-mutuel ask what people actually do in experimental
racetrackmarketsare an example where, interestingly,the environments in which the experimenter in-
environmentis much like the IEM: the settlementsoccur at duces preferences privately on individual sub-
a well-defined end-state known to all agents, unlike stock
market trading where expectations float continuously with jects. We know what is impossible, but what
no clear value revelation endpoint. 'The racetrackbetting is possible in more open-ended systems than
marketis surprisinglyefficient. Marketodds are remarkably are modeled by theory?Is it possible that when
good estimates of winning probabilities.This implies (sic) all are free to choose from a large space of
that racetrackbettors have considerableexpertise, and that
the marketsshould be taken seriously"(Thalerand William strategies, ecologically rational strategies will
T. Ziemba, 1988, p. 169). It is surprising to behavioral emerge that immunize against strategic manip-
economists because their methodology is restrictedto look- ulation? Given that information is inherently
ing for deviations from the standardmodel. What is unusual
here is that in racetracksthey have found reportableevi-
dence for efficient outcomes. For those who follow the
experimental economics, IEM and similar controlled- latter more pronounced in the last two races of the day.
environmentmarketstudies, efficiency is not only common- Various hypotheses have been offered to explain these
place (if not universal),it cannotbe attributedto agents with inefficiencies, but more significant is that computer pro-
"considerableexpertise." The agents are mostly naive, al- grams have been written to arbitrage(yielding returns of
though they get repeat interactionexperience, which, from some 11 percent per bet), the place, show, and long-shot
the evidence, clearly gives them expertiseenough. But, as in inefficiencies. (It is my understandingthatgood profitshave
the IEM and experimental markets, racetrackmarkets are been accumulatedon these programs,so far without neu-
not perfect: there are inefficiencies in the "place" and tralizing the arbitrage opportunities-let the good times
"show" options and the favorite-long-shot bias, with the roll!)
TABLE1-SUMMARYOFRESULTS: UPDA EXPERIMENT order.... Neither all ends pursued, nor all
UP43; 5, 5 means used, are known or need be known to
Pe Qe Pr Qr Eff% Rev%
anybody, in orderfor them to be taken account
of within a spontaneous order. Such an order
1 295 18 300 16 91% 22% forms of itself...."
2 405 18 400 18 100% 7%
3 545 18 540 18 100% 14%
4 460 18 448 18 92% 14% C. Gresham's Law: If It Isn't Cournot-Nash,
5 360 18 350 18 100% 9% WhyIs It a Law?
6 500 18 500 18 98% 12%
7 260 18 250 17 96% 26% In this section I have given many examples of
8 565 18 553 15 92% 28%
9 300 18 300 18 100% 28% institutionsin which the CE theory of markets
10 610 18 610 18 100% 33% predicts their observed behavior. Do we have
11 365 18 350 15 85% 88% contraryexamples? Yes. Gresham's Law: bad
12 550 18 558 15 88% 55%
13 450 18
money drives out good. This "law," while
450 18 100% 31%
14 410 18 410 18 100% 5%
sometimes claimed to be an observed phenom-
15 485 18 484 19 89% 39% ena in countries all over the world, is not a
Cournot-Nash equilibrium.32If currencies A
/1 = 17.3 95% 27%
and B are both available, A having an intrinsic
o-= 1.3 5 21
worth while B is worthless fiat money, then the
Notes: (Pe, Qe) = equilibrium price and quantity. (Pr, theory predicts that A will drive out B. This is
Qr) = realized price and quantity. Eff% = efficiency, % because each agent believes other agents are
max surplus. Rev% = % of surplusrevealed.
rational, and will accept only A in exchange.
Each agent will therefore avoid getting stuck
with the inferiorB by accepting only A, which
can manipulatethe price to his or her advantage. becomes the dominant circulating medium of
But this parableis irrelevant.The relevantques- exchange, while B is "horded."Experiments
tion is what behavior is manifest when every have confirmedthat if both types of money are
agent has the potential for manipulating the initially available,subjectsuse only the superior
price. Without knowledge or understandingof currency (an interest-bearingconsol) as a me-
the whole, and without design or intention, the dium of exchange. But in treatmentsin which
participantsuse the rules at their disposal to subjects first experience a history of using fiat
achieve three propertiesobserved by the exper- money, it being the only medium of exchange
imenter:(1) high efficiency, (2) maximumindi- available, and then the consol is introduced,
vidual profit given the behavior of all other subjects continue trading with the fiat money,
agents, and (3) protectionfrom manipulationby hording the interest-bearingconsol (Gabriele
their protagonists.31This ecological result illus- Cameraet al., 2003). This is entirely rationalif
tratesthe perceptiveinsight of Hayek (1988, pp. each agent believes others will accept the fiat
19-20). "Rules alone can unite an extended money in exchange and this belief is supported
by experience. Think of Gresham's Law as a
31 Space
belief equilibriumin which theory alone is un-
prevents me from dealing fully with the many able to predict when it might occur (Ledyard,
importantissues raised when a subset of agents have asym-
metric advance information on product quality or value 1986).
characteristics. The analysis shows that such conditions Complementingthese results, anotherexper-
generatemarketfailure or inefficiency. Some of these prob- imental study shows thatwhen fiat money is the
lems, however, arise because the analysis is inadequatein
only currency, it will be used even under the
examining both sides of the market,and the implicationsof
the informationcontent of prices. Experimentshave estab-
lished that constructivistinefficiency is often alleviated by
one of several ecologically rationalresponse mechanisms: 32
Hayek (1967, p. 318) notes that Gresham'sLaw is not
competition among sellers for reputations,quality (brand) due to Greshamnor is it a "law"in the theoreticalsense, and
signaling, productwarranties,and the aggregationof private "... as a mere empiricalrule is practicallyworthless."In the
asymmetricinformationinto public price patternsthat self- 1920's when people started using dollars and other hard
correct the alleged problems. See, e.g., Plott and Louis currencies in substitution for the depreciating mark, the
Wilde (1982); Plott and Sunder (1982, 1988); Miller and claim emerged thatGresham'sLaw was wrong-that it was
Plott (1985); Camererand Keith Weigelt (1988). the other way around.
pia. These comments suggest that much insight then we have the standardown-maximizingthe-
mightbe obtainedfromthe systematicstudyof the ory. Consequently, Franciosi et al. (1995) can
conditionsunderwhich surveyresultsarerobustly test the hypothesis, never using the word "fair-
informativeandthe conditionswherethey arenot. ness," that if subject buyers have a utilitarian
concern for profits not being increased relative
F. Fairness: An ExperimentalMarket Test to a baseline then after a change from the base-
line this should alter the observed equilibrium
In developing a descriptive theory48of the relative to the standardpredicted equilibrium
"referencetransaction,"KKT state that what is with no externaleffect, a = 0. In a posted offer
considered "fair" may change: "Terms of ex- marketgiving KKT their best shot (sellers can-
change.that are initially seen as unfair may in not see each other's posted prices, and therefore
time acquire the status of a reference transac- cannot knowingly emulate or undercut each
tion" (KKT, 1991, p. 203). This paves the way other's prices), Franciosi et al. (1995) find that
for the adaptation of "fairness beliefs" to when a = 0 (implementedby either no disclo-
changes in the competitive equilibrium. Al- sure, or by marginalcost-justifying disclosure)
thoughthe competitivemodel is the one thathas the marketconverges quickly to the new com-
static predictive content, its prediction is silent petitive equilibrium.When a > 0 (implemented
as to how long it will take to respond to a by profit 'r and wrodisclosure) prices converge
change in parameters.KKT's argumentsare not more slowly, but precisely, to the new equilib-
predictive, but they tell a story about why mar- rium. Hence, under conditions most favorable
kets might be sluggish in respondingto change. to a "fairness"effect, the response dynamics is
How good is their story? changed,but not the equilibriumas predictedby
Franciosi et al. (1995) state a preference the standardcompetitive model. The discipline
model of optimal choice that allows for a utili- of the market swamps all but a transient"fair-
tariantrade-off between own consumptionand ness" effect. If, realistically,sellers can see each
"fairness."For example, the utility of two com- others prices, I would predict a much smaller
modities (x, z) is given by: u(x, z) = z + ax - "fairness"effect, if any.
(b/2) x2 - ax[(r'/ro) - 1], in which the sell-
er's profit, 7r,relative to a reference profit, iro, m. PersonalSocial Exchange
appearsas an "externality"in the buyer's utility
function. The usual maximizationsubject to an One of the most intriguing discoveries of
income constraint yields the inverse demand experimentaleconomics is that (1) as we have
equation:p = a - bx - a[(r/7r0) - 1]. Thus, seen, people invariably behave noncoopera-
for a > 0 any change in the environmentthat tively in small and large group "impersonal"
increases a firm's profitrelative to the reference market exchange institutions;(2) many (up to
profit has an external effect that lowers the half in single play; over 90 percent in repeat
buyer's inverse demand for units x. If a = 0, play) cooperate in "personal"exchange (two-
person extensive-form games); (3) yet in both
economic environmentsall interactionsare be-
48
This methodology is driven by the untenable belief tween anonymousplayers. In this section I shall
that general theories can be derived directly from observa-
tions if you just have enough data (see Smith, 2002, and the attempt to summarize some of the most com-
references therein). "Perhaps the most important lesson pelling evidence of cooperationin personal ex-
learnedfrom these studies is thatthe rules of fairnesscannot change-in the field as well as the laboratory-
be inferredeither from conventionaleconomic principlesor and review some of the test results designed to
from intuition and introspection.In the words of Sherlock discriminateamong the more prominentpredic-
Holmes in 'The Adventureof the Copper Beaches': 'Data!
Data! Data! I cannot make bricks without clay'" (KKT, tive hypotheses for modeling cooperative be-
1986, pp. 115-16). Neither can a predictivetheory of "fair- havior. Whatevermight be the most useful way
ness" be inferredfrom any amount of the KKT data. If N to model and explain cooperation, unaided by
"fairness"rules are discovered by trial and errormodifica- market incentives, my working hypothesis
tions in the survey questionnaires, you cannot reject the
hypothesis that there is an N + 1 variationthat will identify
throughoutis that it is a productof an unknown
a new one. More data will not help, as the fairness concept mix of cultural and biological evolution, with
is used here as a word that provides no effective means of the biology providingabstractfunction defining
modifying standardtheoryto correctfor its predictiveflaws. potential, and culture shaping the emergent
forms that we observe. But to motivate the informalrules andgave voice to them, as God's,
whole exercise in thought, I will begin with a or natural, law.50 The common lawyer, Sir
discussion of some persistentcross-culturalso- EdwardCoke, championedseventeenth-century
cial practicesfrom business, law, anthropology, social normsas law commandinghigher author-
and Americaneconomic history. ity than the king. Remarkably,these forces pre-
How might a social rule (practice, norm) vailed, paving the way for the rule of law in
emerge, become a cultural fixture, and be England.51Similarly, the cattlemen's associa-
widely emulated?I will use a parableto illus- tions, land clubs, and mining districts in the
trate how a rule for "bargainingin good faith" AmericanWest all fashionedtheirown rules for
might become established. establishingpropertyrights and enforcingthem:
In bargaining over the exchange price be- the brandon the hindquartersof his calf was the
tween a buyer and seller, suppose the seller cattlemen's indelible ownership signature on
begins by announcinga selling price, the buyer his property,enforcedby gunmenhired through
responds with a lower buying price, the seller his cattle club;52squatter'srightswere defended
reduces his asking price, and so on. In this ably (possession is nine points of the law?) by
concessionaryprocess it is consideredbad form the land clubs composed of those brave enough
for the buyer (or seller), once having made a to settle wildernesslands in advanceof veterans
concession, to return to a lower (or higher) exercising their land script claims, and of set-
price. This violates a principleof "bargainingin
good faith" (see Siegel and Fouraker,1960, p.
20). How might this come about?One can sup- gio G. Lazzarini et al. (2002) for new results suggesting
pose thatthose who fail to bargainin good faith that they are complements-contracts facilitate the self-
would be less likely to be sought out by others enforcementof relationalelements beyond contractibility.I
would hypothesize that both must be true: constructivist
for repeat transactions. Such behavior raises rules ultimately must pass fitness tests of ecological ratio-
transactionscost by increasingthe time it takes nality. Formal rules that are incompatible with informal
to complete a sale. Trading pairs would be rules will be modified or eliminated;those that are compat-
expected to self-select, tending to isolate the ible will persist.Hence, at any time slice in history,both must
necessarilybe observedacrossall socioeconomicexperiments.
more time-consumingbargainers,and it would 50"... ([A]ll early (my insertion: as in Sumar with the
take them longer to find those willing to tolerate beginning of writing) 'law-giving' consisted in efforts to
the time cost of bargaining. Such practices- record and make known a law that was conceived as unal-
inherentlyeconomizing in this parable-might terably given. A 'legislator' might endeavor to purge the
law of supposed corruptions,or to restore it to its pristine
then become part of a culturalnorm, powerful
purity, but it was not thought that he could make new
enough to be codified ultimatelyin contractlaw law. ... But if nobody had the power or intentionto change
and in stock exchange rules. Proposition:in this the law ... this does not mean that law did not continue to
mannercollectives discover law in those rules develop" (Hayek, 1973, p. 81).
51
that persist long enough to become entrenched What allowed the rule of "natural"or found law to
prevail in England "... was the deeply entrenchedtradition
practices. In this example the emergent rule of a common law that was not conceived as the productof
reduces transactioncost, leaving open the clas- anyone's will but ratheras a barrierto all power, including
sical question of how equilibriumcan be char- that of the king-a tradition which Edward Coke was to
acterizedin bilateralbargaining. defend against King James I and FrancisBacon, and which
Matthew Hale at the end of the seventeenth century mas-
terly restated in opposition to Thomas Hobbes" (Hayek,
A. SpontaneousOrder Withoutthe Law49 1973, p. 85; also see pp. 167, 173-74).
52 These voluntary private associations for sharing the
The early "law-givers"did not make the law cost of a common good-policing-were subsequentlyun-
dermined by statehood, and the publicly financed local
they "gave";they studied social traditionsand sheriff as the recognizedmonopoly law enforcementofficer.
This observationcontradictsthe myth that a centralfunction
of government is to "solve" the free-riderproblem in the
privateprovisionof public goods. Here we have the reverse:
49 the incentive of the cattlemen'sclubs was to free ride on the
Experimental studies have inquired as to whether
emergentnorms of cooperationand constructivistincentive general taxpayer,assigning the sheriff the task of enforcing
schemes are substitutes,the lattercrowding out the former. propertyrights in cattle. The same free-ridingoccurs with
See Iris Bohnet et al. (2001) and Fehr and Simon Gachter school busing programs,and in publicly providededucation
(2002) for studies suggesting that they are substitutes(for- itself in which governmentfinancingneed not requiregov-
mal rules undermineinformalcooperativenorms), and Ser- ernmentprovision.
tiers under the Homestead Act; mining claims brilliantlyby Coase, was to deal with the ques-
were defined, established, and defended by tion of efficient liability rules in a world of
the guns of the mining clubbers, whose rules significanttransactionscost. He then proceeded
were later to become part of public mining law to use the transactionscost frameworkto exam-
(TerryL. Andersonand PeterJ. Hill, 1975; John ine the problem of social cost in a variety of
Umbeck, 1977). For over a century, the Maine legal precedentsand cases.54
lobstermenhave establishedrights, used threats, In the beginning Shasta County California
then force, to defendexclusive individuallobster- was governed by "open range" law, meaning
fishing territoriesin the ocean (James Acheson, that in principle ranchersare not legally liable
1975). Eskimo polar bear hunting teams for damagesresultingfrom their cattle acciden-
awardedthe upperhalf of the bear's skin (prized tally trespassing on unfenced land. Then, in
for its long mane hairs used to line women's 1945 a California law authorized the Shasta
boots) to thatperson who first fixed his spear in County Board of Supervisors to substitute a
the prey (Peter Freuchen, 1961). Extanthunter- "closed range" ordinance in subregions of the
gatherershave evolved sharingcustoms for the county. Dozens of conversions have occurred
products of communal hunting and gathering. since this enabling law. Under a closed range
For example, the Ache of Eastern Paraguay law the rancher is strictly liable-even if not
share the volatile products of the hunt widely negligent-for damage caused by his livestock.
within the tribe,while the low varianceproducts Robert C. Ellickson (1991), out-Coased Coase
of gatheringare sharedonly within the nuclear by, in effect, asking, "Given that this county
family (Hillard Kaplan and Kim Hill, 1985; applies the polar legal rules used in Coase's
Kristen Hawkes, 1991). illustration,how do neighborsin Shasta County
actually handle the problem of stray cattle?"
B. Ellickson Out-Coases Coase The answer: "Neighbors in fact are strongly
inclined to cooperate,but they achieve cooper-
Using the rancher/farmerparable,Ronald H. ative outcomes not by bargainingfrom legally
Coase (1960) arguedthat if there were no costs established entitlements,55as the parable sup-
of transacting, then theoretically efficiency poses, but rather developing and enforcing
could not depend on who was liable for dam- adaptive norms of neighborliness that trump
ages to crops caused by stray animals. Legal formal legal arrangements.Although the route
liability gives the rancher an incentive to em- chosen is not the one thatthe parableanticipates,
ploy cost-efficient measures to control straying the end reached is exactly the one that Coase
cattle. But if she were not liable, then in a world predicted: coordination to mutual advantage
of zero transactionscost, victims would be led without supervision by the state"56(Ellickson,
in their own interest to negotiate a settlement
paying the rancherto undertakethe same effi-
cient control measures induced by legal liabil-
Roger B. Myerson, 1991, p. 506). These cases may limit
ity. In so doing, trespassvictims save the cost of extensions of the Coase Theorem,but do not, I think,detract
crop damage, assumed to be more than the cost from its essential message that the locus of liability was
of cattle control-otherwise it is inefficient to irrelevant.
control them. The externalityis internalizedby 54Coase (1974) also noticed that the lighthouse was
market negotiation incentives. Curiously, the frequently cited by theorists as an example of a "pure"
public good. As was his style (to confront the casual para-
Coase Theorem-that in the absence of trans- bles of theory that finessed certain costs by fiat), his re-
actions cost efficiency does not dependupon the sponse in effect was, "Well, let's see what people have done
locus of liability-was controversial. It was who actuallyoperatelighthouses,or who use the services of
clearly intended as a kindly spoof of oversim- lighthouses."It turnedout that early lighthouses were pri-
vate enterprise,not government,solutions to a public good
plified theories that, in particular,ignored trans- problem, and the alleged inevitability of free-riding was
actions cost.53 The real problem, addressed solved by the owner who contractedwith port authoritiesto
collect lighthouse fees when ships arrivedportside to load
or unload cargo.
55These are the outside
53 options or threatpoints in game
Later game-theoretic formulationshave allowed that theory.
with two or more alternatives there may exist "standoff 56
The same results emerge in laboratoryexperiments
equilibria" that stall agreement in Coase bargaining (see reportedby Hoffman and Matthew L. Spitzer (1985).
1991, p. 4). Thus, Shasta County citizens, in- ward, and procedural conditions that vary
cluding judges, attorneys,and insuranceadjus- elements of social distance. Again, studying
tors, do not have full working knowledge of what is not helps us to understandwhat is.
formallocal trespasslaw.57Citizensnotifyowners
and help catch the trespassinganimal;use men- D. Perception and the Internal Order of the
tal accounting(reciprocity)to settle debts, e.g., Mind: Why ContextMatters
a rancher,whose cattle have strayedmay tell the
victim to come down and take some hay, or if Two decision tasks, representedby the same
your goat eats my tomato plants, you offer to abstract game tree, may lead to different re-
help me replantthem;use negative gossip, com- sponses because they occur in different con-
plain to officials, submit informal claims for texts. Why? The answer may be found in the
money (but not through a lawyer) to punish process by which we perceive the external
deviant neighbors; rarely use lawyers to seek world. Hayek (1952)58 was a pioneer in devel-
monetary compensation;share the building of oping a theory of perception,which anticipated
fences, most often by a rule of proportionality- recent contributionsto the neuroscienceof per-
you pay more if you have more animals than ception. It is naturalfor our minds to suppose
your neighbor; ignore fence law as irrelevant; that experience is formed from the receipt of
and do not change fence obligations with the sensory impulses that reflect unchanging at-
planting of crops. Finally, contraryto the Coa- tributesof external objects in the environment.
sian parable, the main cost of trespass is not Instead, Hayek proposed that our currentper-
from crop damage,but from highway collisions ception results from a relationshipbetween ex-
that kill animals and damage property. ternal impulses and our past experience of
similar conditions. Categories formed in the
C. Extensive Form InteractionsBetween mind are based on the relative frequency with
AnonymouslyPaired Individuals which current and past perceptions coincide.
Memory consists of external stimuli that have
Cooperationhas also emerged in anonymous been modified by processing systems whose
two-personextensive-formgames in laboratory organizationis conditionedby past experience59
experiments. Although such behavior is con- (Hayek, 1952, pp. 64, 165). Thereis a "constant
traryto rationalprescriptions,it is not inconsis- dynamic interaction between perception and
tent with our examples of spontaneous order memory, which explains the ... identity of pro-
without externally imposed law. cessing and representationalnetworks of the
Why do we study anonymousinteractionsin cortex that modem evidence indicates." "Al-
the laboratory?The model of nonrepeatedgame though devoid of mathematical elaboration,
theory is about strangerswithout a history or a Hayek's model clearly contains most of the
future (Robert W. Rosenthal, 1981), but ano- elements of those later network models of as-
nymity has long been used in small group ex- sociative memory ..." (JoaquinM. Fuster, 1999,
periments to control for the unknown pp. 88-89).
complexities of naturalsocial intercourse(Sie- Hayek's model capturesthe idea that, in the
gel and Fouraker,1960). It is well documented internal order of the mind, perception is self-
that face-to-face interaction swamps subtler organized:abstractfunction combines with ex-
proceduraleffects in yielding cooperative out- perience to determinenetworkconnectivity and
comes (Hoffman and Spitzer, 1985; Kagel and expansion.60Loss can occur eitherfrom lack of
Roth, 1995). But more important,I believe it is
this condition that provides the greatest scope 58
The Sensory Orderwas not publisheduntil 1952 when
for exploring the human instinct for social ex- Hayek revised a manuscript, originally written in the
change, and how it is affected by context, re- 1920's, entitled in English translation,"Contributionsto a
Theory of How Consciousness Develops" (noted in corre-
spondence to me by Bruce Caldwell).
57Under 59
The interdependencebetween perceptionand memory
open range the animal owner is liable for in-
tentionaltrespass,trespassof a lawful fence, and trespassby is revealed by the different descriptionsof the same event
goats, whateverthe circumstances,suggesting the hypothe- by two eyewitnesses (Gazzanigaet al., 1998, pp. 484-86).
60Built into your brainis the maintainedhypothesis that
sis that goat behaviorhad long been recognized in pastoral
norms and now capturedin codified law. the world around you is stationary.Look at the wall and
function or the stimulus of developmental ex- suggestive words like "game," "play,"
perience. Block or distort sensory input, and "players," "opponent," and "partner"(except
function is impaired;impair function by brain where variationson the instructionsare used as
lesions or inherited deficiency, and develop- systematic treatmentsto identify their effect);61
ment is compromised. rather,reference is made to the "decision tree,"
This model is consistent with the hypothesis "decision maker 1" (DM1) or 2 (DM2), and
that the mind is organized by interactivemod- "your counterpart,"etc. The purpose is to pro-
ules (circuits)that are specialized for vision, for vide a baseline context, which avoids emotive
language learning, for socialization, and a host words that might trigger unintendedmeanings
of other functions (see Leda Cosmides and John by the experimenter. I do not mean that the
Tooby, 1992; Pinker, 1994). In this view, mind baseline is "neutral," a concept that is not
is the unconscious product of coevolution be- clearly definable,given that context effects can
tween the biological and cultural development depend on autobiographicalexperience.The ef-
of our brains that distinguished us from other fect of instructionalvariationon decision is an
primates.It was what made reasonpossible. Our empirical matter and any particularset of in-
folk predilection for believing in the "blank structions must always be considered a treat-
slate" concept of mind (Pinker, 2002) makes ment unless the observations are shown to be
plain that this interpretationof mind is just as robustto changes in the instructions.All obser-
consonant with our direct experience as was vations must be seen as a joint product of ex-
once the idea that the earth is flat, or that perimental procedures and the theoretical
witches had to be destroyed. In each case to hypotheses, implementedby particularparame-
escape from the folk perception requires the ters that it is our intention to test. This is not
falsifying indirectevidence, based on reason, to unique to laboratoryobservations,but a charac-
become partof our "felt"experience. Construc- teristic also of field observations,and the whole
tivist rationality then becomes ecologically of science (see Smith, 2002, for examples from
rational. physics, astronomy,and experimentaleconom-
ics). It is thereforeimportantto understandhow
E. ExperimentalProcedures procedures as well as different parameteriza-
tions (games, payoffs) affect behavior.
The experimentsI will reportwill show how Subjects are recruitedin advance for an eco-
social context can be importantin the interactive nomics experiment. Upon arrival at the ap-
decisionbehaviorwe observe.Thispossibilityfol- pointed time they register, receive a show-up
lows from the autobiographicalcharacter of fee, and are assigned to a private computer
memory and the mannerin which past encoded terminalin a large room with 40 stations.Com-
experience interacts with currentsensory input monly there are 11 other people, well spaced
in creating memory. I will be reporting the throughout the room, in the experiments re-
results of decision-makingin single play, two- ported below. After everyone has arrived,each
person, sequential-movegame trees. Subjectin- person logs into the experiment,reads through
structions do not use technical and role- the instructionsfor the experiment,responds to
instructionalquestions,and leams thathe or she
is matchedanonymouslywith anotherperson in
move your eyes back and forth with head still. The wall the room, whose identity will never be known,
does not move. Now press your eyeball with your finger
through the eyelid from the side. The wall moves as you
61
jiggle your eyeball. Why the difference?When you flex the See Terence Bumham et al. (2000), discussed below,
eye muscles and move your eyes back and forth, a copy of for a study comparing the effect of using "partner"and
the signal goes to the occipital cortex to offset apparent "opponent"in a trust game.
movement of the wall so that the net perceptionis that of a 62It is not
meaningful or helpful to talk about "experi-
stationary wall. This stabilizing self-ordered system for menter effects." There are instructionaland proceduralef-
seeing also makes you vulnerable to optical illusions of fects, including the presence or absence of an experimenter,
motion. Moving your eyes back and forth between the what he/she knows or does not know (as in double blind
tunnel gate and your airplaneas it docks, you ambiguously behavioralexperiments),and what he/she does or does not
"sense" that either the gate or the plane, or both, is in do. All of the elementaryoperationsused to implement an
motion. The ambiguity is resolved only when the gate, or experiment are treatments that may or may not have a
plane, stops. significant effect on observed outcomes.
and vice versa. This does not mean that a sub- profitfor each price $0, $1, $2, ..., $10 charged
ject knows nothing about their matched coun- by the seller, and the buyer chooses to buy or
terpart.For example, it may appearevident that not buy. The profit of the seller is the price
he or she is another "like" person, such as an chosen; the profitof the buyer is ($10 - price).
undergraduate,or an industry executive with Each receives nothing if the buyer refuses to
whom one may feel more-or-less an in-group buy.
identity. Obviously, each person imports into
the experimenta host of different past experi- Contest/Exchange.-This treatmentcombines
ences and impressions that are likely to be as- Contestwith Exchange;i.e., the sellers andbuy-
sociated with the currentexperiment. ers in Exchange are selected by the contest
scoring procedure. In one version the total
F. The Contextof Decision: The Ultimatum amountis 10 one-dollarbills, and in the second
Game Example it is 10 ten-dollarbills.
Consider the ultimatum game, a two-stage, Whatever the context there is a game-
two-person game with the following abstract theoretic concept of equilibrium(subgame per-
form: for each pair the experimentermakes a fect) that yields the same predictionin all four
fixed sum of money, m, available (e.g., m will treatments (Reinhard Selten, 1975): Player 1
be 10 one-dollar bills, or 10 ten-dollar bills); offers the minimumunit of account, $1 ($10) if
Player 1 moves first offering 0 ' x < m units m = $10 ($100), and Player 2 accepts the offer.
of the money to Player 2, Player 1 retaining This follows from the assumption that each
m - x; Player 2 then respondsby either accept- player is self-interestedin the narrow sense of
ing the offer, in which case Player 1 is paid m always choosing the largest of two immediate
-x, and Player2 is paid x, or rejectingthe offer, payoffs for herself; that this condition is com-
in which case each player receives nothing. mon knowledge for the two players; and that
Below I report ultimatumresults from four Player 1 applies backwardinduction to the de-
different instructional/procedural treatments cision problem faced by Player 2, conditional
(contexts) that have the same underlying ab- on Player l's offer. Thus Player 1 reasons that
stractgame structure.In each case imagine that any positive payoff is betterthanzero for Player
you are Player 1. See Hoffman et al. (1994; 2 and therefore,Player 1 need only offer x = $1
hereafterHMSS) for instructionaldetails, and ($10).
for referencesto the literatureand the origins of One difficulty with this analysis is that, de-
the ultimatumgame. pending on context, the interactionmay be in-
terpretedas a social exchange between the two
Divide $10.-You and your counterparthave anonymously matched players who in day-to-
been provisionallyallocated $10, and randomly day experience read intentions into the actions
assigned to positions. Your task as Player 1 is to of others (S. Baron-Cohen,1995). Suppose the
divide the $10 by filling out a form that will situation is perceived as a social contract as
then go to your counterpartwho will accept or follows: if Player 2 has an entitlementto more
reject it. thanthe minimumunit of account,then an offer
of less than the perceived entitlement(say, only
Contest Entitlement.-The 12 people in the $1, or even $2-$3) may be rejected by some
room each answer the same 10 questions on a Players 2. Player 1, introspectivelyanticipating
generalknowledge quiz. Your score is the num- this possible mental state of Player 2, might
ber of questions answered correctly; ties are then offer $4 or $5 to insure acceptance of his
brokenin favor of the person who first finished offer. Alternatively, Player 1 might enjoy (get
the quiz. The scores are rankedfrom 1 (highest) utility from) giving money to his counterpart.
through 12 (lowest). Those ranked 1-6 are in- The point is simply that there are alternative
formed that they have earned the right to be models to that of subgame perfection that pre-
Player 1, the other six will be Players 2. dict choices in the ultimatumgame, and these
alternativesleave wide latitudefor the possibil-
Exchange.-Player 1 is a seller, and Player 2 ity of context affecting the behavior of both
is a buyer. A table lists the buyer and seller players. Abstract game theory can embrace
however, this miniscule decline in the mean for context across cultures.In each cultureone
offer causes the rejectionrate to go up from and ob-
needs to vary the instructions/procedures
12.5 percent to 21.7 percent. Three of four serve the sampling distribution of outcomes,
subject Players 1 offering $10 are rejected, then compare the sample distributions across
and one offer of $30 is rejected in the game cultures.
with $100 stakes. As has been shown in These instructionaltreatmenteffects call into
trust/punishmentgames, this behavior is as- question the extent to which one can define
sociated with a strong human propensity to what is meant by "unbiased"instructions. If
incur personal cost to punish those who are results are robust with respect to instructional
perceived as cheaters, even under strict changes, this can only be established empiri-
anonymity. cally. Without such studies no claims can be
3. We note that comparing the Divide $10/ made concerning the relative "neutrality"of
Random entitlementcondition with the Ex- instructions.The main lesson is that,because of
change entitlement, the offer percentage the nature of perception and memory, context
declines from 43.7 percent to 37.1 percent, should matter, and in the ultimatum game the
and comparingthe formerto the Earneden- variation of observed results with systematic
titlementthe decline is from 43.7 percent to instructionalchanges designed to alter context
36.2 percent, both reductionsbeing statisti- shows clearly that context can and does matter.
cally significant.Even more significantis the Experimenters,subject to the same perception/
reductionfrom 43.7 percent to 30.8 percent memory variations,are likely to disagree as to
when the Earnedand Exchange entitlements what is "neutral."
are combined. Moreover,in all four of these
comparisons the rejection rate is null or G. Dictator Games Withand WithoutGains
modest (0 to 12.5 percent). from Exchange
4. The small proportionof the offers that were
rejected, except when the stakes were $100 The ultimatumgame is converted into a dic-
in the Earned/Exchange context and the tatorgame by removing the right of the second
mean offers declined to a low of 27.8 per- mover to veto the offer of the first. Forsythe et
cent, indicatesthatPlayers 1 readtheircoun- al. (1994; hereafterFHSS) note that if the ob-
terpartswell, and as the context is altered, served tendency toward equal split of the prize
normally offer a sufficient amount to avoid is due primarilyto "fairness"-a social norm of
being rejected. The one exception shows just division-then it should be of little conse-
clearly that pushing the edge, even if it quence if this right is eliminated.But if it is the
seems justified by the higher stakes, may prospectof rejection-however irrational-that
invite an escalation of rejections. tempers the amount offered by Player 1, then
the outcomes should be materially affected by
These data indicate that context is important removing the right of rejection, which converts
in the ultimatumgame: the percentage offered the ultimatum game into what is called the
varies by over a third as we move from the dictator game. Thus a significant reduction in
highest (44 percent) to the lowest (28 percent) the mean percent offered in the dictator game
measuredeffect. Also see Hoffmanet al. (2000). would be consistentwith the second hypothesis,
Like variationis reportedin cross-culturalexper- while no significantreductionwould be consis-
iments: a comparisonof two hunter-gatherand tent with the first. Comparing the results in
five modem cultures reveals variation from a Table 3, column 1 with those for Divide $10,
high of 48 percent (Los Angeles subjects) to a Randomentitlementin Table 2, we see that the
low of 26 percent (Machiguengasubjects from mean dictator offer is only 23.3 percent com-
Peru) (Joseph Henrich, 2000). These compari- pared with the mean ultimatum offer of 43.7
sons used care in attemptingto control for in- percent.FHSS conclude that fairness alone can-
structionaldifferencesacrossdifferentlanguages, not accountfor behaviorin the ultimatumgame.
butthis is inherentlyproblematic,given the nature This is correct,but, equally of interest,why are
of perception,in thatone cannotbe sure thatthe dictators giving away nearly a quarterof their
instructions,translations,payoffs, or the proce- endowment? This research puzzle was picked
duresfor handlingthe subjects,controladequately up by HMSS who conjecturedthat such gener-
WITHANDWrrTHOUT
TABLE3-DICTATORGIVING: GAINSFROM ANDSOCIAL
EXCHANGE HISTORY
Double blindb
dictatorgiving under gains from exchangec
Standarddictator
Treatments game giving Baseline Social history
Single Double
Blind Blind 2 Sent Returned Sent Returned
Player role FHSSa HMSSb (Player 1) (Player 2) (Player 1) (Player 2)
Mean percent given 23.3 10.5 51.6 27.2d 53.6 35.5d
of total available
Mean percent given by 38.3 21.0 74.4 49.4 82.7 55.8
top 50 percent givers
a
Data from Forsythe et al. (1994), replicatedby Hoffman et al. (1996b).
b
Data from Hoffman et al. (1994), Double Blind 2.
c Data from
Berg et al. (1995). Their proceduresare different,but are nearest to those of Double Blind 2 in HMSS.
d
Since the sender amountis tripled,if the receivers returnan average of 33.3 percent,then the average amountreturned
will equal (pay back) the amount sent.
McCabeand Smith (2000) examinedthis expla- Note: For abbreviations,see Figure 1 notes.
nation using advanced graduatestudents from
the population,a sample of whom participated
in the $100 Exchange/Entitlementversion of saying that people have a taste for altruismor
the ultimatumgame reportedin Table 2, show- for "fair"outcomes (or a distaste for "unfair"
ing almost identical results for graduate and outcomes), where fairness is understoodas pay-
undergraduatestudents. They used the trust off equity, as in Franciosi et al. (1995). The
game shown in Figure 1. For comparison, hypothesis is that subjects seek to maximize
McCabe and Smith (2000) used the undergrad- these adjusted utilities. It is only the intrinsic
uate data shown in Figure 2. In both groups 50 propertiesof outcomes thatare assumedto drive
percentof Players 1 offer cooperation,while 75 behavior;what alternativesthe players faced at
percent of the undergraduateand 64 percent of a previous decision node are irrelevant. This
the graduate student Players 2 reciprocate. means that intentions, as reflected in the move
However naive undergraduatesare alleged to choices, are assumed to be superfluousin the
be, these tests suggest that graduate students interactions between the parties. The former
with trainingin economic theory are capable of approachidentifiesutility types. The latteriden-
showing very similarbehaviorin this extensive- tifies types who signal intentions, who are into
form game, and in the ultimatumgame reported reading move signals, and risk misidentifying
in Table 2. reciprocity versus defection types. The impor-
tant testable distinctions are that the former
Is It Utility for Other?-Bolton (1991), are immune to instructionalproceduresand to
Rabin (1993), Fehr and Schmidt (1999), and path dependency-the opportunity costs of
Bolton and Ockenfels (2000) have proposed forgone options; this constitutes the core of
useful preferencemodels of decision that aim at the researchprogramof my coauthorsand me.
accounting for behavior in a variety of experi- This programleads naturallyto an understand-
ments, most particularlyultimatumand dictator ing of its own limitations as well as that of
games. The idea behind these models is that we the utilitarianprogram,i.e., both types may be
can explain cooperationin bargaininggames by needed.
N F SPE N F
0.37 $20 27
27 ()
$20 X1
X1
1.0
0.63
C_
C 0.33
0.65 - - 27 $25
17 X2 $25
X2 $25
0.67
0.35
$15 18
6 $30
D
FIGURE TRUSTGAME
4A. VOLUNTARY FIGURE TRUSTGAME
4B. INVOLUNTARY
Note: For abbreviations,see Figure 1 notes. Note: For abbreviations,see Figure 1 notes.
Players' moves signal intentionsthat are not the neuroscientistswho study the deviant behavior
same when actually experienced in extensive of neurological patients with specific brain le-
form as when imagined in a mental experiment sions, such as front lobe (ventromedialprefron-
correspondingto the normal form. I would ar- tal) damage. Such patients have long been
gue that experience and its memory in life is an known to be challenged by tasks involving
extensive process that encodes context along planning and coordinationover time, although
with outcome. The brain is not naturally they score normally on batteriesof psycholog-
adaptedto solve all sequential move problems ical tests (Antonio R. Damasio, 1994). A land-
by reducing them to a single strategyvector as mark experimental study of such patients
in a highly structuredgame. Apparently, we (comparedwith controls)in individualdecision-
have a built-in tendency to wait, observe, then making under uncertainty is that of Antoine
decide-a process that conserves cognitive re- Bechara et al. (1997). Starting with (fictional)
sources by applyingthem only to contingencies endowments of $2,000, each subject on each
that are realized, and avoids the need for revi- trialdrawsa cardfrom one of four freely chosen
sion, given the inevitable surprises in the less decks (A, B, C, D). In decks (A, B) each card
structuredgames of life.70 Constructivistmod- has a payoff value of $50, whereas in decks (C,
eling glosses over distinctions of which we are D), each is worth $100. Also, the $100 decks
unawarethat govern the ecology of choice. Ex- contain occasional large negative payoff cards,
perimental designs conditioned only by con- while the $50 decks have much smallernegative
structivistthinking,ill-prepareus to collect the payoff cards. All this must be learned from
data that can inform needed revisions in our single card draws in a sequence of trials, with a
thinking.It is both cost effective and faithful to running tally of cumulative payoff value. A
game-theoretic assumptions in experiments to subject performs much better by learning to
collect move data from each subject under all avoid the $100 decks in favor of the $50 decks.
contingencies, but it distorts interpretabilityif By period 60, normal control subjects have
game forms are not equivalent. The assump- learnedto drawonly from the $50 (C, D) decks,
tions of game theory, such as those leading to while the brain-damagedsubjects continue to
the logical equivalence of the two game forms, sample disadvantageously the $100 (A, B)
should not be imposed on experimentaldesigns, decks. Furthermore,the control subjects shift to
thereby constrainingour understandingof be- the (C, D) decks before they are able to articu-
havior beyond those assumptions. late why, in periodic questioning. Also, they
pre-registeremotional reactions to the (A, B)
K. Neuroeconomics decks as measuredby real-time skin conductiv-
ity test (SCT) readings. But the brain-damaged
Neuroeconomics is concerned with studying patients tend to verbally rationalize continued
the connections between how the mind/brain sampling of the (A, B) decks, and some types
works-the internal order of the mind-and (with damaged amygdala) register no SCT re-
behavior in (1) individual decision making, (2) sponse. Results consistent with those of Be-
social exchange, and (3) institutions such as charaet al. (1997) have been reportedby Vinod
markets. The working hypothesis is that the Goel et al. (1997) who study patient perfor-
brainhas evolved different,but interdependent, mance in a complex financial planning task.
adaptive mechanisms for each of these tasks Over 50 years ago experiments with animal
involving experience, memory, and perception. behaviordemonstratedthatmotivationwas based
The tools include brain-imagingtechnology and on relative or forgone reward-opportunity
the existence of patients with localized brain cost-and not on an absolute scale of values
lesions associated with specific loss of certain generatedby the brain. Thus David J. Zeaman
mental functions. (1949) reportedexperimentsin which rats were
Decision making has drawn the attention of trainedto runfor a large reward-motivatedgoal.
When shifted to a small reward, the rats re-
sponded by running more slowly than they
would to the small rewardonly. A controlgroup
Any such naturalprocess must be deliberately over-
70
come, constructively, in situations where nature serves us began with a small rewardand shifted to a large
poorly. one, and these rats immediately ran faster than
if the large rewardalone was applied. Monkeys examination, perhaps by imaging split-brain
similarlyrespondto comparisonsof differential subjectsin the same task. Also see Kip Smith et
rewards.It is now establishedthat orbitofrontal al. (2002).
cortex (just above the eyes) neuron activity in The effect of paying subjects is informedby
monkeys enable them to discriminatebetween Thut et al. (1997) who comparebrainactivation
rewardsthat are directly related to the animals' undermonetaryrewardswith the feedbackof an
relative, as distinct from absolute, preference "OK" reinforcementin a dichotomous choice
among rewardssuch as cereal, apple, and raisins task. The monetary rewards yielded a signifi-
(in order of increasing preference in monkeys) cantly higher activationof the orbitofrontalcor-
(Leon Tremblayand Schultz, 1999). tex and other related brain areas (see also
Thus suppose A is preferredto B is preferred Schultz, 2000, 2002).
to C based on choice response. Then neuronal The neural correlatesof individual decision-
activity is greater for A than for B when the making were extended, by McCabe et al.
subject is viewing A and B, and similarly for B (2001), to an fMRI study of behavior in two-
and C when comparingB and C. But the activ- person strategic interactionsin extensive-form
ity associated with B is much greater when trust games like those in Figures 1 to 4. The
comparedwith C than when it is comparedwith prior hypothesis, derived from reciprocity the-
A. This is contraryto what one would expect to ory, the theory-of-mind literature, and sup-
observe if A, B, and C are coded on a fixed ported by imaging results from individual
properties scale rather than a relative scale studies of cued thought processes (P. C.
(Tremblayand Schultz, 1999, p. 706). Fletcher et al., 1995), was that cooperators
These studies have a parallel significance for would show greateractivationin the prefrontal
humans.Prospecttheory proposes that the eval- cortex (specifically BA-8) and supportingcir-
uationof a gamble dependsnot on the total asset cuitry, than noncooperators.The control, for
position but focuses myopically on the oppor- comparison with the mental processes used
tunitycost, gain or loss, relativeto one's current when a subject is playing a human, is for the
asset position. There is also asymmetry-the subject to play a computer knowing the pro-
effect of a loss looms largerthan the effect of a grammed response probabilities and therefore
gain of the same magnitude (Kahneman and having no need to interpretmoves as intentions.
Tversky, 1979). BarbaraA. Mellers et al. (1999) The predicted activations were significantly
have shown that the emotional response to the greater,relativeto controls,for cooperatorsthan
outcome of a gamble depends on the perceived noncooperators,and are consistent with the rec-
value and likelihood of the outcome and on the iprocity interpretationof behavior discussed
forgone outcome. It feels better (less bad) to above.
receive $0 from a gamble when you forgo +$10
than when you forgo + $90. (They use the term IV. Conclusions
"counterfactual"ratherthan "opportunitycost"
to refer to the alternativethat might have pre- Cartesianconstructivismapplies reason to the
vailed.) Thus, our ability to form opportunity design of rulesfor individualaction,to the design
cost comparisonsreceive importantneurophys- of institutions that yield socially optimal out-
iological supportfrom our emotional circuitry. comes, and constitutes the standardsocioeco-
Breiteret al. (2001) use these same principlesin nomic science model. But most of our operating
the design of a functional magnetic imaging knowledge, and ability to decide and performis
(fMRI) study of human hemodynamic re- nondeliberative.Our brainsconserve attentional,
sponses to both the expectation and experience conceptual,and symbolic thoughtresourcesbe-
of monetarygains and losses underuncertainty. cause they are scarce, and proceeds to delegate
They observed significant activation responses most decision-making to autonomic processes
in the amygdala and orbital gyms, with both (including the emotions) that do not require
activations increasing with the expected value conscious attention. Emergent arrangements,
of the gamble. There was also some evidence even if initiallyconstructivist,must have survival
that the right hemisphere is predominantlyac- properties that incorporate opportunity costs
tive for gains, and the left for losses-a partic- and environmentalchallenges invisible to con-
ularly interesting possibility inviting deeper structivistmodeling.This leads to an alternative,
ecological concept of rationality:an emergent not given, or available, to any one mind.
order based on trial-and-errorculturaland bio- Thus, for many, the argumentsof the Scot-
logical evolutionaryprocesses. It yields home- tish philosophersand of Hayek are obscure
and socially grown rules of action, traditions and mystical. But we can design experi-
and moral principles that underlie property ments in which the informationis not given
rights in impersonalexchange, and social cohe- to any participant, then compare market
sion in personal exchange. To study ecological outcomes with efficient competitive out-
rationalitywe use rational reconstruction-for comes and gauge a market institution's
example, reciprocityor other-regardingprefer- performance.
ences-to examine individual behavior, emer- 4. The resulting order is invisible to the par-
gent order in human culture and institutions, ticipants,unlike the visible gains they reap.
and their persistence, diversity, and develop- Agents discover what they need to know to
ment over time. Experimentsenable us to test achieve outcomes optimal against the con-
propositions derived from these rational straininglimits imposed by others.
reconstructions. 5. Rules emerge as a spontaneousorder-they
The study of both kinds of rationality has are found-not deliberately designed by
been prominent in the work of experimental one calculating mind. Initially constructiv-
economists. This is made plain in the many ist institutionsundergoevolutionarychange
direct tests of the observable implications of adapting beyond the circumstances that
propositionsderived from economic and game gave them birth.What emerges is a form of
theory. It is also evident in the great variety of "social mind" that solves complex organi-
experiments that have reached far beyond the zation problems without conscious cogni-
theory to ask why the tests have succeeded, tion. This "social mind" is born of the
failed, or performedbetter (under weaker con- interaction among all individuals through
ditions) than was expected. What have we the rules of institutions that have to date
learned, not as final truth, but as compelling survived culturalselection processes.
workinghypothesesfor continuingexamination? 6. This process accommodates trade-offs be-
tween the cost of transacting,attending,and
1. Marketsconstitute an engine of productiv- monitoring, and the efficiency of the allo-
ity by supporting resource specialization cations so that the institution itself gener-
throughtradeand creatinga diverse wealth ates an order of economy that fits the
of goods and services. problem it evolved to solve. Hence, the
2. Marketsare rule-governedinstitutionspro- hundredsof variationson the fine structure
viding algorithmsthat select, process, and of institutions,each designed without a de-
order the exploratory messages of agents signer to accommodate disparate condi-
who are betterinformedas to theirpersonal tions, but all of them subservient to the
circumstancesthan that of others. As pre- reality of dispersed agent information.
cautionaryprobes by agents yield to con- 7. We understandlittle about how rule sys-
tracts, each becomes more certain of what tems for social interaction and markets
must be given in order to receive. Out of emerge, but it is possible in the laboratory
this interactionbetween minds throughthe to do variations on the rules, and thus to
intermediary of rules the process aggre- study that which is not.
gates the dispersed asymmetric informa- 8. Marketsrequireenforcement-voluntary or
tion, converging more-or-less rapidly to involuntary-of the rules of exchange.
competitive equilibria if they exist. Each These are: the right possession, it's trans-
experimentalmarketcarriesits own unique ference by consent, and the performanceof
mark with a different dynamic path. promises (Hume). Voluntary enforcement
3. All this informationis capturedin the static occurs when people in the market reward
or time-variablesupply and demand envi- good services with gratuities or "tips," an
ronment and must be aggregated to yield example, perhaps, of an emergent cultural
efficient clearingprices. We can never fully norm in which people recognize that tips
understandhow this process works in the are partof an informalexchange. If self- or
world because the requiredinformationis community-enforcementconditions are not
present, the result is unintended conse- Anderson,Terry L. and Hill, Peter J. "The Evo-
quences for the bad, as markets are com- lution of Property Rights: A Study of the
promised or may fail. The game of "trade" American West." Journal of Law and Eco-
must not yield to the game of "steal." nomics, April 1975, 18(1), pp. 163-79.
9. Reciprocity, trust, and trustworthinessare Arrow,KennethJ. "Rationalityof Self and Oth-
importantin personal exchange where for- ers in an Economic System," in Robin M.
mal markets are not worth their cost, yet Hogarthand Melvin WarrenReder, eds., Ra-
there are gains from exchange to be cap- tional choice: The contrast between econom-
tured. They are also importantin contract- ics and psychology. Chicago: University of
ing as not every margin of gain at the Chicago Press, 1987, pp. 201-16.
expense of other can be anticipated and Backerman, Steven R.; Denton, Michael; Ras-
formalized in written contracts. senti, Stephen and Smith, Vernon L. "Market
10. People are not requiredto be selfish; rather Power in a DeregulatedElectrical Industry."
the point of the Scottish philosophers was Journal of Decision Support Systems, 2001,
that people did not have to be good to (30), pp. 357-81.
produce good. Markets economize on in- Banks, Jeffrey; Olson, Mark; Rassenti, Stephen
formation,understanding,rationality,num- and Smith, Vernon L. "Theory, Experiment
bers of agents, and virtue. and the Federal Communications Commis-
11. Markets in no way need destroy the foun- sion Spectrum Auctions." Journal of Eco-
dation upon which they probably nomic Behavior and Organization, 2003
emerged-social exchange between family, (forthcoming).
friends, and associates. This is supportedin Baron-Cohen,S. Mindblindness:An essay on
the studies reported by Henrich (2000). autism and theory of mind. Cambridge,MA:
Thus, individualscan be habitualsocial ex- MIT Press, 1995.
changers and vigorous tradersas well, but Bechara, Antoine; Damasio, Hanna; Tranel,
as in Hayek's "two worlds" text, the eco- Daniel and Damasio, Antonio R. "Deciding
logically rational coexistence of personal AdvantageouslyBefore Knowing the Advan-
and impersonal exchange is not a self- tageous Strategy." Science, February 28,
aware Cartesian construct. Consequently, 1997, 275(5304), pp. 1293-95.
there is the ever-present danger that the Berg, Joyce; Dickhaut, John and McCabe, Kevin.
rules of "personal exchange" will be ap- "Risk Preference Instability Across Institu-
plied inappropriatelyto govern or modify tions: A Dilemma." Working paper, Carlson
the extendedorderof markets.Equallydan- School of Management,University of Min-
gerous, the rules of impersonalmarketex- nesota, 1994.
change may be applied inappropriatelyto . "Trust,Reciprocity, and Social Histo-
our cohesive social networks. ry." Games and Economic Behavior, July
12. New brain-imagingtechnologies have mo- 1995, 10(1), pp. 122-42.
tivated neuroeconomicstudies of the inter- Binswanger,Hans P. "Attitudes Toward Risk:
nal orderof the mind and its links with the ExperimentalMeasurementin Rural India."
spectrum of human decision from choice American Journal of Agricultural Econom-
among fixed gambles to choice mediatedby ics, August 1980, 62(3), pp. 395-407.
marketand other institutionalrules. We are Bohnet, Iris; Frey, Bruno S. and Huck, Steffen.
only at the beginning of this enterprise,but "More Order with Less Law: On Contract
its promise suggests a fundamentalchange Enforcement,Trust, and Crowding."Ameri-
in how we think, observe, and model deci- can Political Science Review, March 2001,
sion in all its contexts. 95(1), pp. 131-44.
Bolton,Gary E. "A ComparativeModel of Bar-
REFERENCES gaining: Theory and Evidence." American
Economic Review, December 1991, 81(5),
Acheson,James. "The Lobster Fiefs: Economic pp. 1096-136.
and Ecological Effects of Territorialityin the Bolton, Gary E; Katok, Elena and Zwick, Rami.
Maine Lobster Industry."Human Ecology, "Dictator Game Giving: Rules of Fairness
September 1975, 2(3), pp. 183-207. versusActs of Kindness."InternationalJournal
of Game Theory,July 1998, 27(2), pp. 269- Chu, Yun Peng and Chu, Ruey Ling. "The Sub-
99. sistance of Preference Reversals in Simpli-
Bolton, Gary E and Ockenfels,Axel. "ERC: A fied and MarketlikeExperimentalSettings:A
Theory of Equity, Reciprocity,and Competi- Note." AmericanEconomic Review, Septem-
tion." American Economic Review, March ber 1990, 80(4), pp. 902-11.
2000, 90(1), pp. 166-93. Coase,RonaldH. "TheProblemof Social Cost."
Breiter, Hans C.; Aharon, Itzhak; Kahneman, Journal of Law and Economics, October
Daniel; Dale, Anders and Shizgal, Peter. 1960, 3(1), pp. 1-44.
"FunctionalImaging of Neural Responses to . "TheLighthousein Economics."Jour-
Expectancy and Experience of Monetary nal of Law and Economics, October 1974,
Gains and Losses." Neuron, May 2001, 17(2), pp. 357-76.
30(2), pp. 619-39. Coricelli, Giorgio; McCabe, Kevin and Smith,
Bronfman, Corrine; McCabe, Kevin; Porter, Vernon L. "Theory-of-MindMechanism in
David;Rassenti,Stephenand Smith,VernonL. PersonalExchange,"in G. Hatano,N. Okada,
"An ExperimentalExaminationof the Wal- and H. Tanabe,eds., Affectiveminds.Amster-
rasian Tatonnement Mechanism." RAND dam: Elsevier, 2000.
Journal of Economics, Winter 1996, 27(4), Cosmides, Leda and Tooby, John. "Cognitive
pp. 681-99. Adaptationsfor Social Exchange,"in Jerome
Burnham, Terence C. "Engineering Altruism: H. Barkow,Leda Cosmides, and JohnTooby,
An ExperimentalInvestigationof Anonymity eds., The adapted mind. New York: Oxford
and Gift Giving." Unpublished manuscript, University Press, 1992, pp. 163-228.
Kennedy School of Government, Harvard Cox,JamesC. and Grether,DavidM. "ThePref-
University, 1998. erence Reversal Phenomenon: Response
Burnham,Terence; McCabe,Kevin and Smith, Mode, Markets and Incentives." Economic
Vernon L. "Friend-or-Foe Intentionality Theory,April 1996, 7(3), pp. 381-405.
Priming in an Extensive Form Trust Game." Damasio,AntonioR. Descartes' error: Emotion,
Journal of Economic Behavior and Organi- reason, and the human brain. New York:
zation, September2000, 43(1), pp. 57-73. G. P. Putnam, 1994.
Camera,Gabriele;Noussair,Charlesand Tucker, Davis, DouglasD. and Holt, CharlesA. Experi-
Steven. "Rate-of-ReturnDominance and Ef- mental economics. Princeton, NJ: Princeton
ficiency in an ExperimentalEconomy."Eco- University Press, 1993.
nomic Theory,2003 (forthcoming). Deck, Cary; McCabe,Kevin and Porter, David.
Camerer,ColinF. Behavioral game theory: Ex- "Why Stable Flat Money Hyperinflates:Re-
perimentson strategic interaction.Princeton, sults from an ExperimentalEconomy." Un-
NJ: PrincetonUniversity Press, 2002. published manuscript, Department of
Camerer,Colinand Ho, Teck-Hua."Experience- Economics, University of Arkansas,2001.
Weighted Attraction Learning in Normal Deck, Cary and Wilson, Bart. "TrackingCus-
Form Games." Econometrica, July 1999, tomerSearchto Price Discriminate."Working
67(4), pp. 827-74. paper, InterdisciplinaryCenterfor Economic
Camerer,Colin;Ho, Teck-Huaand Chong,Juin- Science, George Mason University, 2002.
Kuan. "BehavioralGame Theory: Thinking, Denton, Michael J.; Rassenti, Stephen J. and
Learning,and Teaching."Nobel Symposium Smith, Vernon L. "Spot Market Mechanism
on Behavioral and ExperimentalEconomics, Design and Competitivity Issues in Electric
Stockholm, December 2001. Power." Journal of Economic Behavior and
Camerer,Colinand Weigelt,Keith."Experimen- Organization,April 2001, 44(4), pp. 435-53.
tal Tests of a SequentialEquilibriumModel." de Waal,FransB. M. "Food Sharingand Recip-
Econometrica, January 1988, 56(1), pp. rocal Obligations Among Chimpanzees."
1-36. Journal of Human Evolution, May 1989,
Cason,Timothyand Friedman,Daniel."An Em- 18(5), pp. 433-59.
piricalAnalysisof PriceFormationin Double- . "Food Transfers Through Mesh in
Auction Markets," in D. Friedman and J. Brown Capuchins."Journal of Comparative
Rust, eds., The double auction. Redwood Psychology, December 1997, 111(4), pp.
City, CA: Addison-Wesley,1993, pp. 253-83. 370-78.
Ostrom, Elinor. Governing the commons: The InternationalJournal of Game Theory, 1997,
evolution of institutionsfor collective action. 26(1), pp. 113-36.
Cambridge: Cambridge University Press, Rassenti,Stephen;Smith, VernonL. and Bulfin,
1990. R. "A CombinatorialAuction Mechanismfor
Pillutla, Madan; Malhotra, Deepak and Mur- Airport Time Slot Allocation." Bell Journal
nighan, J. Keith. "Attributionsof Trust and of Economics, Autumn 1982, 13(2), pp. 402-
the Calculus of Reciprocity."Journal of Ex- 17.
perimental Social Psychology, 2003 (forth- Rassenti,Stephen;Smith,VernonL. and Wilson,
coming). Bart J. "Using Experiments to Inform the
Pinker, Steven. The language instinct. New Privatization/Deregulation Movement in
York: W. Morrow and Co., 1994. Electricity." Cato Journal, Winter 2002,
. The blank slate: The modem denial of 21(3), pp. 515-44.
humannature. New York: Viking, 2002. . "ControllingMarketPower and Price
Plott, CharlesR. "Researchon Pricing in a Gas Spikes in ElectricityNetworks:Demand-side
Transportation Network." Office of Eco- Bidding."Proceedings of the National Acad-
nomic Policy Technical Report No. 88-2. emy of Sciences, March 2003, 100(5), pp.
Federal Energy Regulatory Commission, 2998-3003.
Washington,DC, July 1988. Rosenthal,Robert W. "Games of Perfect Infor-
. "Equilibrium,Equilibration,Information mation, Predatory Pricing and the Chain-
and MultipleMarkets:FromBasic Science to StoreParadox."Journalof Economic Theory,
InstitutionalDesign." Nobel Symposium,Be- August 1981, 25(1), pp. 92-100.
havioraland ExperimentalEconomics, Grand Satterthwaite,Mark."Strategy-ProofAllocation
Hotel Saltsjobaden,December 4-6, 2001. Mechanisms,"in J. Eatwell, M. Milgate, and
Plott,CharlesR. and Sunder,Shyam."Efficiency P. Newman, eds., The new Palgrave, volume
of Experimental Security Markets with In- 4. London: Macmillan, 1987, pp. 518-20.
siderInformation:An Applicationof Rational- Schotter, Andrew; Weigelt, Keith and Wilson,
Expectations Models." Journal of Political Charles."A LaboratoryInvestigationof Mul-
Economy, August 1982, 90(4), pp. 663-98. tiperson Rationality and Presentation Ef-
. "RationalExpectationsand the Aggre- fects." Games and Economic Behavior, May
gation of Diverse Informationin Laboratory 1994, 6(3), pp. 445-68.
Securities Markets."Econometrica,Septem- Schultz,Wolfram."MultipleReward Signals in
ber 1988, 56(5), pp. 1085-118. the Brain." Nature Reviews: Neuroscience,
Plott, CharlesR. and Uhl, JonathanT. "Compet- December 2000, 1(3), pp. 199-207.
itive Equilibriumwith Middlemen: An Em- . "Getting Formal with Dopamine and
pirical Study." Southern Economic Journal, Reward."Neuron, October 2002, 36(2), pp.
April 1981, 47(4), pp. 1063-71. 241-63.
Plott, CharlesR. and Wilde,Louis."Professional Schwartz,T. and Ang,J. S. "SpeculativeBubbles
Diagnosis vs. Self-Diagnosis:An Experimen- in the Asset Market: An Experimental
tal Examinationof Some Special Featuresof Study."Paper presentedat the American Fi-
Markets with Uncertainty,"in V. L. Smith, nance Association meeting, Atlanta, Decem-
ed., Research in experimental economics, ber, 1989.
volume 2. Greenwich, CT: JAI Press, 1982, Segal,NancyL. Entwinedlives: Twinsand what
pp. 63-112. they tell us about human behavior. New
Porter,Davidand Smith,VernonL. "StockMar- York: Plume, 1999.
ket Bubbles in the Laboratory." Applied Selten, Reinhard. "Reexaminationof the Per-
Mathematical Finance, December 1994, 1, fectness Concept for Equilibrium Points in
pp. 111-27. Extensive Games." InternationalJournal of
Rabin, Matthew. "IncorporatingFairness into Game Theory, 1975, 4, pp. 25-55.
Game Theory and Economics." American Siegel, Sidney. "TheoreticalModels of Choice
Economic Review, December 1993, 83(5), and Strategy Behavior." Psychometrika,
pp. 1281-302. 1959, 24, pp. 303-16.
Rapoport,Amnon. "Orderof Play in Strategi- Siegel, Sidney and Fouraker,LawrenceE. Bar-
cally EquivalentGames in Extensive Form." gaining and group decision making: Experi-